anhtld commited on
Commit
053a443
·
verified ·
1 Parent(s): 90a3a20

Auto-sync: 2026-06-27 11:24:43

Browse files
results/README.md CHANGED
@@ -15,6 +15,7 @@ Use these summaries for the current paper tables:
15
  - `h16_policy_ckpt_near_miss_policy_field_k32_sigma0p35_summary.md`
16
  - `h16_policy_ckpt_near_miss_policy_fieldckpt_field_k32_sigma0p35_summary.md`
17
  - `h16_policy_ckpt_near_miss_policy_bc5_fieldckpt_field_k32_sigma0p35_summary.md`
 
18
 
19
  Known stale or incomplete files:
20
 
@@ -26,8 +27,8 @@ Known stale or incomplete files:
26
  absent/incomplete because the final seed was canceled after the stronger `best.pt`
27
  field-checkpoint variant completed.
28
 
29
- Pending:
30
 
31
- - Job `14837136` is sweeping K/sigma for the current best deployment-clean route
32
- (`near_miss_policy_bc5` + `best.pt` + Gaussian field selection). Summary job:
33
- `14837137`.
 
15
  - `h16_policy_ckpt_near_miss_policy_field_k32_sigma0p35_summary.md`
16
  - `h16_policy_ckpt_near_miss_policy_fieldckpt_field_k32_sigma0p35_summary.md`
17
  - `h16_policy_ckpt_near_miss_policy_bc5_fieldckpt_field_k32_sigma0p35_summary.md`
18
+ - `h16_policy_ckpt_near_miss_policy_bc5_bestpt_field_sweep_summary.md`
19
 
20
  Known stale or incomplete files:
21
 
 
27
  absent/incomplete because the final seed was canceled after the stronger `best.pt`
28
  field-checkpoint variant completed.
29
 
30
+ Completed sweep:
31
 
32
+ - Job `14837136` swept K/sigma for the current best deployment-clean route
33
+ (`near_miss_policy_bc5` + `best.pt` + Gaussian field selection). Summary job
34
+ `14837137` wrote `h16_policy_ckpt_near_miss_policy_bc5_bestpt_field_sweep_summary.*`.
results/dovla_cil_run_report_2026-06-27.md CHANGED
@@ -13,7 +13,7 @@ The deployed h=16 behavior-cloning policy remains weak, but the interventional f
13
  | Near-miss distillation policy, BC x5 | 28.29% | -1.45 pp | Stronger BC improves slightly but remains below baseline |
14
  | Near-miss proposal + field, best-policy checkpoint | 26.32% | -3.42 pp | Field scoring around the BC-selected checkpoint is worse |
15
  | Near-miss proposal + field, field checkpoint | 30.14% | +0.41 pp | Deployment-clean route begins to recover the mechanism |
16
- | Near-miss proposal + field, BC x5 field checkpoint | 32.64% | +2.90 pp | Best deployment-clean result so far |
17
  | Lattice field selection, no expert candidates | 56.99% | +27.25 pp | Conservative, reviewer-safe result |
18
  | Lattice field selection, near-miss only | 55.94% | +26.20 pp | Minimal local counterfactual proposal family |
19
  | Lattice field selection, no expert and no near-miss | 25.57% | -4.17 pp | Confirms near-miss proposals carry the gain |
@@ -36,6 +36,7 @@ Canonical summaries:
36
  - `results/h16_policy_ckpt_near_miss_policy_field_k32_sigma0p35_summary.md`
37
  - `results/h16_policy_ckpt_near_miss_policy_fieldckpt_field_k32_sigma0p35_summary.md`
38
  - `results/h16_policy_ckpt_near_miss_policy_bc5_fieldckpt_field_k32_sigma0p35_summary.md`
 
39
 
40
  Non-canonical/stale summaries:
41
 
@@ -70,8 +71,8 @@ Non-canonical/stale summaries:
70
  | 14836240 | Summarize near-miss `best.pt` field selection | Completed |
71
  | 14836241 | Evaluate near-miss BC x5 `best.pt` with field Gaussian selection | Completed |
72
  | 14836242 | Summarize near-miss BC x5 `best.pt` field selection | Completed |
73
- | 14837136 | Sweep near-miss BC x5 `best.pt` field selection over K/sigma | Pending |
74
- | 14837137 | Summarize near-miss BC x5 `best.pt` field sweep | Pending |
75
 
76
  ## Interpretation
77
 
@@ -93,7 +94,7 @@ Deployment-clean distillation is partially but not fully successful. Directly tr
93
  policy to imitate near-miss candidates stays below baseline (27.48%, or 28.29% with BC x5).
94
  Using the BC-selected `best_policy.pt` checkpoint as a Gaussian proposal center also fails
95
  (26.32%). However, selecting the field-ranked checkpoint `best.pt` and increasing BC weight
96
- does produce a small clean gain: 32.64%, +2.90 pp over the h=16 policy. This is promising
97
  as a deployment path, but it is still far below the same-state no-expert lattice result
98
  (56.99%), so the paper should frame it as evidence for the proposal bottleneck rather than
99
  as the main result.
@@ -107,11 +108,10 @@ as the main result.
107
  - Negative but useful ablation: Gaussian field search fails, showing the field is not a generic black-box action optimizer off the data manifold.
108
  - Negative but useful ablation: nearest train-state lattice retrieval fails, showing the gain is not from a generic action library.
109
  - Negative checkpoint ablation: selecting checkpoint by lower BC validation loss does not improve online rollout.
110
- - Deployment-clean positive control: near-miss proposal policy plus field selection reaches 32.64% when using the field-selected checkpoint and BC x5, but the gap to 56.99% shows that proposal quality remains the bottleneck.
111
 
112
  ## Next Best Experiments
113
 
114
  1. Generate an on-the-fly non-expert proposal lattice that does not depend on stored validation candidates, then evaluate with `selection_mode=lattice`.
115
- 2. Finish job `14837136` to tune K/sigma for the promising near-miss BC x5 `best.pt` proposal+field route.
116
- 3. Train a proposal model to imitate successful non-expert lattice candidates so deployment does not require an explicit lattice store.
117
- 4. Add final paper tables separating `policy`, `proposal_field`, `field_lattice_no_expert`, `field_lattice_full`, and `oracle`.
 
13
  | Near-miss distillation policy, BC x5 | 28.29% | -1.45 pp | Stronger BC improves slightly but remains below baseline |
14
  | Near-miss proposal + field, best-policy checkpoint | 26.32% | -3.42 pp | Field scoring around the BC-selected checkpoint is worse |
15
  | Near-miss proposal + field, field checkpoint | 30.14% | +0.41 pp | Deployment-clean route begins to recover the mechanism |
16
+ | Near-miss proposal + field, BC x5 field checkpoint | 32.93% | +3.19 pp | Best deployment-clean result so far (`k64_sigma0.50`) |
17
  | Lattice field selection, no expert candidates | 56.99% | +27.25 pp | Conservative, reviewer-safe result |
18
  | Lattice field selection, near-miss only | 55.94% | +26.20 pp | Minimal local counterfactual proposal family |
19
  | Lattice field selection, no expert and no near-miss | 25.57% | -4.17 pp | Confirms near-miss proposals carry the gain |
 
36
  - `results/h16_policy_ckpt_near_miss_policy_field_k32_sigma0p35_summary.md`
37
  - `results/h16_policy_ckpt_near_miss_policy_fieldckpt_field_k32_sigma0p35_summary.md`
38
  - `results/h16_policy_ckpt_near_miss_policy_bc5_fieldckpt_field_k32_sigma0p35_summary.md`
39
+ - `results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_field_sweep_summary.md`
40
 
41
  Non-canonical/stale summaries:
42
 
 
71
  | 14836240 | Summarize near-miss `best.pt` field selection | Completed |
72
  | 14836241 | Evaluate near-miss BC x5 `best.pt` with field Gaussian selection | Completed |
73
  | 14836242 | Summarize near-miss BC x5 `best.pt` field selection | Completed |
74
+ | 14837136 | Sweep near-miss BC x5 `best.pt` field selection over K/sigma | Completed |
75
+ | 14837137 | Summarize near-miss BC x5 `best.pt` field sweep | Completed |
76
 
77
  ## Interpretation
78
 
 
94
  policy to imitate near-miss candidates stays below baseline (27.48%, or 28.29% with BC x5).
95
  Using the BC-selected `best_policy.pt` checkpoint as a Gaussian proposal center also fails
96
  (26.32%). However, selecting the field-ranked checkpoint `best.pt` and increasing BC weight
97
+ does produce a small clean gain: 32.93%, +3.19 pp over the h=16 policy. This is promising
98
  as a deployment path, but it is still far below the same-state no-expert lattice result
99
  (56.99%), so the paper should frame it as evidence for the proposal bottleneck rather than
100
  as the main result.
 
108
  - Negative but useful ablation: Gaussian field search fails, showing the field is not a generic black-box action optimizer off the data manifold.
109
  - Negative but useful ablation: nearest train-state lattice retrieval fails, showing the gain is not from a generic action library.
110
  - Negative checkpoint ablation: selecting checkpoint by lower BC validation loss does not improve online rollout.
111
+ - Deployment-clean positive control: near-miss proposal policy plus field selection reaches 32.93% when using the field-selected checkpoint and BC x5, but the gap to 56.99% shows that proposal quality remains the bottleneck.
112
 
113
  ## Next Best Experiments
114
 
115
  1. Generate an on-the-fly non-expert proposal lattice that does not depend on stored validation candidates, then evaluate with `selection_mode=lattice`.
116
+ 2. Train a proposal model to imitate successful non-expert lattice candidates so deployment does not require an explicit lattice store.
117
+ 3. Add final paper tables separating `policy`, `proposal_field`, `field_lattice_no_expert`, `field_lattice_full`, and `oracle`.
 
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_field_sweep_summary.json ADDED
@@ -0,0 +1,1150 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt",
3
+ "baseline_h4_policy_success": 0.2967,
4
+ "baseline_h16_policy_success": 0.29739130434782607,
5
+ "num_result_files": 15,
6
+ "configs": [
7
+ {
8
+ "config": "k64_sigma0.50",
9
+ "completed_seeds": [
10
+ 0,
11
+ 1,
12
+ 2
13
+ ],
14
+ "num_completed": 3,
15
+ "mean_success": 0.3292753623188406,
16
+ "std_success": 0.01233843284277841,
17
+ "mean_progress": 0.5483272921758826,
18
+ "mean_action_mse_to_best": 0.4194377801362155,
19
+ "absolute_gain_vs_h4_baseline": 0.03257536231884056,
20
+ "absolute_gain_vs_h16_policy": 0.03188405797101451
21
+ },
22
+ {
23
+ "config": "k32_sigma0.35",
24
+ "completed_seeds": [
25
+ 0,
26
+ 1,
27
+ 2
28
+ ],
29
+ "num_completed": 3,
30
+ "mean_success": 0.32811594202898553,
31
+ "std_success": 0.016159257814221877,
32
+ "mean_progress": 0.5461846023357074,
33
+ "mean_action_mse_to_best": 0.4120204289451889,
34
+ "absolute_gain_vs_h4_baseline": 0.03141594202898551,
35
+ "absolute_gain_vs_h16_policy": 0.03072463768115946
36
+ },
37
+ {
38
+ "config": "k64_sigma0.35",
39
+ "completed_seeds": [
40
+ 0,
41
+ 1,
42
+ 2
43
+ ],
44
+ "num_completed": 3,
45
+ "mean_success": 0.32753623188405795,
46
+ "std_success": 0.020306411464506676,
47
+ "mean_progress": 0.5457847676730542,
48
+ "mean_action_mse_to_best": 0.4149229990820522,
49
+ "absolute_gain_vs_h4_baseline": 0.03083623188405793,
50
+ "absolute_gain_vs_h16_policy": 0.03014492753623188
51
+ },
52
+ {
53
+ "config": "k16_sigma0.20",
54
+ "completed_seeds": [
55
+ 0,
56
+ 1,
57
+ 2
58
+ ],
59
+ "num_completed": 3,
60
+ "mean_success": 0.3263768115942029,
61
+ "std_success": 0.01859595934184978,
62
+ "mean_progress": 0.5471449069725687,
63
+ "mean_action_mse_to_best": 0.40453435571999224,
64
+ "absolute_gain_vs_h4_baseline": 0.02967681159420288,
65
+ "absolute_gain_vs_h16_policy": 0.02898550724637683
66
+ },
67
+ {
68
+ "config": "k32_sigma0.20",
69
+ "completed_seeds": [
70
+ 0,
71
+ 1,
72
+ 2
73
+ ],
74
+ "num_completed": 3,
75
+ "mean_success": 0.32231884057971016,
76
+ "std_success": 0.020748440774693663,
77
+ "mean_progress": 0.5414390760636103,
78
+ "mean_action_mse_to_best": 0.40601564239008703,
79
+ "absolute_gain_vs_h4_baseline": 0.025618840579710145,
80
+ "absolute_gain_vs_h16_policy": 0.024927536231884095
81
+ }
82
+ ],
83
+ "best": {
84
+ "config": "k64_sigma0.50",
85
+ "completed_seeds": [
86
+ 0,
87
+ 1,
88
+ 2
89
+ ],
90
+ "num_completed": 3,
91
+ "mean_success": 0.3292753623188406,
92
+ "std_success": 0.01233843284277841,
93
+ "mean_progress": 0.5483272921758826,
94
+ "mean_action_mse_to_best": 0.4194377801362155,
95
+ "absolute_gain_vs_h4_baseline": 0.03257536231884056,
96
+ "absolute_gain_vs_h16_policy": 0.03188405797101451
97
+ },
98
+ "rows": [
99
+ {
100
+ "config": "k16_sigma0.20",
101
+ "seed": 0,
102
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k16_sigma0.20/seed_0/online_rollout.json",
103
+ "selection_mode": "field",
104
+ "num_candidates": 16,
105
+ "candidate_sigma": 0.2,
106
+ "num_groups": 575,
107
+ "policy_rollout_success_rate": 0.33043478260869563,
108
+ "policy_rollout_progress": 0.544193181817764,
109
+ "oracle_success_rate": 0.8573913043478261,
110
+ "action_mse_to_best": 0.3899358766955202,
111
+ "per_task": {
112
+ "LiftPegUpright-v1": {
113
+ "action_mse_to_best": 0.32247859436884213,
114
+ "expert_success_rate": 0.8865979381443299,
115
+ "num_groups": 97,
116
+ "oracle_success_rate": 0.9278350515463918,
117
+ "policy_expert_regret": 1.07642500809173,
118
+ "policy_oracle_regret": 1.1028404858001728,
119
+ "policy_rollout_progress": 0.5733718855172089,
120
+ "policy_rollout_success_rate": 0.2268041237113402,
121
+ "restore_max_error": 4.76837158203125e-07
122
+ },
123
+ "PickCube-v1": {
124
+ "action_mse_to_best": 0.3079392760633849,
125
+ "expert_success_rate": 0.9375,
126
+ "num_groups": 208,
127
+ "oracle_success_rate": 0.9471153846153846,
128
+ "policy_expert_regret": 0.9954389036507704,
129
+ "policy_oracle_regret": 1.0088617710772758,
130
+ "policy_rollout_progress": 0.601058509406777,
131
+ "policy_rollout_success_rate": 0.32211538461538464,
132
+ "restore_max_error": 4.76837158203125e-07
133
+ },
134
+ "PullCube-v1": {
135
+ "action_mse_to_best": 0.6287145626235318,
136
+ "expert_success_rate": 0.19480519480519481,
137
+ "num_groups": 77,
138
+ "oracle_success_rate": 0.36363636363636365,
139
+ "policy_expert_regret": 0.28835389711485887,
140
+ "policy_oracle_regret": 0.4758843815524143,
141
+ "policy_rollout_progress": 0.31379556818789645,
142
+ "policy_rollout_success_rate": 0.19480519480519481,
143
+ "restore_max_error": 4.76837158203125e-07
144
+ },
145
+ "PushCube-v1": {
146
+ "action_mse_to_best": 0.3850305494158826,
147
+ "expert_success_rate": 0.8617021276595744,
148
+ "num_groups": 94,
149
+ "oracle_success_rate": 0.9893617021276596,
150
+ "policy_expert_regret": 0.5150462303389894,
151
+ "policy_oracle_regret": 0.6138637159416016,
152
+ "policy_rollout_progress": 0.6964525897452172,
153
+ "policy_rollout_success_rate": 0.6702127659574468,
154
+ "restore_max_error": 4.76837158203125e-07
155
+ },
156
+ "StackCube-v1": {
157
+ "action_mse_to_best": 0.44724689957431774,
158
+ "expert_success_rate": 0.6767676767676768,
159
+ "num_groups": 99,
160
+ "oracle_success_rate": 0.8585858585858586,
161
+ "policy_expert_regret": 0.9836689957765617,
162
+ "policy_oracle_regret": 1.157828590334064,
163
+ "policy_rollout_progress": 0.43075792431229293,
164
+ "policy_rollout_success_rate": 0.23232323232323232,
165
+ "restore_max_error": 3.948807716369629e-07
166
+ }
167
+ }
168
+ },
169
+ {
170
+ "config": "k16_sigma0.20",
171
+ "seed": 1,
172
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k16_sigma0.20/seed_1/online_rollout.json",
173
+ "selection_mode": "field",
174
+ "num_candidates": 16,
175
+ "candidate_sigma": 0.2,
176
+ "num_groups": 575,
177
+ "policy_rollout_success_rate": 0.3060869565217391,
178
+ "policy_rollout_progress": 0.5316903001021431,
179
+ "oracle_success_rate": 0.8695652173913043,
180
+ "action_mse_to_best": 0.39861446821495244,
181
+ "per_task": {
182
+ "LiftPegUpright-v1": {
183
+ "action_mse_to_best": 0.33289363966282226,
184
+ "expert_success_rate": 0.8584070796460177,
185
+ "num_groups": 113,
186
+ "oracle_success_rate": 0.9380530973451328,
187
+ "policy_expert_regret": 1.048153557608613,
188
+ "policy_oracle_regret": 1.1386048915111913,
189
+ "policy_rollout_progress": 0.5989050643633952,
190
+ "policy_rollout_success_rate": 0.19469026548672566,
191
+ "restore_max_error": 4.76837158203125e-07
192
+ },
193
+ "PickCube-v1": {
194
+ "action_mse_to_best": 0.3264182775443339,
195
+ "expert_success_rate": 0.9402173913043478,
196
+ "num_groups": 184,
197
+ "oracle_success_rate": 0.9456521739130435,
198
+ "policy_expert_regret": 1.1100710043076263,
199
+ "policy_oracle_regret": 1.1100950087825565,
200
+ "policy_rollout_progress": 0.5545998569780394,
201
+ "policy_rollout_success_rate": 0.266304347826087,
202
+ "restore_max_error": 4.76837158203125e-07
203
+ },
204
+ "PullCube-v1": {
205
+ "action_mse_to_best": 0.5894409823966654,
206
+ "expert_success_rate": 0.25,
207
+ "num_groups": 76,
208
+ "oracle_success_rate": 0.40789473684210525,
209
+ "policy_expert_regret": 0.3966844152708195,
210
+ "policy_oracle_regret": 0.66795516218745,
211
+ "policy_rollout_progress": 0.2253140292596072,
212
+ "policy_rollout_success_rate": 0.10526315789473684,
213
+ "restore_max_error": 4.76837158203125e-07
214
+ },
215
+ "PushCube-v1": {
216
+ "action_mse_to_best": 0.3773898981377348,
217
+ "expert_success_rate": 0.8198198198198198,
218
+ "num_groups": 111,
219
+ "oracle_success_rate": 1.0,
220
+ "policy_expert_regret": 0.3984359223563392,
221
+ "policy_oracle_regret": 0.4819212639117026,
222
+ "policy_rollout_progress": 0.7703309883405497,
223
+ "policy_rollout_success_rate": 0.7477477477477478,
224
+ "restore_max_error": 4.76837158203125e-07
225
+ },
226
+ "StackCube-v1": {
227
+ "action_mse_to_best": 0.4927206760231446,
228
+ "expert_success_rate": 0.6923076923076923,
229
+ "num_groups": 91,
230
+ "oracle_success_rate": 0.8571428571428571,
231
+ "policy_expert_regret": 1.1646305985830643,
232
+ "policy_oracle_regret": 1.3239166384542382,
233
+ "policy_rollout_progress": 0.36668868870525567,
234
+ "policy_rollout_success_rate": 0.15384615384615385,
235
+ "restore_max_error": 3.948807716369629e-07
236
+ }
237
+ }
238
+ },
239
+ {
240
+ "config": "k16_sigma0.20",
241
+ "seed": 2,
242
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k16_sigma0.20/seed_2/online_rollout.json",
243
+ "selection_mode": "field",
244
+ "num_candidates": 16,
245
+ "candidate_sigma": 0.2,
246
+ "num_groups": 575,
247
+ "policy_rollout_success_rate": 0.3426086956521739,
248
+ "policy_rollout_progress": 0.5655512389977989,
249
+ "oracle_success_rate": 0.8765217391304347,
250
+ "action_mse_to_best": 0.425052722249504,
251
+ "per_task": {
252
+ "LiftPegUpright-v1": {
253
+ "action_mse_to_best": 0.3572826692106901,
254
+ "expert_success_rate": 0.8229166666666666,
255
+ "num_groups": 96,
256
+ "oracle_success_rate": 0.9270833333333334,
257
+ "policy_expert_regret": 0.8819095164847871,
258
+ "policy_oracle_regret": 0.9881842317990959,
259
+ "policy_rollout_progress": 0.6433409581271311,
260
+ "policy_rollout_success_rate": 0.3020833333333333,
261
+ "restore_max_error": 3.5762786865234375e-07
262
+ },
263
+ "PickCube-v1": {
264
+ "action_mse_to_best": 0.33731283482653324,
265
+ "expert_success_rate": 0.9444444444444444,
266
+ "num_groups": 198,
267
+ "oracle_success_rate": 0.9595959595959596,
268
+ "policy_expert_regret": 1.0088076980295355,
269
+ "policy_oracle_regret": 1.015891148558244,
270
+ "policy_rollout_progress": 0.6228325308554552,
271
+ "policy_rollout_success_rate": 0.32323232323232326,
272
+ "restore_max_error": 4.76837158203125e-07
273
+ },
274
+ "PullCube-v1": {
275
+ "action_mse_to_best": 0.6599690684013897,
276
+ "expert_success_rate": 0.24444444444444444,
277
+ "num_groups": 90,
278
+ "oracle_success_rate": 0.4666666666666667,
279
+ "policy_expert_regret": 0.3202306208821634,
280
+ "policy_oracle_regret": 0.5816346477613681,
281
+ "policy_rollout_progress": 0.31432361036746037,
282
+ "policy_rollout_success_rate": 0.2,
283
+ "restore_max_error": 4.0978193283081055e-07
284
+ },
285
+ "PushCube-v1": {
286
+ "action_mse_to_best": 0.39746504446657577,
287
+ "expert_success_rate": 0.8514851485148515,
288
+ "num_groups": 101,
289
+ "oracle_success_rate": 1.0,
290
+ "policy_expert_regret": 0.4742610312924527,
291
+ "policy_oracle_regret": 0.5289383319639923,
292
+ "policy_rollout_progress": 0.7482893908082848,
293
+ "policy_rollout_success_rate": 0.7227722772277227,
294
+ "restore_max_error": 4.76837158203125e-07
295
+ },
296
+ "StackCube-v1": {
297
+ "action_mse_to_best": 0.4864116900703973,
298
+ "expert_success_rate": 0.7666666666666667,
299
+ "num_groups": 90,
300
+ "oracle_success_rate": 0.9111111111111111,
301
+ "policy_expert_regret": 1.1857868118418589,
302
+ "policy_oracle_regret": 1.3348589769668049,
303
+ "policy_rollout_progress": 0.40271151032712726,
304
+ "policy_rollout_success_rate": 0.14444444444444443,
305
+ "restore_max_error": 4.76837158203125e-07
306
+ }
307
+ }
308
+ },
309
+ {
310
+ "config": "k32_sigma0.20",
311
+ "seed": 0,
312
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k32_sigma0.20/seed_0/online_rollout.json",
313
+ "selection_mode": "field",
314
+ "num_candidates": 32,
315
+ "candidate_sigma": 0.2,
316
+ "num_groups": 575,
317
+ "policy_rollout_success_rate": 0.30782608695652175,
318
+ "policy_rollout_progress": 0.5236520005676198,
319
+ "oracle_success_rate": 0.8573913043478261,
320
+ "action_mse_to_best": 0.3907121669816906,
321
+ "per_task": {
322
+ "LiftPegUpright-v1": {
323
+ "action_mse_to_best": 0.32411609940979746,
324
+ "expert_success_rate": 0.8865979381443299,
325
+ "num_groups": 97,
326
+ "oracle_success_rate": 0.9278350515463918,
327
+ "policy_expert_regret": 1.096901917734097,
328
+ "policy_oracle_regret": 1.12331739544254,
329
+ "policy_rollout_progress": 0.5632042542253572,
330
+ "policy_rollout_success_rate": 0.21649484536082475,
331
+ "restore_max_error": 4.76837158203125e-07
332
+ },
333
+ "PickCube-v1": {
334
+ "action_mse_to_best": 0.3089638287559725,
335
+ "expert_success_rate": 0.9375,
336
+ "num_groups": 208,
337
+ "oracle_success_rate": 0.9471153846153846,
338
+ "policy_expert_regret": 0.9997012414521753,
339
+ "policy_oracle_regret": 1.0131241088786807,
340
+ "policy_rollout_progress": 0.594855451822066,
341
+ "policy_rollout_success_rate": 0.3173076923076923,
342
+ "restore_max_error": 4.76837158203125e-07
343
+ },
344
+ "PullCube-v1": {
345
+ "action_mse_to_best": 0.620038845709392,
346
+ "expert_success_rate": 0.19480519480519481,
347
+ "num_groups": 77,
348
+ "oracle_success_rate": 0.36363636363636365,
349
+ "policy_expert_regret": 0.34222419101954143,
350
+ "policy_oracle_regret": 0.6057183034637363,
351
+ "policy_rollout_progress": 0.21818976864638404,
352
+ "policy_rollout_success_rate": 0.09090909090909091,
353
+ "restore_max_error": 4.76837158203125e-07
354
+ },
355
+ "PushCube-v1": {
356
+ "action_mse_to_best": 0.3883775261012798,
357
+ "expert_success_rate": 0.8617021276595744,
358
+ "num_groups": 94,
359
+ "oracle_success_rate": 0.9893617021276596,
360
+ "policy_expert_regret": 0.5517117910562678,
361
+ "policy_oracle_regret": 0.6486242381816215,
362
+ "policy_rollout_progress": 0.6828119799177698,
363
+ "policy_rollout_success_rate": 0.648936170212766,
364
+ "restore_max_error": 4.76837158203125e-07
365
+ },
366
+ "StackCube-v1": {
367
+ "action_mse_to_best": 0.4515684789629898,
368
+ "expert_success_rate": 0.6767676767676768,
369
+ "num_groups": 99,
370
+ "oracle_success_rate": 0.8585858585858586,
371
+ "policy_expert_regret": 1.0139289311688355,
372
+ "policy_oracle_regret": 1.1769277733383756,
373
+ "policy_rollout_progress": 0.4217597514089912,
374
+ "policy_rollout_success_rate": 0.2222222222222222,
375
+ "restore_max_error": 3.948807716369629e-07
376
+ }
377
+ }
378
+ },
379
+ {
380
+ "config": "k32_sigma0.20",
381
+ "seed": 1,
382
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k32_sigma0.20/seed_1/online_rollout.json",
383
+ "selection_mode": "field",
384
+ "num_candidates": 32,
385
+ "candidate_sigma": 0.2,
386
+ "num_groups": 575,
387
+ "policy_rollout_success_rate": 0.3130434782608696,
388
+ "policy_rollout_progress": 0.5412584817804316,
389
+ "oracle_success_rate": 0.8695652173913043,
390
+ "action_mse_to_best": 0.4014194884442765,
391
+ "per_task": {
392
+ "LiftPegUpright-v1": {
393
+ "action_mse_to_best": 0.3384467677122592,
394
+ "expert_success_rate": 0.8584070796460177,
395
+ "num_groups": 113,
396
+ "oracle_success_rate": 0.9380530973451328,
397
+ "policy_expert_regret": 1.025868605855292,
398
+ "policy_oracle_regret": 1.105070760951633,
399
+ "policy_rollout_progress": 0.6147400798787058,
400
+ "policy_rollout_success_rate": 0.21238938053097345,
401
+ "restore_max_error": 4.76837158203125e-07
402
+ },
403
+ "PickCube-v1": {
404
+ "action_mse_to_best": 0.32736685282915184,
405
+ "expert_success_rate": 0.9402173913043478,
406
+ "num_groups": 184,
407
+ "oracle_success_rate": 0.9456521739130435,
408
+ "policy_expert_regret": 1.11036436841317,
409
+ "policy_oracle_regret": 1.1103883728881,
410
+ "policy_rollout_progress": 0.552188413313833,
411
+ "policy_rollout_success_rate": 0.2608695652173913,
412
+ "restore_max_error": 4.76837158203125e-07
413
+ },
414
+ "PullCube-v1": {
415
+ "action_mse_to_best": 0.593048064920463,
416
+ "expert_success_rate": 0.25,
417
+ "num_groups": 76,
418
+ "oracle_success_rate": 0.40789473684210525,
419
+ "policy_expert_regret": 0.29840909580142533,
420
+ "policy_oracle_regret": 0.5478567443129386,
421
+ "policy_rollout_progress": 0.28396381365937967,
422
+ "policy_rollout_success_rate": 0.17105263157894737,
423
+ "restore_max_error": 4.76837158203125e-07
424
+ },
425
+ "PushCube-v1": {
426
+ "action_mse_to_best": 0.3798570418814281,
427
+ "expert_success_rate": 0.8198198198198198,
428
+ "num_groups": 111,
429
+ "oracle_success_rate": 1.0,
430
+ "policy_expert_regret": 0.4019931071513408,
431
+ "policy_oracle_regret": 0.4871569598043287,
432
+ "policy_rollout_progress": 0.7650952924479235,
433
+ "policy_rollout_success_rate": 0.7477477477477478,
434
+ "restore_max_error": 4.76837158203125e-07
435
+ },
436
+ "StackCube-v1": {
437
+ "action_mse_to_best": 0.4956091824243521,
438
+ "expert_success_rate": 0.6923076923076923,
439
+ "num_groups": 91,
440
+ "oracle_success_rate": 0.8571428571428571,
441
+ "policy_expert_regret": 1.183210695019135,
442
+ "policy_oracle_regret": 1.342819583776233,
443
+ "policy_rollout_progress": 0.36976376536128286,
444
+ "policy_rollout_success_rate": 0.13186813186813187,
445
+ "restore_max_error": 3.948807716369629e-07
446
+ }
447
+ }
448
+ },
449
+ {
450
+ "config": "k32_sigma0.20",
451
+ "seed": 2,
452
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k32_sigma0.20/seed_2/online_rollout.json",
453
+ "selection_mode": "field",
454
+ "num_candidates": 32,
455
+ "candidate_sigma": 0.2,
456
+ "num_groups": 575,
457
+ "policy_rollout_success_rate": 0.34608695652173915,
458
+ "policy_rollout_progress": 0.5594067458427795,
459
+ "oracle_success_rate": 0.8765217391304347,
460
+ "action_mse_to_best": 0.425915271744294,
461
+ "per_task": {
462
+ "LiftPegUpright-v1": {
463
+ "action_mse_to_best": 0.35779982555444195,
464
+ "expert_success_rate": 0.8229166666666666,
465
+ "num_groups": 96,
466
+ "oracle_success_rate": 0.9270833333333334,
467
+ "policy_expert_regret": 0.8545110411942005,
468
+ "policy_oracle_regret": 0.9607857565085093,
469
+ "policy_rollout_progress": 0.6394894334177176,
470
+ "policy_rollout_success_rate": 0.3333333333333333,
471
+ "restore_max_error": 3.5762786865234375e-07
472
+ },
473
+ "PickCube-v1": {
474
+ "action_mse_to_best": 0.3381687065755779,
475
+ "expert_success_rate": 0.9444444444444444,
476
+ "num_groups": 198,
477
+ "oracle_success_rate": 0.9595959595959596,
478
+ "policy_expert_regret": 1.00870233607676,
479
+ "policy_oracle_regret": 1.0157857866054683,
480
+ "policy_rollout_progress": 0.6229378928082309,
481
+ "policy_rollout_success_rate": 0.32323232323232326,
482
+ "restore_max_error": 4.76837158203125e-07
483
+ },
484
+ "PullCube-v1": {
485
+ "action_mse_to_best": 0.6543993772731886,
486
+ "expert_success_rate": 0.24444444444444444,
487
+ "num_groups": 90,
488
+ "oracle_success_rate": 0.4666666666666667,
489
+ "policy_expert_regret": 0.3404962347147779,
490
+ "policy_oracle_regret": 0.5860447995105965,
491
+ "policy_rollout_progress": 0.3017854992972894,
492
+ "policy_rollout_success_rate": 0.18888888888888888,
493
+ "restore_max_error": 4.0978193283081055e-07
494
+ },
495
+ "PushCube-v1": {
496
+ "action_mse_to_best": 0.4016442125534067,
497
+ "expert_success_rate": 0.8514851485148515,
498
+ "num_groups": 101,
499
+ "oracle_success_rate": 1.0,
500
+ "policy_expert_regret": 0.47577363680494894,
501
+ "policy_oracle_regret": 0.5299919200415658,
502
+ "policy_rollout_progress": 0.7472358027307113,
503
+ "policy_rollout_success_rate": 0.7227722772277227,
504
+ "restore_max_error": 4.76837158203125e-07
505
+ },
506
+ "StackCube-v1": {
507
+ "action_mse_to_best": 0.49036760750330155,
508
+ "expert_success_rate": 0.7666666666666667,
509
+ "num_groups": 90,
510
+ "oracle_success_rate": 0.9111111111111111,
511
+ "policy_expert_regret": 1.1946870706147619,
512
+ "policy_oracle_regret": 1.356807397802671,
513
+ "policy_rollout_progress": 0.38105199403233,
514
+ "policy_rollout_success_rate": 0.14444444444444443,
515
+ "restore_max_error": 4.76837158203125e-07
516
+ }
517
+ }
518
+ },
519
+ {
520
+ "config": "k32_sigma0.35",
521
+ "seed": 0,
522
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k32_sigma0.35/seed_0/online_rollout.json",
523
+ "selection_mode": "field",
524
+ "num_candidates": 32,
525
+ "candidate_sigma": 0.35,
526
+ "num_groups": 575,
527
+ "policy_rollout_success_rate": 0.3234782608695652,
528
+ "policy_rollout_progress": 0.5365680560134893,
529
+ "oracle_success_rate": 0.8573913043478261,
530
+ "action_mse_to_best": 0.39553123654797673,
531
+ "per_task": {
532
+ "LiftPegUpright-v1": {
533
+ "action_mse_to_best": 0.3222495113672285,
534
+ "expert_success_rate": 0.8865979381443299,
535
+ "num_groups": 97,
536
+ "oracle_success_rate": 0.9278350515463918,
537
+ "policy_expert_regret": 1.0924583412322801,
538
+ "policy_oracle_regret": 1.118873818940723,
539
+ "policy_rollout_progress": 0.5676478307271741,
540
+ "policy_rollout_success_rate": 0.21649484536082475,
541
+ "restore_max_error": 4.76837158203125e-07
542
+ },
543
+ "PickCube-v1": {
544
+ "action_mse_to_best": 0.31544953831829703,
545
+ "expert_success_rate": 0.9375,
546
+ "num_groups": 208,
547
+ "oracle_success_rate": 0.9471153846153846,
548
+ "policy_expert_regret": 0.9939068040470235,
549
+ "policy_oracle_regret": 1.007329671473529,
550
+ "policy_rollout_progress": 0.5977829167028316,
551
+ "policy_rollout_success_rate": 0.3269230769230769,
552
+ "restore_max_error": 4.76837158203125e-07
553
+ },
554
+ "PullCube-v1": {
555
+ "action_mse_to_best": 0.6550499460139831,
556
+ "expert_success_rate": 0.19480519480519481,
557
+ "num_groups": 77,
558
+ "oracle_success_rate": 0.36363636363636365,
559
+ "policy_expert_regret": 0.2455574702068305,
560
+ "policy_oracle_regret": 0.4835427300238486,
561
+ "policy_rollout_progress": 0.29021572308752125,
562
+ "policy_rollout_success_rate": 0.16883116883116883,
563
+ "restore_max_error": 4.76837158203125e-07
564
+ },
565
+ "PushCube-v1": {
566
+ "action_mse_to_best": 0.3813233734603892,
567
+ "expert_success_rate": 0.8617021276595744,
568
+ "num_groups": 94,
569
+ "oracle_success_rate": 0.9893617021276596,
570
+ "policy_expert_regret": 0.5716793151929024,
571
+ "policy_oracle_regret": 0.6685689863689402,
572
+ "policy_rollout_progress": 0.6824948434182938,
573
+ "policy_rollout_success_rate": 0.648936170212766,
574
+ "restore_max_error": 4.76837158203125e-07
575
+ },
576
+ "StackCube-v1": {
577
+ "action_mse_to_best": 0.44722738882733715,
578
+ "expert_success_rate": 0.6767676767676768,
579
+ "num_groups": 99,
580
+ "oracle_success_rate": 0.8585858585858586,
581
+ "policy_expert_regret": 0.9838731827759983,
582
+ "policy_oracle_regret": 1.1580327773335004,
583
+ "policy_rollout_progress": 0.43055373731285634,
584
+ "policy_rollout_success_rate": 0.23232323232323232,
585
+ "restore_max_error": 3.948807716369629e-07
586
+ }
587
+ }
588
+ },
589
+ {
590
+ "config": "k32_sigma0.35",
591
+ "seed": 1,
592
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k32_sigma0.35/seed_1/online_rollout.json",
593
+ "selection_mode": "field",
594
+ "num_candidates": 32,
595
+ "candidate_sigma": 0.35,
596
+ "num_groups": 575,
597
+ "policy_rollout_success_rate": 0.31478260869565217,
598
+ "policy_rollout_progress": 0.5412338370824422,
599
+ "oracle_success_rate": 0.8695652173913043,
600
+ "action_mse_to_best": 0.4067689944188232,
601
+ "per_task": {
602
+ "LiftPegUpright-v1": {
603
+ "action_mse_to_best": 0.33181332383898243,
604
+ "expert_success_rate": 0.8584070796460177,
605
+ "num_groups": 113,
606
+ "oracle_success_rate": 0.9380530973451328,
607
+ "policy_expert_regret": 1.004089695426215,
608
+ "policy_oracle_regret": 1.0832918505225562,
609
+ "policy_rollout_progress": 0.6276694327856587,
610
+ "policy_rollout_success_rate": 0.22123893805309736,
611
+ "restore_max_error": 4.76837158203125e-07
612
+ },
613
+ "PickCube-v1": {
614
+ "action_mse_to_best": 0.33753558987265697,
615
+ "expert_success_rate": 0.9402173913043478,
616
+ "num_groups": 184,
617
+ "oracle_success_rate": 0.9456521739130435,
618
+ "policy_expert_regret": 1.1125094104487367,
619
+ "policy_oracle_regret": 1.112533414923667,
620
+ "policy_rollout_progress": 0.5523134896521578,
621
+ "policy_rollout_success_rate": 0.266304347826087,
622
+ "restore_max_error": 4.76837158203125e-07
623
+ },
624
+ "PullCube-v1": {
625
+ "action_mse_to_best": 0.6109341694728324,
626
+ "expert_success_rate": 0.25,
627
+ "num_groups": 76,
628
+ "oracle_success_rate": 0.40789473684210525,
629
+ "policy_expert_regret": 0.3739064343829165,
630
+ "policy_oracle_regret": 0.6021578795881124,
631
+ "policy_rollout_progress": 0.25026103445704573,
632
+ "policy_rollout_success_rate": 0.14473684210526316,
633
+ "restore_max_error": 4.76837158203125e-07
634
+ },
635
+ "PushCube-v1": {
636
+ "action_mse_to_best": 0.3871889783254078,
637
+ "expert_success_rate": 0.8198198198198198,
638
+ "num_groups": 111,
639
+ "oracle_success_rate": 1.0,
640
+ "policy_expert_regret": 0.4335025233459902,
641
+ "policy_oracle_regret": 0.4999616445453317,
642
+ "policy_rollout_progress": 0.7612996167159295,
643
+ "policy_rollout_success_rate": 0.7387387387387387,
644
+ "restore_max_error": 4.76837158203125e-07
645
+ },
646
+ "StackCube-v1": {
647
+ "action_mse_to_best": 0.49320598007026284,
648
+ "expert_success_rate": 0.6923076923076923,
649
+ "num_groups": 91,
650
+ "oracle_success_rate": 0.8571428571428571,
651
+ "policy_expert_regret": 1.1323802955858,
652
+ "policy_oracle_regret": 1.3045280497152727,
653
+ "policy_rollout_progress": 0.38607727744422116,
654
+ "policy_rollout_success_rate": 0.15384615384615385,
655
+ "restore_max_error": 3.948807716369629e-07
656
+ }
657
+ }
658
+ },
659
+ {
660
+ "config": "k32_sigma0.35",
661
+ "seed": 2,
662
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k32_sigma0.35/seed_2/online_rollout.json",
663
+ "selection_mode": "field",
664
+ "num_candidates": 32,
665
+ "candidate_sigma": 0.35,
666
+ "num_groups": 575,
667
+ "policy_rollout_success_rate": 0.34608695652173915,
668
+ "policy_rollout_progress": 0.5607519139111908,
669
+ "oracle_success_rate": 0.8765217391304347,
670
+ "action_mse_to_best": 0.4337610558687669,
671
+ "per_task": {
672
+ "LiftPegUpright-v1": {
673
+ "action_mse_to_best": 0.3546609722592014,
674
+ "expert_success_rate": 0.8229166666666666,
675
+ "num_groups": 96,
676
+ "oracle_success_rate": 0.9270833333333334,
677
+ "policy_expert_regret": 0.8603796567767859,
678
+ "policy_oracle_regret": 0.9666543720910946,
679
+ "policy_rollout_progress": 0.644037484501799,
680
+ "policy_rollout_success_rate": 0.3229166666666667,
681
+ "restore_max_error": 3.5762786865234375e-07
682
+ },
683
+ "PickCube-v1": {
684
+ "action_mse_to_best": 0.34091578990296284,
685
+ "expert_success_rate": 0.9444444444444444,
686
+ "num_groups": 198,
687
+ "oracle_success_rate": 0.9595959595959596,
688
+ "policy_expert_regret": 1.0118255193599246,
689
+ "policy_oracle_regret": 1.0189089698886329,
690
+ "policy_rollout_progress": 0.6198147095250662,
691
+ "policy_rollout_success_rate": 0.32323232323232326,
692
+ "restore_max_error": 4.76837158203125e-07
693
+ },
694
+ "PullCube-v1": {
695
+ "action_mse_to_best": 0.7086632399095429,
696
+ "expert_success_rate": 0.24444444444444444,
697
+ "num_groups": 90,
698
+ "oracle_success_rate": 0.4666666666666667,
699
+ "policy_expert_regret": 0.3871518444545826,
700
+ "policy_oracle_regret": 0.6527724535790842,
701
+ "policy_rollout_progress": 0.26429835981517097,
702
+ "policy_rollout_success_rate": 0.15555555555555556,
703
+ "restore_max_error": 4.0978193283081055e-07
704
+ },
705
+ "PushCube-v1": {
706
+ "action_mse_to_best": 0.39258715377585723,
707
+ "expert_success_rate": 0.8514851485148515,
708
+ "num_groups": 101,
709
+ "oracle_success_rate": 1.0,
710
+ "policy_expert_regret": 0.40123750564485494,
711
+ "policy_oracle_regret": 0.49343014014239356,
712
+ "policy_rollout_progress": 0.7639956024318638,
713
+ "policy_rollout_success_rate": 0.7425742574257426,
714
+ "restore_max_error": 4.76837158203125e-07
715
+ },
716
+ "StackCube-v1": {
717
+ "action_mse_to_best": 0.49369814737389484,
718
+ "expert_success_rate": 0.7666666666666667,
719
+ "num_groups": 90,
720
+ "oracle_success_rate": 0.9111111111111111,
721
+ "policy_expert_regret": 1.155931308037705,
722
+ "policy_oracle_regret": 1.305003473162651,
723
+ "policy_rollout_progress": 0.41034479190905887,
724
+ "policy_rollout_success_rate": 0.16666666666666666,
725
+ "restore_max_error": 4.76837158203125e-07
726
+ }
727
+ }
728
+ },
729
+ {
730
+ "config": "k64_sigma0.35",
731
+ "seed": 0,
732
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k64_sigma0.35/seed_0/online_rollout.json",
733
+ "selection_mode": "field",
734
+ "num_candidates": 64,
735
+ "candidate_sigma": 0.35,
736
+ "num_groups": 575,
737
+ "policy_rollout_success_rate": 0.3234782608695652,
738
+ "policy_rollout_progress": 0.5343159465103726,
739
+ "oracle_success_rate": 0.8573913043478261,
740
+ "action_mse_to_best": 0.39918596936227835,
741
+ "per_task": {
742
+ "LiftPegUpright-v1": {
743
+ "action_mse_to_best": 0.32200342137846594,
744
+ "expert_success_rate": 0.8865979381443299,
745
+ "num_groups": 97,
746
+ "oracle_success_rate": 0.9278350515463918,
747
+ "policy_expert_regret": 1.0924534333735396,
748
+ "policy_oracle_regret": 1.1188689110819827,
749
+ "policy_rollout_progress": 0.5676527385859146,
750
+ "policy_rollout_success_rate": 0.21649484536082475,
751
+ "restore_max_error": 4.76837158203125e-07
752
+ },
753
+ "PickCube-v1": {
754
+ "action_mse_to_best": 0.32040938953510845,
755
+ "expert_success_rate": 0.9375,
756
+ "num_groups": 208,
757
+ "oracle_success_rate": 0.9471153846153846,
758
+ "policy_expert_regret": 1.0003741416238392,
759
+ "policy_oracle_regret": 1.0137970090503445,
760
+ "policy_rollout_progress": 0.5913155791260158,
761
+ "policy_rollout_success_rate": 0.3269230769230769,
762
+ "restore_max_error": 4.76837158203125e-07
763
+ },
764
+ "PullCube-v1": {
765
+ "action_mse_to_best": 0.6566622284325686,
766
+ "expert_success_rate": 0.19480519480519481,
767
+ "num_groups": 77,
768
+ "oracle_success_rate": 0.36363636363636365,
769
+ "policy_expert_regret": 0.2460009390846041,
770
+ "policy_oracle_regret": 0.4629497372528128,
771
+ "policy_rollout_progress": 0.29306630275033213,
772
+ "policy_rollout_success_rate": 0.18181818181818182,
773
+ "restore_max_error": 4.76837158203125e-07
774
+ },
775
+ "PushCube-v1": {
776
+ "action_mse_to_best": 0.39111159397091005,
777
+ "expert_success_rate": 0.8617021276595744,
778
+ "num_groups": 94,
779
+ "oracle_success_rate": 0.9893617021276596,
780
+ "policy_expert_regret": 0.5936297910644653,
781
+ "policy_oracle_regret": 0.6903621123192158,
782
+ "policy_rollout_progress": 0.6713400153403587,
783
+ "policy_rollout_success_rate": 0.6382978723404256,
784
+ "restore_max_error": 4.76837158203125e-07
785
+ },
786
+ "StackCube-v1": {
787
+ "action_mse_to_best": 0.44772692993659563,
788
+ "expert_success_rate": 0.6767676767676768,
789
+ "num_groups": 99,
790
+ "oracle_success_rate": 0.8585858585858586,
791
+ "policy_expert_regret": 0.9749961484863301,
792
+ "policy_oracle_regret": 1.1491557430438322,
793
+ "policy_rollout_progress": 0.43943077160252464,
794
+ "policy_rollout_success_rate": 0.23232323232323232,
795
+ "restore_max_error": 3.948807716369629e-07
796
+ }
797
+ }
798
+ },
799
+ {
800
+ "config": "k64_sigma0.35",
801
+ "seed": 1,
802
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k64_sigma0.35/seed_1/online_rollout.json",
803
+ "selection_mode": "field",
804
+ "num_candidates": 64,
805
+ "candidate_sigma": 0.35,
806
+ "num_groups": 575,
807
+ "policy_rollout_success_rate": 0.3095652173913043,
808
+ "policy_rollout_progress": 0.536779355444554,
809
+ "oracle_success_rate": 0.8695652173913043,
810
+ "action_mse_to_best": 0.4101900245184484,
811
+ "per_task": {
812
+ "LiftPegUpright-v1": {
813
+ "action_mse_to_best": 0.33171445011795886,
814
+ "expert_success_rate": 0.8584070796460177,
815
+ "num_groups": 113,
816
+ "oracle_success_rate": 0.9380530973451328,
817
+ "policy_expert_regret": 1.0022590173556742,
818
+ "policy_oracle_regret": 1.0813931261016205,
819
+ "policy_rollout_progress": 0.6295681572065944,
820
+ "policy_rollout_success_rate": 0.22123893805309736,
821
+ "restore_max_error": 4.76837158203125e-07
822
+ },
823
+ "PickCube-v1": {
824
+ "action_mse_to_best": 0.3401717414229136,
825
+ "expert_success_rate": 0.9402173913043478,
826
+ "num_groups": 184,
827
+ "oracle_success_rate": 0.9456521739130435,
828
+ "policy_expert_regret": 1.1126799705215609,
829
+ "policy_oracle_regret": 1.1127039749964909,
830
+ "policy_rollout_progress": 0.5521429295793338,
831
+ "policy_rollout_success_rate": 0.266304347826087,
832
+ "restore_max_error": 4.76837158203125e-07
833
+ },
834
+ "PullCube-v1": {
835
+ "action_mse_to_best": 0.6175777537650183,
836
+ "expert_success_rate": 0.25,
837
+ "num_groups": 76,
838
+ "oracle_success_rate": 0.40789473684210525,
839
+ "policy_expert_regret": 0.3696385342464373,
840
+ "policy_oracle_regret": 0.6190723324227227,
841
+ "policy_rollout_progress": 0.24086668692006366,
842
+ "policy_rollout_success_rate": 0.11842105263157894,
843
+ "restore_max_error": 4.76837158203125e-07
844
+ },
845
+ "PushCube-v1": {
846
+ "action_mse_to_best": 0.3934205474483,
847
+ "expert_success_rate": 0.8198198198198198,
848
+ "num_groups": 111,
849
+ "oracle_success_rate": 1.0,
850
+ "policy_expert_regret": 0.4525444950606372,
851
+ "policy_oracle_regret": 0.5380019815386953,
852
+ "policy_rollout_progress": 0.741277297740584,
853
+ "policy_rollout_success_rate": 0.7207207207207207,
854
+ "restore_max_error": 4.76837158203125e-07
855
+ },
856
+ "StackCube-v1": {
857
+ "action_mse_to_best": 0.49646528307757865,
858
+ "expert_success_rate": 0.6923076923076923,
859
+ "num_groups": 91,
860
+ "oracle_success_rate": 0.8571428571428571,
861
+ "policy_expert_regret": 1.131820830506283,
862
+ "policy_oracle_regret": 1.2914297192633808,
863
+ "policy_rollout_progress": 0.38818659690710217,
864
+ "policy_rollout_success_rate": 0.16483516483516483,
865
+ "restore_max_error": 3.948807716369629e-07
866
+ }
867
+ }
868
+ },
869
+ {
870
+ "config": "k64_sigma0.35",
871
+ "seed": 2,
872
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k64_sigma0.35/seed_2/online_rollout.json",
873
+ "selection_mode": "field",
874
+ "num_candidates": 64,
875
+ "candidate_sigma": 0.35,
876
+ "num_groups": 575,
877
+ "policy_rollout_success_rate": 0.34956521739130436,
878
+ "policy_rollout_progress": 0.5662590010642358,
879
+ "oracle_success_rate": 0.8765217391304347,
880
+ "action_mse_to_best": 0.43539300336542985,
881
+ "per_task": {
882
+ "LiftPegUpright-v1": {
883
+ "action_mse_to_best": 0.3568755660720247,
884
+ "expert_success_rate": 0.8229166666666666,
885
+ "num_groups": 96,
886
+ "oracle_success_rate": 0.9270833333333334,
887
+ "policy_expert_regret": 0.8741882933924595,
888
+ "policy_oracle_regret": 0.9804630087067684,
889
+ "policy_rollout_progress": 0.640645514552792,
890
+ "policy_rollout_success_rate": 0.3125,
891
+ "restore_max_error": 3.5762786865234375e-07
892
+ },
893
+ "PickCube-v1": {
894
+ "action_mse_to_best": 0.3404735067313669,
895
+ "expert_success_rate": 0.9444444444444444,
896
+ "num_groups": 198,
897
+ "oracle_success_rate": 0.9595959595959596,
898
+ "policy_expert_regret": 1.0108400425600885,
899
+ "policy_oracle_regret": 1.0179234930887968,
900
+ "policy_rollout_progress": 0.6208001863249023,
901
+ "policy_rollout_success_rate": 0.32323232323232326,
902
+ "restore_max_error": 4.76837158203125e-07
903
+ },
904
+ "PullCube-v1": {
905
+ "action_mse_to_best": 0.7115254996551408,
906
+ "expert_success_rate": 0.24444444444444444,
907
+ "num_groups": 90,
908
+ "oracle_success_rate": 0.4666666666666667,
909
+ "policy_expert_regret": 0.38466353309858176,
910
+ "policy_oracle_regret": 0.6054818535947966,
911
+ "policy_rollout_progress": 0.30913451097698674,
912
+ "policy_rollout_success_rate": 0.2,
913
+ "restore_max_error": 4.0978193283081055e-07
914
+ },
915
+ "PushCube-v1": {
916
+ "action_mse_to_best": 0.396122483481275,
917
+ "expert_success_rate": 0.8514851485148515,
918
+ "num_groups": 101,
919
+ "oracle_success_rate": 1.0,
920
+ "policy_expert_regret": 0.43801331372544317,
921
+ "policy_oracle_regret": 0.5106629358954949,
922
+ "policy_rollout_progress": 0.7566637967777724,
923
+ "policy_rollout_success_rate": 0.7326732673267327,
924
+ "restore_max_error": 4.76837158203125e-07
925
+ },
926
+ "StackCube-v1": {
927
+ "action_mse_to_best": 0.4959055828758412,
928
+ "expert_success_rate": 0.7666666666666667,
929
+ "num_groups": 90,
930
+ "oracle_success_rate": 0.9111111111111111,
931
+ "policy_expert_regret": 1.155905323723952,
932
+ "policy_oracle_regret": 1.304977488848898,
933
+ "policy_rollout_progress": 0.4103707762228118,
934
+ "policy_rollout_success_rate": 0.16666666666666666,
935
+ "restore_max_error": 4.76837158203125e-07
936
+ }
937
+ }
938
+ },
939
+ {
940
+ "config": "k64_sigma0.50",
941
+ "seed": 0,
942
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k64_sigma0.50/seed_0/online_rollout.json",
943
+ "selection_mode": "field",
944
+ "num_candidates": 64,
945
+ "candidate_sigma": 0.5,
946
+ "num_groups": 575,
947
+ "policy_rollout_success_rate": 0.3269565217391304,
948
+ "policy_rollout_progress": 0.5399010283508054,
949
+ "oracle_success_rate": 0.8573913043478261,
950
+ "action_mse_to_best": 0.4025516371615231,
951
+ "per_task": {
952
+ "LiftPegUpright-v1": {
953
+ "action_mse_to_best": 0.32251464834283156,
954
+ "expert_success_rate": 0.8865979381443299,
955
+ "num_groups": 97,
956
+ "oracle_success_rate": 0.9278350515463918,
957
+ "policy_expert_regret": 1.0791288545143973,
958
+ "policy_oracle_regret": 1.1055443322228402,
959
+ "policy_rollout_progress": 0.5707339151311166,
960
+ "policy_rollout_success_rate": 0.2268041237113402,
961
+ "restore_max_error": 4.76837158203125e-07
962
+ },
963
+ "PickCube-v1": {
964
+ "action_mse_to_best": 0.3262789388760351,
965
+ "expert_success_rate": 0.9375,
966
+ "num_groups": 208,
967
+ "oracle_success_rate": 0.9471153846153846,
968
+ "policy_expert_regret": 0.9832543106995917,
969
+ "policy_oracle_regret": 0.9966771781260971,
970
+ "policy_rollout_progress": 0.6036277177425711,
971
+ "policy_rollout_success_rate": 0.3317307692307692,
972
+ "restore_max_error": 4.76837158203125e-07
973
+ },
974
+ "PullCube-v1": {
975
+ "action_mse_to_best": 0.6741709269873508,
976
+ "expert_success_rate": 0.19480519480519481,
977
+ "num_groups": 77,
978
+ "oracle_success_rate": 0.36363636363636365,
979
+ "policy_expert_regret": 0.21915740969699699,
980
+ "policy_oracle_regret": 0.43092472498099527,
981
+ "policy_rollout_progress": 0.32208714377763986,
982
+ "policy_rollout_success_rate": 0.2077922077922078,
983
+ "restore_max_error": 4.76837158203125e-07
984
+ },
985
+ "PushCube-v1": {
986
+ "action_mse_to_best": 0.3830847614543869,
987
+ "expert_success_rate": 0.8617021276595744,
988
+ "num_groups": 94,
989
+ "oracle_success_rate": 0.9893617021276596,
990
+ "policy_expert_regret": 0.6339274873124793,
991
+ "policy_oracle_regret": 0.7308171584885171,
992
+ "policy_rollout_progress": 0.6521615649157382,
993
+ "policy_rollout_success_rate": 0.6170212765957447,
994
+ "restore_max_error": 4.76837158203125e-07
995
+ },
996
+ "StackCube-v1": {
997
+ "action_mse_to_best": 0.448445679168358,
998
+ "expert_success_rate": 0.6767676767676768,
999
+ "num_groups": 99,
1000
+ "oracle_success_rate": 0.8585858585858586,
1001
+ "policy_expert_regret": 0.9758063234163054,
1002
+ "policy_oracle_regret": 1.1499659179738073,
1003
+ "policy_rollout_progress": 0.4386205966725494,
1004
+ "policy_rollout_success_rate": 0.23232323232323232,
1005
+ "restore_max_error": 3.948807716369629e-07
1006
+ }
1007
+ }
1008
+ },
1009
+ {
1010
+ "config": "k64_sigma0.50",
1011
+ "seed": 1,
1012
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k64_sigma0.50/seed_1/online_rollout.json",
1013
+ "selection_mode": "field",
1014
+ "num_candidates": 64,
1015
+ "candidate_sigma": 0.5,
1016
+ "num_groups": 575,
1017
+ "policy_rollout_success_rate": 0.3182608695652174,
1018
+ "policy_rollout_progress": 0.5455222901711647,
1019
+ "oracle_success_rate": 0.8695652173913043,
1020
+ "action_mse_to_best": 0.41242632866877577,
1021
+ "per_task": {
1022
+ "LiftPegUpright-v1": {
1023
+ "action_mse_to_best": 0.3303305860361562,
1024
+ "expert_success_rate": 0.8584070796460177,
1025
+ "num_groups": 113,
1026
+ "oracle_success_rate": 0.9380530973451328,
1027
+ "policy_expert_regret": 0.9963025565432236,
1028
+ "policy_oracle_regret": 1.0867589768846477,
1029
+ "policy_rollout_progress": 0.624202306423567,
1030
+ "policy_rollout_success_rate": 0.22123893805309736,
1031
+ "restore_max_error": 4.76837158203125e-07
1032
+ },
1033
+ "PickCube-v1": {
1034
+ "action_mse_to_best": 0.33823254261084873,
1035
+ "expert_success_rate": 0.9402173913043478,
1036
+ "num_groups": 184,
1037
+ "oracle_success_rate": 0.9456521739130435,
1038
+ "policy_expert_regret": 1.0973767103475474,
1039
+ "policy_oracle_regret": 1.0974007148224774,
1040
+ "policy_rollout_progress": 0.5641294867033144,
1041
+ "policy_rollout_success_rate": 0.27717391304347827,
1042
+ "restore_max_error": 4.76837158203125e-07
1043
+ },
1044
+ "PullCube-v1": {
1045
+ "action_mse_to_best": 0.6391991545495234,
1046
+ "expert_success_rate": 0.25,
1047
+ "num_groups": 76,
1048
+ "oracle_success_rate": 0.40789473684210525,
1049
+ "policy_expert_regret": 0.3485522249125336,
1050
+ "policy_oracle_regret": 0.5971745297214702,
1051
+ "policy_rollout_progress": 0.2562023993875635,
1052
+ "policy_rollout_success_rate": 0.14473684210526316,
1053
+ "restore_max_error": 4.76837158203125e-07
1054
+ },
1055
+ "PushCube-v1": {
1056
+ "action_mse_to_best": 0.3912716278688865,
1057
+ "expert_success_rate": 0.8198198198198198,
1058
+ "num_groups": 111,
1059
+ "oracle_success_rate": 1.0,
1060
+ "policy_expert_regret": 0.4345502409043613,
1061
+ "policy_oracle_regret": 0.5010873861946501,
1062
+ "policy_rollout_progress": 0.7601738750666112,
1063
+ "policy_rollout_success_rate": 0.7387387387387387,
1064
+ "restore_max_error": 4.76837158203125e-07
1065
+ },
1066
+ "StackCube-v1": {
1067
+ "action_mse_to_best": 0.5007989943170776,
1068
+ "expert_success_rate": 0.6923076923076923,
1069
+ "num_groups": 91,
1070
+ "oracle_success_rate": 0.8571428571428571,
1071
+ "policy_expert_regret": 1.1284581424443276,
1072
+ "policy_oracle_regret": 1.3006058965738003,
1073
+ "policy_rollout_progress": 0.38999943058569353,
1074
+ "policy_rollout_success_rate": 0.15384615384615385,
1075
+ "restore_max_error": 3.948807716369629e-07
1076
+ }
1077
+ }
1078
+ },
1079
+ {
1080
+ "config": "k64_sigma0.50",
1081
+ "seed": 2,
1082
+ "path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt/k64_sigma0.50/seed_2/online_rollout.json",
1083
+ "selection_mode": "field",
1084
+ "num_candidates": 64,
1085
+ "candidate_sigma": 0.5,
1086
+ "num_groups": 575,
1087
+ "policy_rollout_success_rate": 0.3426086956521739,
1088
+ "policy_rollout_progress": 0.5595585580056777,
1089
+ "oracle_success_rate": 0.8765217391304347,
1090
+ "action_mse_to_best": 0.44333537457834765,
1091
+ "per_task": {
1092
+ "LiftPegUpright-v1": {
1093
+ "action_mse_to_best": 0.35361564260771655,
1094
+ "expert_success_rate": 0.8229166666666666,
1095
+ "num_groups": 96,
1096
+ "oracle_success_rate": 0.9270833333333334,
1097
+ "policy_expert_regret": 0.8470402403424183,
1098
+ "policy_oracle_regret": 0.9533149556567272,
1099
+ "policy_rollout_progress": 0.6469602342694998,
1100
+ "policy_rollout_success_rate": 0.3333333333333333,
1101
+ "restore_max_error": 3.5762786865234375e-07
1102
+ },
1103
+ "PickCube-v1": {
1104
+ "action_mse_to_best": 0.34493389291067916,
1105
+ "expert_success_rate": 0.9444444444444444,
1106
+ "num_groups": 198,
1107
+ "oracle_success_rate": 0.9595959595959596,
1108
+ "policy_expert_regret": 1.007205421227322,
1109
+ "policy_oracle_regret": 1.0142888717560306,
1110
+ "policy_rollout_progress": 0.6244348076576687,
1111
+ "policy_rollout_success_rate": 0.32323232323232326,
1112
+ "restore_max_error": 4.76837158203125e-07
1113
+ },
1114
+ "PullCube-v1": {
1115
+ "action_mse_to_best": 0.75293321294917,
1116
+ "expert_success_rate": 0.24444444444444444,
1117
+ "num_groups": 90,
1118
+ "oracle_success_rate": 0.4666666666666667,
1119
+ "policy_expert_regret": 0.4486654932610691,
1120
+ "policy_oracle_regret": 0.7107819825928244,
1121
+ "policy_rollout_progress": 0.250149907451123,
1122
+ "policy_rollout_success_rate": 0.13333333333333333,
1123
+ "restore_max_error": 4.0978193283081055e-07
1124
+ },
1125
+ "PushCube-v1": {
1126
+ "action_mse_to_best": 0.3872078667477806,
1127
+ "expert_success_rate": 0.8514851485148515,
1128
+ "num_groups": 101,
1129
+ "oracle_success_rate": 1.0,
1130
+ "policy_expert_regret": 0.4194727140192938,
1131
+ "policy_oracle_regret": 0.5103210249159595,
1132
+ "policy_rollout_progress": 0.7570057077573078,
1133
+ "policy_rollout_success_rate": 0.7326732673267327,
1134
+ "restore_max_error": 4.76837158203125e-07
1135
+ },
1136
+ "StackCube-v1": {
1137
+ "action_mse_to_best": 0.50890937987715,
1138
+ "expert_success_rate": 0.7666666666666667,
1139
+ "num_groups": 90,
1140
+ "oracle_success_rate": 0.9111111111111111,
1141
+ "policy_expert_regret": 1.1548446744680405,
1142
+ "policy_oracle_regret": 1.3039168395929865,
1143
+ "policy_rollout_progress": 0.4114314254787233,
1144
+ "policy_rollout_success_rate": 0.16666666666666666,
1145
+ "restore_max_error": 4.76837158203125e-07
1146
+ }
1147
+ }
1148
+ }
1149
+ ]
1150
+ }
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_field_sweep_summary.md ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # h=16 Field-Guided Rollout Sweep
2
+
3
+ Result root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_field_sweep/near_miss_policy_bc5_bestpt`
4
+ Completed result files: 15
5
+ Baseline h=4 policy success: 29.67%
6
+ Baseline h=16 policy success: 29.74%
7
+
8
+ | config | seeds | mean success | gain vs h=16 | progress | action MSE |
9
+ |---|---:|---:|---:|---:|---:|
10
+ | k64_sigma0.50 | 3 | 32.93% | +3.19% | 54.83% | 0.419 |
11
+ | k32_sigma0.35 | 3 | 32.81% | +3.07% | 54.62% | 0.412 |
12
+ | k64_sigma0.35 | 3 | 32.75% | +3.01% | 54.58% | 0.415 |
13
+ | k16_sigma0.20 | 3 | 32.64% | +2.90% | 54.71% | 0.405 |
14
+ | k32_sigma0.20 | 3 | 32.23% | +2.49% | 54.14% | 0.406 |
15
+
16
+ Best config:
17
+ - k64_sigma0.50
18
+ - mean success: 32.93%
19
+ - gain vs h=16 policy: +3.19%