Hugging Face's logo

UncleStrawberryBear
/
FlyControl-models

Reinforcement Learning
English
neuroscience
drosophila
simulation
Model card Files
xet
 Community
UncleStrawberryBear commited on
Commit
d1c6c1b
·
verified ·
1 Parent(s): 42ec738

Add Figure 5 Ant flat-ground and maze experiments (part 3)

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. configs/figure5/flat/mlp/seed_9/final.yaml +62 -0
  2. configs/figure5/large/adult/seed_0/final.yaml +176 -0
  3. configs/figure5/large/adult/seed_1/final.yaml +176 -0
  4. configs/figure5/large/adult/seed_10/final.yaml +176 -0
  5. configs/figure5/large/adult/seed_11/final.yaml +176 -0
  6. configs/figure5/large/adult/seed_12/final.yaml +176 -0
  7. configs/figure5/large/adult/seed_13/final.yaml +176 -0
  8. configs/figure5/large/adult/seed_14/final.yaml +176 -0
  9. configs/figure5/large/adult/seed_15/final.yaml +176 -0
  10. configs/figure5/large/adult/seed_16/final.yaml +176 -0
  11. configs/figure5/large/adult/seed_17/final.yaml +176 -0
  12. configs/figure5/large/adult/seed_18/final.yaml +176 -0
  13. configs/figure5/large/adult/seed_19/final.yaml +176 -0
  14. configs/figure5/large/adult/seed_2/final.yaml +176 -0
  15. configs/figure5/large/adult/seed_3/final.yaml +176 -0
  16. configs/figure5/large/adult/seed_4/final.yaml +176 -0
  17. configs/figure5/large/adult/seed_5/final.yaml +176 -0
  18. configs/figure5/large/adult/seed_6/final.yaml +176 -0
  19. configs/figure5/large/adult/seed_7/final.yaml +176 -0
  20. configs/figure5/large/adult/seed_8/final.yaml +176 -0
  21. configs/figure5/large/adult/seed_9/final.yaml +176 -0
  22. configs/figure5/large/gru/seed_0/final.yaml +141 -0
  23. configs/figure5/large/gru/seed_1/final.yaml +141 -0
  24. configs/figure5/large/gru/seed_10/final.yaml +141 -0
  25. configs/figure5/large/gru/seed_11/final.yaml +141 -0
  26. configs/figure5/large/gru/seed_12/final.yaml +141 -0
  27. configs/figure5/large/gru/seed_13/final.yaml +141 -0
  28. configs/figure5/large/gru/seed_14/final.yaml +141 -0
  29. configs/figure5/large/gru/seed_15/final.yaml +141 -0
  30. configs/figure5/large/gru/seed_16/final.yaml +141 -0
  31. configs/figure5/large/gru/seed_17/final.yaml +141 -0
  32. configs/figure5/large/gru/seed_18/final.yaml +141 -0
  33. configs/figure5/large/gru/seed_19/final.yaml +141 -0
  34. configs/figure5/large/gru/seed_2/final.yaml +141 -0
  35. configs/figure5/large/gru/seed_3/final.yaml +141 -0
  36. configs/figure5/large/gru/seed_4/final.yaml +141 -0
  37. configs/figure5/large/gru/seed_5/final.yaml +141 -0
  38. configs/figure5/large/gru/seed_6/final.yaml +141 -0
  39. configs/figure5/large/gru/seed_7/final.yaml +141 -0
  40. configs/figure5/large/gru/seed_8/final.yaml +141 -0
  41. configs/figure5/large/gru/seed_9/final.yaml +141 -0
  42. configs/figure5/large/lstm/seed_0/final.yaml +141 -0
  43. configs/figure5/large/lstm/seed_1/final.yaml +141 -0
  44. configs/figure5/large/lstm/seed_10/final.yaml +141 -0
  45. configs/figure5/large/lstm/seed_11/final.yaml +141 -0
  46. configs/figure5/large/lstm/seed_12/final.yaml +141 -0
  47. configs/figure5/large/lstm/seed_13/final.yaml +141 -0
  48. configs/figure5/large/lstm/seed_14/final.yaml +141 -0
  49. configs/figure5/large/lstm/seed_15/final.yaml +141 -0
  50. configs/figure5/large/lstm/seed_16/final.yaml +141 -0
configs/figure5/flat/mlp/seed_9/final.yaml ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ policy:
2
+ name: mlp
3
+ actor_kwargs:
4
+ hidden:
5
+ - 1378
6
+ - 1378
7
+ activation: relu
8
+ value_hidden:
9
+ - 256
10
+ - 256
11
+ value_activation: relu
12
+ log_std_init: -1.0
13
+ env:
14
+ env_id: Ant-v5
15
+ friction_values: &id001
16
+ - 1.0
17
+ steps_per_context: 1000000000
18
+ cycle: false
19
+ append_zero_goal_features: true
20
+ stabilization: &id002
21
+ enabled: true
22
+ forward_cap_mps: 2.2
23
+ inversion_threshold_degrees: 90.0
24
+ inversion_persistence_steps: 10
25
+ action_scale: 0.9
26
+ resolved_env:
27
+ env_id: Ant-v5
28
+ friction_values: *id001
29
+ steps_per_context: 1000000000
30
+ cycle: false
31
+ append_zero_goal_features: true
32
+ stabilization: *id002
33
+ schedule_seed: 9
34
+ ppo:
35
+ total_timesteps: 200000
36
+ n_envs: 4
37
+ n_steps: 512
38
+ seq_len: 64
39
+ n_epochs: 10
40
+ n_minibatches: 4
41
+ gamma: 0.99
42
+ lam: 0.95
43
+ clip_range: 0.2
44
+ vf_clip_range: 0.2
45
+ lr: 0.0001
46
+ vf_coef: 0.5
47
+ ent_coef: 0.0
48
+ max_grad_norm: 0.5
49
+ normalize_reward: true
50
+ normalize_observation: true
51
+ observation_clip: 10.0
52
+ rollout_heartbeat: 128
53
+ best_checkpoint_mode: stability_velocity
54
+ best_checkpoint_window: 100
55
+ best_checkpoint_min_episodes: 20
56
+ best_checkpoint_min_forward_velocity: 1.5
57
+ best_checkpoint_max_high_z_termination_rate: 0.15
58
+ best_checkpoint_max_inverted_completion_rate: 0.05
59
+ best_checkpoint_max_functional_failure_rate: 0.2
60
+ evaluation:
61
+ run_robustness: true
62
+ seed: 9
configs/figure5/large/adult/seed_0/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 0
172
+ seed_roles:
173
+ optimization: 0
174
+ sampling: 0
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_1/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 1
172
+ seed_roles:
173
+ optimization: 1
174
+ sampling: 1
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_10/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 10
172
+ seed_roles:
173
+ optimization: 10
174
+ sampling: 10
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_11/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 11
172
+ seed_roles:
173
+ optimization: 11
174
+ sampling: 11
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_12/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 12
172
+ seed_roles:
173
+ optimization: 12
174
+ sampling: 12
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_13/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 13
172
+ seed_roles:
173
+ optimization: 13
174
+ sampling: 13
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_14/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 14
172
+ seed_roles:
173
+ optimization: 14
174
+ sampling: 14
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_15/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 15
172
+ seed_roles:
173
+ optimization: 15
174
+ sampling: 15
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_16/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 16
172
+ seed_roles:
173
+ optimization: 16
174
+ sampling: 16
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_17/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 17
172
+ seed_roles:
173
+ optimization: 17
174
+ sampling: 17
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_18/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 18
172
+ seed_roles:
173
+ optimization: 18
174
+ sampling: 18
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_19/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 19
172
+ seed_roles:
173
+ optimization: 19
174
+ sampling: 19
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_2/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 2
172
+ seed_roles:
173
+ optimization: 2
174
+ sampling: 2
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_3/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 3
172
+ seed_roles:
173
+ optimization: 3
174
+ sampling: 3
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_4/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 4
172
+ seed_roles:
173
+ optimization: 4
174
+ sampling: 4
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_5/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 5
172
+ seed_roles:
173
+ optimization: 5
174
+ sampling: 5
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_6/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 6
172
+ seed_roles:
173
+ optimization: 6
174
+ sampling: 6
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_7/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 7
172
+ seed_roles:
173
+ optimization: 7
174
+ sampling: 7
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_8/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 8
172
+ seed_roles:
173
+ optimization: 8
174
+ sampling: 8
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/adult/seed_9/final.yaml ADDED
@@ -0,0 +1,176 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: connectome_rate
5
+ actor_kwargs:
6
+ dt: 0.01
7
+ tau_init: 0.02
8
+ use_lora: false
9
+ input_target: sensory
10
+ goal_adapter_hidden:
11
+ - 256
12
+ - 128
13
+ goal_adapter_features: 109
14
+ goal_router_hidden:
15
+ - 256
16
+ - 256
17
+ goal_router_codebook:
18
+ - - -18.0
19
+ - 12.0
20
+ - - -14.0
21
+ - 12.0
22
+ - - -10.0
23
+ - 12.0
24
+ - - -6.0
25
+ - 12.0
26
+ - - 2.0
27
+ - 12.0
28
+ - - 6.0
29
+ - 12.0
30
+ - - 10.0
31
+ - 12.0
32
+ - - 14.0
33
+ - 12.0
34
+ - - 18.0
35
+ - 12.0
36
+ - - -18.0
37
+ - 8.0
38
+ - - -6.0
39
+ - 8.0
40
+ - - 2.0
41
+ - 8.0
42
+ - - 10.0
43
+ - 8.0
44
+ - - 18.0
45
+ - 8.0
46
+ - - -18.0
47
+ - 4.0
48
+ - - -14.0
49
+ - 4.0
50
+ - - -10.0
51
+ - 4.0
52
+ - - -6.0
53
+ - 4.0
54
+ - - -2.0
55
+ - 4.0
56
+ - - 2.0
57
+ - 4.0
58
+ - - 10.0
59
+ - 4.0
60
+ - - 14.0
61
+ - 4.0
62
+ - - 18.0
63
+ - 4.0
64
+ - - -18.0
65
+ - 0.0
66
+ - - 2.0
67
+ - 0.0
68
+ - - 18.0
69
+ - 0.0
70
+ - - -18.0
71
+ - -4.0
72
+ - - -14.0
73
+ - -4.0
74
+ - - -6.0
75
+ - -4.0
76
+ - - 2.0
77
+ - -4.0
78
+ - - 6.0
79
+ - -4.0
80
+ - - 10.0
81
+ - -4.0
82
+ - - 14.0
83
+ - -4.0
84
+ - - 18.0
85
+ - -4.0
86
+ - - -14.0
87
+ - -8.0
88
+ - - -6.0
89
+ - -8.0
90
+ - - 2.0
91
+ - -8.0
92
+ - - 10.0
93
+ - -8.0
94
+ - - -18.0
95
+ - -12.0
96
+ - - -14.0
97
+ - -12.0
98
+ - - -6.0
99
+ - -12.0
100
+ - - -2.0
101
+ - -12.0
102
+ - - 2.0
103
+ - -12.0
104
+ - - 10.0
105
+ - -12.0
106
+ - - 14.0
107
+ - -12.0
108
+ - - 18.0
109
+ - -12.0
110
+ goal_router_codebook_latch_radius: 0.45
111
+ value_hidden:
112
+ - 256
113
+ - 256
114
+ value_activation: relu
115
+ connectome:
116
+ species: adult
117
+ kind: original
118
+ adult_dir: larva/data/adult
119
+ chunksize: 250000
120
+ min_syn_count: 1
121
+ normalization: row_l1
122
+ global_gain: 1.0
123
+ sensory_selector:
124
+ source: classification
125
+ field: flow
126
+ include:
127
+ - afferent
128
+ output_selector:
129
+ source: classification
130
+ field: super_class
131
+ include:
132
+ - motor
133
+ polarity:
134
+ unknown_action: zero
135
+ nt_signs:
136
+ ACH: 1.0
137
+ GLUT: 1.0
138
+ GABA: -1.0
139
+ DA: 1.0
140
+ SER: 1.0
141
+ OCT: 1.0
142
+ env:
143
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
144
+ env_id: AntMaze_Large_Diverse_GR-v5
145
+ goal_feature_mode: world
146
+ continuing_task: false
147
+ resolved_env:
148
+ env_id: AntMaze_Large_Diverse_GR-v5
149
+ goal_feature_mode: world
150
+ continuing_task: false
151
+ imitation:
152
+ demonstration_steps: 60000
153
+ dagger_steps: 30000
154
+ dagger_iterations: 0
155
+ updates_per_stage: 5000
156
+ batch_size: 64
157
+ burn_in: 8
158
+ learning_rate: 0.0001
159
+ gradient_clip: 1.0
160
+ actor_train_mode: goal_adapter_only
161
+ goal_input_features: 4
162
+ loss_space: pre_tanh
163
+ sampling_mode: episode_uniform
164
+ checkpoint_interval_updates: 500
165
+ navigation:
166
+ waypoint_planner: official_grid_waypoints
167
+ student_goal: final
168
+ evaluation_waypoint_planner: none
169
+ evaluation:
170
+ action_smoothing: 0.95
171
+ seed: 9
172
+ seed_roles:
173
+ optimization: 9
174
+ sampling: 9
175
+ collection: 9420000
176
+ evaluation: 9410000
configs/figure5/large/gru/seed_0/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 0
137
+ seed_roles:
138
+ optimization: 0
139
+ sampling: 0
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_1/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 1
137
+ seed_roles:
138
+ optimization: 1
139
+ sampling: 1
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_10/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 10
137
+ seed_roles:
138
+ optimization: 10
139
+ sampling: 10
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_11/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 11
137
+ seed_roles:
138
+ optimization: 11
139
+ sampling: 11
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_12/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 12
137
+ seed_roles:
138
+ optimization: 12
139
+ sampling: 12
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_13/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 13
137
+ seed_roles:
138
+ optimization: 13
139
+ sampling: 13
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_14/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 14
137
+ seed_roles:
138
+ optimization: 14
139
+ sampling: 14
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_15/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 15
137
+ seed_roles:
138
+ optimization: 15
139
+ sampling: 15
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_16/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 16
137
+ seed_roles:
138
+ optimization: 16
139
+ sampling: 16
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_17/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 17
137
+ seed_roles:
138
+ optimization: 17
139
+ sampling: 17
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_18/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 18
137
+ seed_roles:
138
+ optimization: 18
139
+ sampling: 18
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_19/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 19
137
+ seed_roles:
138
+ optimization: 19
139
+ sampling: 19
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_2/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 2
137
+ seed_roles:
138
+ optimization: 2
139
+ sampling: 2
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_3/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 3
137
+ seed_roles:
138
+ optimization: 3
139
+ sampling: 3
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_4/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 4
137
+ seed_roles:
138
+ optimization: 4
139
+ sampling: 4
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_5/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 5
137
+ seed_roles:
138
+ optimization: 5
139
+ sampling: 5
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_6/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 6
137
+ seed_roles:
138
+ optimization: 6
139
+ sampling: 6
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_7/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 7
137
+ seed_roles:
138
+ optimization: 7
139
+ sampling: 7
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_8/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 8
137
+ seed_roles:
138
+ optimization: 8
139
+ sampling: 8
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/gru/seed_9/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: gru
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 9
137
+ seed_roles:
138
+ optimization: 9
139
+ sampling: 9
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/lstm/seed_0/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: lstm
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 0
137
+ seed_roles:
138
+ optimization: 0
139
+ sampling: 0
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/lstm/seed_1/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: lstm
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 1
137
+ seed_roles:
138
+ optimization: 1
139
+ sampling: 1
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/lstm/seed_10/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: lstm
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 10
137
+ seed_roles:
138
+ optimization: 10
139
+ sampling: 10
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/lstm/seed_11/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: lstm
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 11
137
+ seed_roles:
138
+ optimization: 11
139
+ sampling: 11
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/lstm/seed_12/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: lstm
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 12
137
+ seed_roles:
138
+ optimization: 12
139
+ sampling: 12
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/lstm/seed_13/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: lstm
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 13
137
+ seed_roles:
138
+ optimization: 13
139
+ sampling: 13
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/lstm/seed_14/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: lstm
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 14
137
+ seed_roles:
138
+ optimization: 14
139
+ sampling: 14
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/lstm/seed_15/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: lstm
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 15
137
+ seed_roles:
138
+ optimization: 15
139
+ sampling: 15
140
+ collection: 9420000
141
+ evaluation: 9410000
configs/figure5/large/lstm/seed_16/final.yaml ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ algorithm: antmaze_teacher_distillation
2
+ observation_clip: 60.0
3
+ policy:
4
+ name: lstm
5
+ actor_kwargs:
6
+ goal_router_hidden:
7
+ - 256
8
+ - 256
9
+ goal_router_codebook:
10
+ - - -18.0
11
+ - 12.0
12
+ - - -14.0
13
+ - 12.0
14
+ - - -10.0
15
+ - 12.0
16
+ - - -6.0
17
+ - 12.0
18
+ - - 2.0
19
+ - 12.0
20
+ - - 6.0
21
+ - 12.0
22
+ - - 10.0
23
+ - 12.0
24
+ - - 14.0
25
+ - 12.0
26
+ - - 18.0
27
+ - 12.0
28
+ - - -18.0
29
+ - 8.0
30
+ - - -6.0
31
+ - 8.0
32
+ - - 2.0
33
+ - 8.0
34
+ - - 10.0
35
+ - 8.0
36
+ - - 18.0
37
+ - 8.0
38
+ - - -18.0
39
+ - 4.0
40
+ - - -14.0
41
+ - 4.0
42
+ - - -10.0
43
+ - 4.0
44
+ - - -6.0
45
+ - 4.0
46
+ - - -2.0
47
+ - 4.0
48
+ - - 2.0
49
+ - 4.0
50
+ - - 10.0
51
+ - 4.0
52
+ - - 14.0
53
+ - 4.0
54
+ - - 18.0
55
+ - 4.0
56
+ - - -18.0
57
+ - 0.0
58
+ - - 2.0
59
+ - 0.0
60
+ - - 18.0
61
+ - 0.0
62
+ - - -18.0
63
+ - -4.0
64
+ - - -14.0
65
+ - -4.0
66
+ - - -6.0
67
+ - -4.0
68
+ - - 2.0
69
+ - -4.0
70
+ - - 6.0
71
+ - -4.0
72
+ - - 10.0
73
+ - -4.0
74
+ - - 14.0
75
+ - -4.0
76
+ - - 18.0
77
+ - -4.0
78
+ - - -14.0
79
+ - -8.0
80
+ - - -6.0
81
+ - -8.0
82
+ - - 2.0
83
+ - -8.0
84
+ - - 10.0
85
+ - -8.0
86
+ - - -18.0
87
+ - -12.0
88
+ - - -14.0
89
+ - -12.0
90
+ - - -6.0
91
+ - -12.0
92
+ - - -2.0
93
+ - -12.0
94
+ - - 2.0
95
+ - -12.0
96
+ - - 10.0
97
+ - -12.0
98
+ - - 14.0
99
+ - -12.0
100
+ - - 18.0
101
+ - -12.0
102
+ goal_router_codebook_latch_radius: 0.45
103
+ value_hidden:
104
+ - 256
105
+ - 256
106
+ value_activation: relu
107
+ env:
108
+ class_path: larva.ant_transfer.env.AntMazeTransferEnv
109
+ env_id: AntMaze_Large_Diverse_GR-v5
110
+ goal_feature_mode: world
111
+ continuing_task: false
112
+ resolved_env:
113
+ env_id: AntMaze_Large_Diverse_GR-v5
114
+ goal_feature_mode: world
115
+ continuing_task: false
116
+ imitation:
117
+ demonstration_steps: 60000
118
+ dagger_steps: 30000
119
+ dagger_iterations: 0
120
+ updates_per_stage: 5000
121
+ batch_size: 64
122
+ burn_in: 16
123
+ learning_rate: 0.0001
124
+ gradient_clip: 1.0
125
+ actor_train_mode: base_all
126
+ goal_input_features: 4
127
+ loss_space: pre_tanh
128
+ sampling_mode: episode_uniform
129
+ checkpoint_interval_updates: 500
130
+ navigation:
131
+ waypoint_planner: official_grid_waypoints
132
+ student_goal: final
133
+ evaluation_waypoint_planner: none
134
+ evaluation:
135
+ action_smoothing: 0.95
136
+ seed: 16
137
+ seed_roles:
138
+ optimization: 16
139
+ sampling: 16
140
+ collection: 9420000
141
+ evaluation: 9410000