Auto-sync: 2026-06-30 15:20:36
Browse files- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_stacknowg_summary.json +475 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_stacknowg_summary.md +19 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_challenger0p01_norm_summary.json +478 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_challenger0p01_norm_summary.md +19 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_norm_summary.json +472 -0
- results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_norm_summary.md +19 -0
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_stacknowg_summary.json
ADDED
|
@@ -0,0 +1,475 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
|
| 3 |
+
"objective": "near_miss_policy_bc5",
|
| 4 |
+
"out_name": "policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_stacknowg.json",
|
| 5 |
+
"num_completed": 3,
|
| 6 |
+
"baseline_h4_policy_success": 0.2967,
|
| 7 |
+
"baseline_h16_rank_checkpoint_success": 0.29739130434782607,
|
| 8 |
+
"mean_success": 0.36,
|
| 9 |
+
"std_success": 0.017650246200160393,
|
| 10 |
+
"mean_progress": 0.571213552200006,
|
| 11 |
+
"mean_action_mse_to_best": 0.4006344801130826,
|
| 12 |
+
"gain_vs_h4": 0.06329999999999997,
|
| 13 |
+
"gain_vs_h16_rank_checkpoint": 0.06260869565217392,
|
| 14 |
+
"rows": [
|
| 15 |
+
{
|
| 16 |
+
"seed": 0,
|
| 17 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_0/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_stacknowg.json",
|
| 18 |
+
"num_groups": 575,
|
| 19 |
+
"selection_mode": "retrieval_residual",
|
| 20 |
+
"num_candidates": 48,
|
| 21 |
+
"candidate_sigma": 0.0,
|
| 22 |
+
"proposal_types": [],
|
| 23 |
+
"proposal_lattice_types": [],
|
| 24 |
+
"selection_margin": 0.2,
|
| 25 |
+
"prepend_policy_candidate": false,
|
| 26 |
+
"field_optim_steps": 0,
|
| 27 |
+
"field_optim_step_size": 0.0,
|
| 28 |
+
"field_optim_trust_radius": 0.0,
|
| 29 |
+
"field_optim_l2_penalty": 0.0,
|
| 30 |
+
"retrieval_neighbors": 4,
|
| 31 |
+
"retrieval_metric": "raw",
|
| 32 |
+
"retrieval_type_min_success": 0.0,
|
| 33 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 34 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 35 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 36 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 37 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 38 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 39 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 40 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 41 |
+
"retrieval_residual_scale": 1.0,
|
| 42 |
+
"retrieval_residual_scales": [
|
| 43 |
+
0.35,
|
| 44 |
+
0.4,
|
| 45 |
+
0.45
|
| 46 |
+
],
|
| 47 |
+
"retrieval_residual_anchor": "expert",
|
| 48 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 49 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 50 |
+
"retrieval_residual_challenger_types": [
|
| 51 |
+
"residual_near_miss"
|
| 52 |
+
],
|
| 53 |
+
"retrieval_residual_challenger_scales": [],
|
| 54 |
+
"retrieval_residual_challenger_margin": 0.01,
|
| 55 |
+
"retrieval_residual_challenger_tasks": [],
|
| 56 |
+
"retrieval_residual_challenger_type_tasks": {},
|
| 57 |
+
"lattice_exclude_types": [
|
| 58 |
+
"residual_random_negative",
|
| 59 |
+
"residual_wrong_direction",
|
| 60 |
+
"residual_near_miss+residual_no_op"
|
| 61 |
+
],
|
| 62 |
+
"lattice_exclude_type_tasks": {
|
| 63 |
+
"residual_wrong_gripper": [
|
| 64 |
+
"StackCube-v1"
|
| 65 |
+
]
|
| 66 |
+
},
|
| 67 |
+
"candidate_type_bonuses": {
|
| 68 |
+
"retrieval_residual_no_op": 0.03
|
| 69 |
+
},
|
| 70 |
+
"candidate_type_bonus_components": false,
|
| 71 |
+
"candidate_oracle_rollouts": 0,
|
| 72 |
+
"candidate_oracle_unique_tolerance": 1e-06,
|
| 73 |
+
"candidate_oracle_success_rate": null,
|
| 74 |
+
"candidate_oracle_progress": null,
|
| 75 |
+
"candidate_oracle_score": null,
|
| 76 |
+
"candidate_oracle_regret": null,
|
| 77 |
+
"candidate_oracle_score_gain_over_selected": null,
|
| 78 |
+
"candidate_oracle_unique_count": null,
|
| 79 |
+
"candidate_oracle_improvement_rate": null,
|
| 80 |
+
"candidate_oracle_selected_branch_success_rate": null,
|
| 81 |
+
"candidate_oracle_selected_branch_progress": null,
|
| 82 |
+
"candidate_oracle_type_counts": {},
|
| 83 |
+
"candidate_oracle_best_branch_rank": null,
|
| 84 |
+
"candidate_oracle_best_branch_rank_counts": {},
|
| 85 |
+
"candidate_oracle_branch_success_rates": [],
|
| 86 |
+
"candidate_oracle_branch_progress": [],
|
| 87 |
+
"candidate_oracle_branch_score_gains_over_selected": [],
|
| 88 |
+
"selected_residual_scale_counts": {
|
| 89 |
+
"0.35": 525,
|
| 90 |
+
"0.45": 45,
|
| 91 |
+
"0.4": 5
|
| 92 |
+
},
|
| 93 |
+
"policy_rollout_success_rate": 0.3565217391304348,
|
| 94 |
+
"policy_rollout_progress": 0.5547143837315557,
|
| 95 |
+
"oracle_success_rate": 0.8573913043478261,
|
| 96 |
+
"action_mse_to_best": 0.3848872706310257,
|
| 97 |
+
"best_policy_val": {
|
| 98 |
+
"bc_loss": 0.13721593966086706,
|
| 99 |
+
"field_effect_loss": 0.009290305380192068,
|
| 100 |
+
"field_potential_loss": 0.2666468388504452,
|
| 101 |
+
"field_preference_loss": 0.5130573478009965,
|
| 102 |
+
"lattice_edges": 3833.3333333333335,
|
| 103 |
+
"progress_mae": 0.1933159919248687,
|
| 104 |
+
"rank_acc": 0.8265031774838766,
|
| 105 |
+
"rank_loss": 0.5130523675017886,
|
| 106 |
+
"regret_mae": 0.3756548762321472,
|
| 107 |
+
"success_accuracy": 0.8773836526605818,
|
| 108 |
+
"total_loss": 1.5581054819954767
|
| 109 |
+
},
|
| 110 |
+
"per_task": {
|
| 111 |
+
"LiftPegUpright-v1": {
|
| 112 |
+
"action_mse_to_best": 0.32276735697709713,
|
| 113 |
+
"expert_success_rate": 0.8865979381443299,
|
| 114 |
+
"num_groups": 97,
|
| 115 |
+
"oracle_success_rate": 0.9278350515463918,
|
| 116 |
+
"policy_expert_regret": 1.0783334703175063,
|
| 117 |
+
"policy_oracle_regret": 1.1047489480259491,
|
| 118 |
+
"policy_rollout_progress": 0.5714634232914325,
|
| 119 |
+
"policy_rollout_success_rate": 0.2268041237113402,
|
| 120 |
+
"restore_max_error": 2.384185791015625e-07
|
| 121 |
+
},
|
| 122 |
+
"PickCube-v1": {
|
| 123 |
+
"action_mse_to_best": 0.3027741009584413,
|
| 124 |
+
"expert_success_rate": 0.9375,
|
| 125 |
+
"num_groups": 208,
|
| 126 |
+
"oracle_success_rate": 0.9471153846153846,
|
| 127 |
+
"policy_expert_regret": 0.9442090743866104,
|
| 128 |
+
"policy_oracle_regret": 0.9576319418131158,
|
| 129 |
+
"policy_rollout_progress": 0.6186344925170908,
|
| 130 |
+
"policy_rollout_success_rate": 0.3557692307692308,
|
| 131 |
+
"restore_max_error": 2.384185791015625e-07
|
| 132 |
+
},
|
| 133 |
+
"PullCube-v1": {
|
| 134 |
+
"action_mse_to_best": 0.5978217993850832,
|
| 135 |
+
"expert_success_rate": 0.19480519480519481,
|
| 136 |
+
"num_groups": 77,
|
| 137 |
+
"oracle_success_rate": 0.36363636363636365,
|
| 138 |
+
"policy_expert_regret": 0.2471449282322708,
|
| 139 |
+
"policy_oracle_regret": 0.4208110002740347,
|
| 140 |
+
"policy_rollout_progress": 0.31884596189000197,
|
| 141 |
+
"policy_rollout_success_rate": 0.22077922077922077,
|
| 142 |
+
"restore_max_error": 2.3096799850463867e-07
|
| 143 |
+
},
|
| 144 |
+
"PushCube-v1": {
|
| 145 |
+
"action_mse_to_best": 0.3675119048578942,
|
| 146 |
+
"expert_success_rate": 0.8617021276595744,
|
| 147 |
+
"num_groups": 94,
|
| 148 |
+
"oracle_success_rate": 0.9893617021276596,
|
| 149 |
+
"policy_expert_regret": 0.5508309629686335,
|
| 150 |
+
"policy_oracle_regret": 0.628381609441118,
|
| 151 |
+
"policy_rollout_progress": 0.7014056246014352,
|
| 152 |
+
"policy_rollout_success_rate": 0.6702127659574468,
|
| 153 |
+
"restore_max_error": 2.384185791015625e-07
|
| 154 |
+
},
|
| 155 |
+
"StackCube-v1": {
|
| 156 |
+
"action_mse_to_best": 0.4691549129031523,
|
| 157 |
+
"expert_success_rate": 0.6767676767676768,
|
| 158 |
+
"num_groups": 99,
|
| 159 |
+
"oracle_success_rate": 0.8585858585858586,
|
| 160 |
+
"policy_expert_regret": 0.9402529505467174,
|
| 161 |
+
"policy_oracle_regret": 1.091359183944837,
|
| 162 |
+
"policy_rollout_progress": 0.44817753903793567,
|
| 163 |
+
"policy_rollout_success_rate": 0.29292929292929293,
|
| 164 |
+
"restore_max_error": 2.384185791015625e-07
|
| 165 |
+
}
|
| 166 |
+
}
|
| 167 |
+
},
|
| 168 |
+
{
|
| 169 |
+
"seed": 1,
|
| 170 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_stacknowg.json",
|
| 171 |
+
"num_groups": 575,
|
| 172 |
+
"selection_mode": "retrieval_residual",
|
| 173 |
+
"num_candidates": 48,
|
| 174 |
+
"candidate_sigma": 0.0,
|
| 175 |
+
"proposal_types": [],
|
| 176 |
+
"proposal_lattice_types": [],
|
| 177 |
+
"selection_margin": 0.2,
|
| 178 |
+
"prepend_policy_candidate": false,
|
| 179 |
+
"field_optim_steps": 0,
|
| 180 |
+
"field_optim_step_size": 0.0,
|
| 181 |
+
"field_optim_trust_radius": 0.0,
|
| 182 |
+
"field_optim_l2_penalty": 0.0,
|
| 183 |
+
"retrieval_neighbors": 4,
|
| 184 |
+
"retrieval_metric": "raw",
|
| 185 |
+
"retrieval_type_min_success": 0.0,
|
| 186 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 187 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 188 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 189 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 190 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 191 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 192 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 193 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 194 |
+
"retrieval_residual_scale": 1.0,
|
| 195 |
+
"retrieval_residual_scales": [
|
| 196 |
+
0.35,
|
| 197 |
+
0.4,
|
| 198 |
+
0.45
|
| 199 |
+
],
|
| 200 |
+
"retrieval_residual_anchor": "expert",
|
| 201 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 202 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 203 |
+
"retrieval_residual_challenger_types": [
|
| 204 |
+
"residual_near_miss"
|
| 205 |
+
],
|
| 206 |
+
"retrieval_residual_challenger_scales": [],
|
| 207 |
+
"retrieval_residual_challenger_margin": 0.01,
|
| 208 |
+
"retrieval_residual_challenger_tasks": [],
|
| 209 |
+
"retrieval_residual_challenger_type_tasks": {},
|
| 210 |
+
"lattice_exclude_types": [
|
| 211 |
+
"residual_random_negative",
|
| 212 |
+
"residual_wrong_direction",
|
| 213 |
+
"residual_near_miss+residual_no_op"
|
| 214 |
+
],
|
| 215 |
+
"lattice_exclude_type_tasks": {
|
| 216 |
+
"residual_wrong_gripper": [
|
| 217 |
+
"StackCube-v1"
|
| 218 |
+
]
|
| 219 |
+
},
|
| 220 |
+
"candidate_type_bonuses": {
|
| 221 |
+
"retrieval_residual_no_op": 0.03
|
| 222 |
+
},
|
| 223 |
+
"candidate_type_bonus_components": false,
|
| 224 |
+
"candidate_oracle_rollouts": 0,
|
| 225 |
+
"candidate_oracle_unique_tolerance": 1e-06,
|
| 226 |
+
"candidate_oracle_success_rate": null,
|
| 227 |
+
"candidate_oracle_progress": null,
|
| 228 |
+
"candidate_oracle_score": null,
|
| 229 |
+
"candidate_oracle_regret": null,
|
| 230 |
+
"candidate_oracle_score_gain_over_selected": null,
|
| 231 |
+
"candidate_oracle_unique_count": null,
|
| 232 |
+
"candidate_oracle_improvement_rate": null,
|
| 233 |
+
"candidate_oracle_selected_branch_success_rate": null,
|
| 234 |
+
"candidate_oracle_selected_branch_progress": null,
|
| 235 |
+
"candidate_oracle_type_counts": {},
|
| 236 |
+
"candidate_oracle_best_branch_rank": null,
|
| 237 |
+
"candidate_oracle_best_branch_rank_counts": {},
|
| 238 |
+
"candidate_oracle_branch_success_rates": [],
|
| 239 |
+
"candidate_oracle_branch_progress": [],
|
| 240 |
+
"candidate_oracle_branch_score_gains_over_selected": [],
|
| 241 |
+
"selected_residual_scale_counts": {
|
| 242 |
+
"0.35": 534,
|
| 243 |
+
"0.45": 38,
|
| 244 |
+
"0.4": 3
|
| 245 |
+
},
|
| 246 |
+
"policy_rollout_success_rate": 0.3443478260869565,
|
| 247 |
+
"policy_rollout_progress": 0.5689966015874044,
|
| 248 |
+
"oracle_success_rate": 0.8695652173913043,
|
| 249 |
+
"action_mse_to_best": 0.39591813460478315,
|
| 250 |
+
"best_policy_val": {
|
| 251 |
+
"bc_loss": 0.1315989200439718,
|
| 252 |
+
"field_effect_loss": 0.010110370814800262,
|
| 253 |
+
"field_potential_loss": 0.31546536915832096,
|
| 254 |
+
"field_preference_loss": 0.4960637440284093,
|
| 255 |
+
"lattice_edges": 3833.3333333333335,
|
| 256 |
+
"progress_mae": 0.19023226698239645,
|
| 257 |
+
"rank_acc": 0.8255469501018524,
|
| 258 |
+
"rank_loss": 0.4960531195004781,
|
| 259 |
+
"regret_mae": 0.42156532737943864,
|
| 260 |
+
"success_accuracy": 0.8807473679383596,
|
| 261 |
+
"total_loss": 1.561984618504842
|
| 262 |
+
},
|
| 263 |
+
"per_task": {
|
| 264 |
+
"LiftPegUpright-v1": {
|
| 265 |
+
"action_mse_to_best": 0.3281630301287612,
|
| 266 |
+
"expert_success_rate": 0.8584070796460177,
|
| 267 |
+
"num_groups": 113,
|
| 268 |
+
"oracle_success_rate": 0.9380530973451328,
|
| 269 |
+
"policy_expert_regret": 0.987445328615408,
|
| 270 |
+
"policy_oracle_regret": 1.0666473587002374,
|
| 271 |
+
"policy_rollout_progress": 0.6354643670858535,
|
| 272 |
+
"policy_rollout_success_rate": 0.23008849557522124,
|
| 273 |
+
"restore_max_error": 2.384185791015625e-07
|
| 274 |
+
},
|
| 275 |
+
"PickCube-v1": {
|
| 276 |
+
"action_mse_to_best": 0.3249674804063271,
|
| 277 |
+
"expert_success_rate": 0.9402173913043478,
|
| 278 |
+
"num_groups": 184,
|
| 279 |
+
"oracle_success_rate": 0.9456521739130435,
|
| 280 |
+
"policy_expert_regret": 1.0883876778003152,
|
| 281 |
+
"policy_oracle_regret": 1.0884116822752454,
|
| 282 |
+
"policy_rollout_progress": 0.5676837366418508,
|
| 283 |
+
"policy_rollout_success_rate": 0.2826086956521739,
|
| 284 |
+
"restore_max_error": 2.384185791015625e-07
|
| 285 |
+
},
|
| 286 |
+
"PullCube-v1": {
|
| 287 |
+
"action_mse_to_best": 0.594155711464976,
|
| 288 |
+
"expert_success_rate": 0.25,
|
| 289 |
+
"num_groups": 76,
|
| 290 |
+
"oracle_success_rate": 0.40789473684210525,
|
| 291 |
+
"policy_expert_regret": 0.2993461788680993,
|
| 292 |
+
"policy_oracle_regret": 0.5232914631303988,
|
| 293 |
+
"policy_rollout_progress": 0.2899509874221526,
|
| 294 |
+
"policy_rollout_success_rate": 0.18421052631578946,
|
| 295 |
+
"restore_max_error": 2.384185791015625e-07
|
| 296 |
+
},
|
| 297 |
+
"PushCube-v1": {
|
| 298 |
+
"action_mse_to_best": 0.3595503710948669,
|
| 299 |
+
"expert_success_rate": 0.8198198198198198,
|
| 300 |
+
"num_groups": 111,
|
| 301 |
+
"oracle_success_rate": 1.0,
|
| 302 |
+
"policy_expert_regret": 0.3128753544779511,
|
| 303 |
+
"policy_oracle_regret": 0.33004292161078064,
|
| 304 |
+
"policy_rollout_progress": 0.8411282495603906,
|
| 305 |
+
"policy_rollout_success_rate": 0.8288288288288288,
|
| 306 |
+
"restore_max_error": 2.3795291781425476e-07
|
| 307 |
+
},
|
| 308 |
+
"StackCube-v1": {
|
| 309 |
+
"action_mse_to_best": 0.5023138828084364,
|
| 310 |
+
"expert_success_rate": 0.6923076923076923,
|
| 311 |
+
"num_groups": 91,
|
| 312 |
+
"oracle_success_rate": 0.8571428571428571,
|
| 313 |
+
"policy_expert_regret": 1.140720906984675,
|
| 314 |
+
"policy_oracle_regret": 1.300382754141158,
|
| 315 |
+
"policy_rollout_progress": 0.390222573018336,
|
| 316 |
+
"policy_rollout_success_rate": 0.15384615384615385,
|
| 317 |
+
"restore_max_error": 1.9371509552001953e-07
|
| 318 |
+
}
|
| 319 |
+
}
|
| 320 |
+
},
|
| 321 |
+
{
|
| 322 |
+
"seed": 2,
|
| 323 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_stacknowg.json",
|
| 324 |
+
"num_groups": 575,
|
| 325 |
+
"selection_mode": "retrieval_residual",
|
| 326 |
+
"num_candidates": 48,
|
| 327 |
+
"candidate_sigma": 0.0,
|
| 328 |
+
"proposal_types": [],
|
| 329 |
+
"proposal_lattice_types": [],
|
| 330 |
+
"selection_margin": 0.2,
|
| 331 |
+
"prepend_policy_candidate": false,
|
| 332 |
+
"field_optim_steps": 0,
|
| 333 |
+
"field_optim_step_size": 0.0,
|
| 334 |
+
"field_optim_trust_radius": 0.0,
|
| 335 |
+
"field_optim_l2_penalty": 0.0,
|
| 336 |
+
"retrieval_neighbors": 4,
|
| 337 |
+
"retrieval_metric": "raw",
|
| 338 |
+
"retrieval_type_min_success": 0.0,
|
| 339 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 340 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 341 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 342 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 343 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 344 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 345 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 346 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 347 |
+
"retrieval_residual_scale": 1.0,
|
| 348 |
+
"retrieval_residual_scales": [
|
| 349 |
+
0.35,
|
| 350 |
+
0.4,
|
| 351 |
+
0.45
|
| 352 |
+
],
|
| 353 |
+
"retrieval_residual_anchor": "expert",
|
| 354 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 355 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 356 |
+
"retrieval_residual_challenger_types": [
|
| 357 |
+
"residual_near_miss"
|
| 358 |
+
],
|
| 359 |
+
"retrieval_residual_challenger_scales": [],
|
| 360 |
+
"retrieval_residual_challenger_margin": 0.01,
|
| 361 |
+
"retrieval_residual_challenger_tasks": [],
|
| 362 |
+
"retrieval_residual_challenger_type_tasks": {},
|
| 363 |
+
"lattice_exclude_types": [
|
| 364 |
+
"residual_random_negative",
|
| 365 |
+
"residual_wrong_direction",
|
| 366 |
+
"residual_near_miss+residual_no_op"
|
| 367 |
+
],
|
| 368 |
+
"lattice_exclude_type_tasks": {
|
| 369 |
+
"residual_wrong_gripper": [
|
| 370 |
+
"StackCube-v1"
|
| 371 |
+
]
|
| 372 |
+
},
|
| 373 |
+
"candidate_type_bonuses": {
|
| 374 |
+
"retrieval_residual_no_op": 0.03
|
| 375 |
+
},
|
| 376 |
+
"candidate_type_bonus_components": false,
|
| 377 |
+
"candidate_oracle_rollouts": 0,
|
| 378 |
+
"candidate_oracle_unique_tolerance": 1e-06,
|
| 379 |
+
"candidate_oracle_success_rate": null,
|
| 380 |
+
"candidate_oracle_progress": null,
|
| 381 |
+
"candidate_oracle_score": null,
|
| 382 |
+
"candidate_oracle_regret": null,
|
| 383 |
+
"candidate_oracle_score_gain_over_selected": null,
|
| 384 |
+
"candidate_oracle_unique_count": null,
|
| 385 |
+
"candidate_oracle_improvement_rate": null,
|
| 386 |
+
"candidate_oracle_selected_branch_success_rate": null,
|
| 387 |
+
"candidate_oracle_selected_branch_progress": null,
|
| 388 |
+
"candidate_oracle_type_counts": {},
|
| 389 |
+
"candidate_oracle_best_branch_rank": null,
|
| 390 |
+
"candidate_oracle_best_branch_rank_counts": {},
|
| 391 |
+
"candidate_oracle_branch_success_rates": [],
|
| 392 |
+
"candidate_oracle_branch_progress": [],
|
| 393 |
+
"candidate_oracle_branch_score_gains_over_selected": [],
|
| 394 |
+
"selected_residual_scale_counts": {
|
| 395 |
+
"0.35": 526,
|
| 396 |
+
"0.45": 40,
|
| 397 |
+
"0.4": 9
|
| 398 |
+
},
|
| 399 |
+
"policy_rollout_success_rate": 0.3791304347826087,
|
| 400 |
+
"policy_rollout_progress": 0.5899296712810579,
|
| 401 |
+
"oracle_success_rate": 0.8765217391304347,
|
| 402 |
+
"action_mse_to_best": 0.42109803510343896,
|
| 403 |
+
"best_policy_val": {
|
| 404 |
+
"bc_loss": 0.11367896075050037,
|
| 405 |
+
"field_effect_loss": 0.009670218582161598,
|
| 406 |
+
"field_potential_loss": 0.2641640139950646,
|
| 407 |
+
"field_preference_loss": 0.5130490180518892,
|
| 408 |
+
"lattice_edges": 3833.3333333333335,
|
| 409 |
+
"progress_mae": 0.2021729110015763,
|
| 410 |
+
"rank_acc": 0.8333857821093665,
|
| 411 |
+
"rank_loss": 0.5130119257503085,
|
| 412 |
+
"regret_mae": 0.3958987047274907,
|
| 413 |
+
"success_accuracy": 0.8680730561415354,
|
| 414 |
+
"total_loss": 1.4394984311527677
|
| 415 |
+
},
|
| 416 |
+
"per_task": {
|
| 417 |
+
"LiftPegUpright-v1": {
|
| 418 |
+
"action_mse_to_best": 0.35602021238203935,
|
| 419 |
+
"expert_success_rate": 0.8229166666666666,
|
| 420 |
+
"num_groups": 96,
|
| 421 |
+
"oracle_success_rate": 0.9270833333333334,
|
| 422 |
+
"policy_expert_regret": 0.8476849501021206,
|
| 423 |
+
"policy_oracle_regret": 0.9539546399998168,
|
| 424 |
+
"policy_rollout_progress": 0.6463205499264101,
|
| 425 |
+
"policy_rollout_success_rate": 0.3333333333333333,
|
| 426 |
+
"restore_max_error": 1.955777406692505e-07
|
| 427 |
+
},
|
| 428 |
+
"PickCube-v1": {
|
| 429 |
+
"action_mse_to_best": 0.33825299484335414,
|
| 430 |
+
"expert_success_rate": 0.9444444444444444,
|
| 431 |
+
"num_groups": 198,
|
| 432 |
+
"oracle_success_rate": 0.9595959595959596,
|
| 433 |
+
"policy_expert_regret": 0.9734464705065645,
|
| 434 |
+
"policy_oracle_regret": 0.9805299210352729,
|
| 435 |
+
"policy_rollout_progress": 0.6379917381764061,
|
| 436 |
+
"policy_rollout_success_rate": 0.3434343434343434,
|
| 437 |
+
"restore_max_error": 2.384185791015625e-07
|
| 438 |
+
},
|
| 439 |
+
"PullCube-v1": {
|
| 440 |
+
"action_mse_to_best": 0.639584848864211,
|
| 441 |
+
"expert_success_rate": 0.24444444444444444,
|
| 442 |
+
"num_groups": 90,
|
| 443 |
+
"oracle_success_rate": 0.4666666666666667,
|
| 444 |
+
"policy_expert_regret": 0.26856607250455355,
|
| 445 |
+
"policy_oracle_regret": 0.454090747775303,
|
| 446 |
+
"policy_rollout_progress": 0.37102277845972115,
|
| 447 |
+
"policy_rollout_success_rate": 0.26666666666666666,
|
| 448 |
+
"restore_max_error": 2.384185791015625e-07
|
| 449 |
+
},
|
| 450 |
+
"PushCube-v1": {
|
| 451 |
+
"action_mse_to_best": 0.397049247997232,
|
| 452 |
+
"expert_success_rate": 0.8514851485148515,
|
| 453 |
+
"num_groups": 101,
|
| 454 |
+
"oracle_success_rate": 1.0,
|
| 455 |
+
"policy_expert_regret": 0.3614020819711213,
|
| 456 |
+
"policy_oracle_regret": 0.41630433337523204,
|
| 457 |
+
"policy_rollout_progress": 0.8015174488029858,
|
| 458 |
+
"policy_rollout_success_rate": 0.7821782178217822,
|
| 459 |
+
"restore_max_error": 2.384185791015625e-07
|
| 460 |
+
},
|
| 461 |
+
"StackCube-v1": {
|
| 462 |
+
"action_mse_to_best": 0.48127473745909,
|
| 463 |
+
"expert_success_rate": 0.7666666666666667,
|
| 464 |
+
"num_groups": 90,
|
| 465 |
+
"oracle_success_rate": 0.9111111111111111,
|
| 466 |
+
"policy_expert_regret": 1.1606415937344232,
|
| 467 |
+
"policy_oracle_regret": 1.3098470245798428,
|
| 468 |
+
"policy_rollout_progress": 0.4055012404918671,
|
| 469 |
+
"policy_rollout_success_rate": 0.16666666666666666,
|
| 470 |
+
"restore_max_error": 2.2351741790771484e-07
|
| 471 |
+
}
|
| 472 |
+
}
|
| 473 |
+
}
|
| 474 |
+
]
|
| 475 |
+
}
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_stacknowg_summary.md
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# h=16 Best-Policy Checkpoint Rollout
|
| 2 |
+
|
| 3 |
+
Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
|
| 4 |
+
Objective: `near_miss_policy_bc5`
|
| 5 |
+
Result file: `policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p01_stacknowg.json`
|
| 6 |
+
Completed seeds: 3
|
| 7 |
+
Baseline h=4 policy success: 29.67%
|
| 8 |
+
Baseline h=16 rank-checkpoint success: 29.74%
|
| 9 |
+
|
| 10 |
+
Mean success: 36.00% +/- 1.77%
|
| 11 |
+
Gain vs h=16 rank checkpoint: +6.26%
|
| 12 |
+
Mean progress: 57.12%
|
| 13 |
+
Mean action MSE to best: 0.401
|
| 14 |
+
|
| 15 |
+
| seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
|
| 16 |
+
|---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 17 |
+
| 0 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type mask residual_wrong_gripper=StackCube-v1 + challenger residual_near_miss@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 35.65% | 55.47% | 85.74% | n/a | n/a | 0.385 |
|
| 18 |
+
| 1 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type mask residual_wrong_gripper=StackCube-v1 + challenger residual_near_miss@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 34.43% | 56.90% | 86.96% | n/a | n/a | 0.396 |
|
| 19 |
+
| 2 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type mask residual_wrong_gripper=StackCube-v1 + challenger residual_near_miss@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 37.91% | 58.99% | 87.65% | n/a | n/a | 0.421 |
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_challenger0p01_norm_summary.json
ADDED
|
@@ -0,0 +1,478 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
|
| 3 |
+
"objective": "near_miss_policy_bc5",
|
| 4 |
+
"out_name": "policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_challenger0p01_norm.json",
|
| 5 |
+
"num_completed": 3,
|
| 6 |
+
"baseline_h4_policy_success": 0.2967,
|
| 7 |
+
"baseline_h16_rank_checkpoint_success": 0.29739130434782607,
|
| 8 |
+
"mean_success": 0.36057971014492757,
|
| 9 |
+
"std_success": 0.015780472566743406,
|
| 10 |
+
"mean_progress": 0.574088785921579,
|
| 11 |
+
"mean_action_mse_to_best": 0.41923150534080206,
|
| 12 |
+
"gain_vs_h4": 0.06387971014492755,
|
| 13 |
+
"gain_vs_h16_rank_checkpoint": 0.0631884057971015,
|
| 14 |
+
"rows": [
|
| 15 |
+
{
|
| 16 |
+
"seed": 0,
|
| 17 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_0/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_challenger0p01_norm.json",
|
| 18 |
+
"num_groups": 575,
|
| 19 |
+
"selection_mode": "retrieval_residual",
|
| 20 |
+
"num_candidates": 48,
|
| 21 |
+
"candidate_sigma": 0.0,
|
| 22 |
+
"proposal_types": [],
|
| 23 |
+
"proposal_lattice_types": [],
|
| 24 |
+
"selection_margin": 0.2,
|
| 25 |
+
"prepend_policy_candidate": false,
|
| 26 |
+
"field_optim_steps": 0,
|
| 27 |
+
"field_optim_step_size": 0.0,
|
| 28 |
+
"field_optim_trust_radius": 0.0,
|
| 29 |
+
"field_optim_l2_penalty": 0.0,
|
| 30 |
+
"retrieval_neighbors": 4,
|
| 31 |
+
"retrieval_metric": "raw",
|
| 32 |
+
"retrieval_type_min_success": 0.0,
|
| 33 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 34 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 35 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 36 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 37 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 38 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 39 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 40 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 41 |
+
"retrieval_residual_scale": 1.0,
|
| 42 |
+
"retrieval_residual_scales": [
|
| 43 |
+
0.35,
|
| 44 |
+
0.4,
|
| 45 |
+
0.45
|
| 46 |
+
],
|
| 47 |
+
"retrieval_residual_anchor": "expert",
|
| 48 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 49 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 50 |
+
"retrieval_residual_challenger_types": [
|
| 51 |
+
"residual_near_miss",
|
| 52 |
+
"residual_wrong_gripper"
|
| 53 |
+
],
|
| 54 |
+
"retrieval_residual_challenger_scales": [],
|
| 55 |
+
"retrieval_residual_challenger_margin": 0.01,
|
| 56 |
+
"retrieval_residual_challenger_tasks": [],
|
| 57 |
+
"retrieval_residual_challenger_type_tasks": {
|
| 58 |
+
"residual_wrong_gripper": [
|
| 59 |
+
"PickCube-v1",
|
| 60 |
+
"PullCube-v1"
|
| 61 |
+
]
|
| 62 |
+
},
|
| 63 |
+
"lattice_exclude_types": [
|
| 64 |
+
"residual_random_negative",
|
| 65 |
+
"residual_wrong_direction",
|
| 66 |
+
"residual_near_miss+residual_no_op"
|
| 67 |
+
],
|
| 68 |
+
"candidate_type_bonuses": {
|
| 69 |
+
"retrieval_residual_no_op": 0.03
|
| 70 |
+
},
|
| 71 |
+
"candidate_type_bonus_components": false,
|
| 72 |
+
"candidate_oracle_rollouts": 0,
|
| 73 |
+
"candidate_oracle_unique_tolerance": 1e-06,
|
| 74 |
+
"candidate_oracle_success_rate": null,
|
| 75 |
+
"candidate_oracle_progress": null,
|
| 76 |
+
"candidate_oracle_score": null,
|
| 77 |
+
"candidate_oracle_regret": null,
|
| 78 |
+
"candidate_oracle_score_gain_over_selected": null,
|
| 79 |
+
"candidate_oracle_unique_count": null,
|
| 80 |
+
"candidate_oracle_improvement_rate": null,
|
| 81 |
+
"candidate_oracle_selected_branch_success_rate": null,
|
| 82 |
+
"candidate_oracle_selected_branch_progress": null,
|
| 83 |
+
"candidate_oracle_type_counts": {},
|
| 84 |
+
"candidate_oracle_best_branch_rank": null,
|
| 85 |
+
"candidate_oracle_best_branch_rank_counts": {},
|
| 86 |
+
"candidate_oracle_branch_success_rates": [],
|
| 87 |
+
"candidate_oracle_branch_progress": [],
|
| 88 |
+
"candidate_oracle_branch_score_gains_over_selected": [],
|
| 89 |
+
"selected_residual_scale_counts": {
|
| 90 |
+
"0.35": 505,
|
| 91 |
+
"0.45": 62,
|
| 92 |
+
"0.4": 8
|
| 93 |
+
},
|
| 94 |
+
"policy_rollout_success_rate": 0.3582608695652174,
|
| 95 |
+
"policy_rollout_progress": 0.559952266274722,
|
| 96 |
+
"oracle_success_rate": 0.8573913043478261,
|
| 97 |
+
"action_mse_to_best": 0.39629442195691494,
|
| 98 |
+
"best_policy_val": {
|
| 99 |
+
"bc_loss": 0.13721593966086706,
|
| 100 |
+
"field_effect_loss": 0.009290305380192068,
|
| 101 |
+
"field_potential_loss": 0.2666468388504452,
|
| 102 |
+
"field_preference_loss": 0.5130573478009965,
|
| 103 |
+
"lattice_edges": 3833.3333333333335,
|
| 104 |
+
"progress_mae": 0.1933159919248687,
|
| 105 |
+
"rank_acc": 0.8265031774838766,
|
| 106 |
+
"rank_loss": 0.5130523675017886,
|
| 107 |
+
"regret_mae": 0.3756548762321472,
|
| 108 |
+
"success_accuracy": 0.8773836526605818,
|
| 109 |
+
"total_loss": 1.5581054819954767
|
| 110 |
+
},
|
| 111 |
+
"per_task": {
|
| 112 |
+
"LiftPegUpright-v1": {
|
| 113 |
+
"action_mse_to_best": 0.32276735697709713,
|
| 114 |
+
"expert_success_rate": 0.8865979381443299,
|
| 115 |
+
"num_groups": 97,
|
| 116 |
+
"oracle_success_rate": 0.9278350515463918,
|
| 117 |
+
"policy_expert_regret": 1.0783334703175063,
|
| 118 |
+
"policy_oracle_regret": 1.1047489480259491,
|
| 119 |
+
"policy_rollout_progress": 0.5714634232914325,
|
| 120 |
+
"policy_rollout_success_rate": 0.2268041237113402,
|
| 121 |
+
"restore_max_error": 2.384185791015625e-07
|
| 122 |
+
},
|
| 123 |
+
"PickCube-v1": {
|
| 124 |
+
"action_mse_to_best": 0.3064204964142006,
|
| 125 |
+
"expert_success_rate": 0.9375,
|
| 126 |
+
"num_groups": 208,
|
| 127 |
+
"oracle_success_rate": 0.9471153846153846,
|
| 128 |
+
"policy_expert_regret": 0.9335763099818275,
|
| 129 |
+
"policy_oracle_regret": 0.9469991774083331,
|
| 130 |
+
"policy_rollout_progress": 0.6244595646141813,
|
| 131 |
+
"policy_rollout_success_rate": 0.3605769230769231,
|
| 132 |
+
"restore_max_error": 2.384185791015625e-07
|
| 133 |
+
},
|
| 134 |
+
"PullCube-v1": {
|
| 135 |
+
"action_mse_to_best": 0.6037605747387007,
|
| 136 |
+
"expert_success_rate": 0.19480519480519481,
|
| 137 |
+
"num_groups": 77,
|
| 138 |
+
"oracle_success_rate": 0.36363636363636365,
|
| 139 |
+
"policy_expert_regret": 0.24754985438728278,
|
| 140 |
+
"policy_oracle_regret": 0.4211023997069932,
|
| 141 |
+
"policy_rollout_progress": 0.31856763445232034,
|
| 142 |
+
"policy_rollout_success_rate": 0.22077922077922077,
|
| 143 |
+
"restore_max_error": 2.3096799850463867e-07
|
| 144 |
+
},
|
| 145 |
+
"PushCube-v1": {
|
| 146 |
+
"action_mse_to_best": 0.3675119048578942,
|
| 147 |
+
"expert_success_rate": 0.8617021276595744,
|
| 148 |
+
"num_groups": 94,
|
| 149 |
+
"oracle_success_rate": 0.9893617021276596,
|
| 150 |
+
"policy_expert_regret": 0.5508309629686335,
|
| 151 |
+
"policy_oracle_regret": 0.628381609441118,
|
| 152 |
+
"policy_rollout_progress": 0.7014056246014352,
|
| 153 |
+
"policy_rollout_success_rate": 0.6702127659574468,
|
| 154 |
+
"restore_max_error": 2.384185791015625e-07
|
| 155 |
+
},
|
| 156 |
+
"StackCube-v1": {
|
| 157 |
+
"action_mse_to_best": 0.5231284084118376,
|
| 158 |
+
"expert_success_rate": 0.6767676767676768,
|
| 159 |
+
"num_groups": 99,
|
| 160 |
+
"oracle_success_rate": 0.8585858585858586,
|
| 161 |
+
"policy_expert_regret": 0.9252789386294105,
|
| 162 |
+
"policy_oracle_regret": 1.0729591972900159,
|
| 163 |
+
"policy_rollout_progress": 0.46657752569275673,
|
| 164 |
+
"policy_rollout_success_rate": 0.29292929292929293,
|
| 165 |
+
"restore_max_error": 2.384185791015625e-07
|
| 166 |
+
}
|
| 167 |
+
}
|
| 168 |
+
},
|
| 169 |
+
{
|
| 170 |
+
"seed": 1,
|
| 171 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_challenger0p01_norm.json",
|
| 172 |
+
"num_groups": 575,
|
| 173 |
+
"selection_mode": "retrieval_residual",
|
| 174 |
+
"num_candidates": 48,
|
| 175 |
+
"candidate_sigma": 0.0,
|
| 176 |
+
"proposal_types": [],
|
| 177 |
+
"proposal_lattice_types": [],
|
| 178 |
+
"selection_margin": 0.2,
|
| 179 |
+
"prepend_policy_candidate": false,
|
| 180 |
+
"field_optim_steps": 0,
|
| 181 |
+
"field_optim_step_size": 0.0,
|
| 182 |
+
"field_optim_trust_radius": 0.0,
|
| 183 |
+
"field_optim_l2_penalty": 0.0,
|
| 184 |
+
"retrieval_neighbors": 4,
|
| 185 |
+
"retrieval_metric": "raw",
|
| 186 |
+
"retrieval_type_min_success": 0.0,
|
| 187 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 188 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 189 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 190 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 191 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 192 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 193 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 194 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 195 |
+
"retrieval_residual_scale": 1.0,
|
| 196 |
+
"retrieval_residual_scales": [
|
| 197 |
+
0.35,
|
| 198 |
+
0.4,
|
| 199 |
+
0.45
|
| 200 |
+
],
|
| 201 |
+
"retrieval_residual_anchor": "expert",
|
| 202 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 203 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 204 |
+
"retrieval_residual_challenger_types": [
|
| 205 |
+
"residual_near_miss",
|
| 206 |
+
"residual_wrong_gripper"
|
| 207 |
+
],
|
| 208 |
+
"retrieval_residual_challenger_scales": [],
|
| 209 |
+
"retrieval_residual_challenger_margin": 0.01,
|
| 210 |
+
"retrieval_residual_challenger_tasks": [],
|
| 211 |
+
"retrieval_residual_challenger_type_tasks": {
|
| 212 |
+
"residual_wrong_gripper": [
|
| 213 |
+
"PickCube-v1",
|
| 214 |
+
"PullCube-v1"
|
| 215 |
+
]
|
| 216 |
+
},
|
| 217 |
+
"lattice_exclude_types": [
|
| 218 |
+
"residual_random_negative",
|
| 219 |
+
"residual_wrong_direction",
|
| 220 |
+
"residual_near_miss+residual_no_op"
|
| 221 |
+
],
|
| 222 |
+
"candidate_type_bonuses": {
|
| 223 |
+
"retrieval_residual_no_op": 0.03
|
| 224 |
+
},
|
| 225 |
+
"candidate_type_bonus_components": false,
|
| 226 |
+
"candidate_oracle_rollouts": 0,
|
| 227 |
+
"candidate_oracle_unique_tolerance": 1e-06,
|
| 228 |
+
"candidate_oracle_success_rate": null,
|
| 229 |
+
"candidate_oracle_progress": null,
|
| 230 |
+
"candidate_oracle_score": null,
|
| 231 |
+
"candidate_oracle_regret": null,
|
| 232 |
+
"candidate_oracle_score_gain_over_selected": null,
|
| 233 |
+
"candidate_oracle_unique_count": null,
|
| 234 |
+
"candidate_oracle_improvement_rate": null,
|
| 235 |
+
"candidate_oracle_selected_branch_success_rate": null,
|
| 236 |
+
"candidate_oracle_selected_branch_progress": null,
|
| 237 |
+
"candidate_oracle_type_counts": {},
|
| 238 |
+
"candidate_oracle_best_branch_rank": null,
|
| 239 |
+
"candidate_oracle_best_branch_rank_counts": {},
|
| 240 |
+
"candidate_oracle_branch_success_rates": [],
|
| 241 |
+
"candidate_oracle_branch_progress": [],
|
| 242 |
+
"candidate_oracle_branch_score_gains_over_selected": [],
|
| 243 |
+
"selected_residual_scale_counts": {
|
| 244 |
+
"0.35": 516,
|
| 245 |
+
"0.45": 51,
|
| 246 |
+
"0.4": 8
|
| 247 |
+
},
|
| 248 |
+
"policy_rollout_success_rate": 0.34608695652173915,
|
| 249 |
+
"policy_rollout_progress": 0.5719372708739145,
|
| 250 |
+
"oracle_success_rate": 0.8695652173913043,
|
| 251 |
+
"action_mse_to_best": 0.41279180422911177,
|
| 252 |
+
"best_policy_val": {
|
| 253 |
+
"bc_loss": 0.1315989200439718,
|
| 254 |
+
"field_effect_loss": 0.010110370814800262,
|
| 255 |
+
"field_potential_loss": 0.31546536915832096,
|
| 256 |
+
"field_preference_loss": 0.4960637440284093,
|
| 257 |
+
"lattice_edges": 3833.3333333333335,
|
| 258 |
+
"progress_mae": 0.19023226698239645,
|
| 259 |
+
"rank_acc": 0.8255469501018524,
|
| 260 |
+
"rank_loss": 0.4960531195004781,
|
| 261 |
+
"regret_mae": 0.42156532737943864,
|
| 262 |
+
"success_accuracy": 0.8807473679383596,
|
| 263 |
+
"total_loss": 1.561984618504842
|
| 264 |
+
},
|
| 265 |
+
"per_task": {
|
| 266 |
+
"LiftPegUpright-v1": {
|
| 267 |
+
"action_mse_to_best": 0.3281630301287612,
|
| 268 |
+
"expert_success_rate": 0.8584070796460177,
|
| 269 |
+
"num_groups": 113,
|
| 270 |
+
"oracle_success_rate": 0.9380530973451328,
|
| 271 |
+
"policy_expert_regret": 0.987445328615408,
|
| 272 |
+
"policy_oracle_regret": 1.0666473587002374,
|
| 273 |
+
"policy_rollout_progress": 0.6354643670858535,
|
| 274 |
+
"policy_rollout_success_rate": 0.23008849557522124,
|
| 275 |
+
"restore_max_error": 2.384185791015625e-07
|
| 276 |
+
},
|
| 277 |
+
"PickCube-v1": {
|
| 278 |
+
"action_mse_to_best": 0.36477837357265147,
|
| 279 |
+
"expert_success_rate": 0.9402173913043478,
|
| 280 |
+
"num_groups": 184,
|
| 281 |
+
"oracle_success_rate": 0.9456521739130435,
|
| 282 |
+
"policy_expert_regret": 1.08854141729155,
|
| 283 |
+
"policy_oracle_regret": 1.08856542176648,
|
| 284 |
+
"policy_rollout_progress": 0.5675299971506161,
|
| 285 |
+
"policy_rollout_success_rate": 0.2826086956521739,
|
| 286 |
+
"restore_max_error": 2.384185791015625e-07
|
| 287 |
+
},
|
| 288 |
+
"PullCube-v1": {
|
| 289 |
+
"action_mse_to_best": 0.5960272875075278,
|
| 290 |
+
"expert_success_rate": 0.25,
|
| 291 |
+
"num_groups": 76,
|
| 292 |
+
"oracle_success_rate": 0.40789473684210525,
|
| 293 |
+
"policy_expert_regret": 0.27355212223176895,
|
| 294 |
+
"policy_oracle_regret": 0.4974974064940685,
|
| 295 |
+
"policy_rollout_progress": 0.30258720147570495,
|
| 296 |
+
"policy_rollout_success_rate": 0.19736842105263158,
|
| 297 |
+
"restore_max_error": 2.384185791015625e-07
|
| 298 |
+
},
|
| 299 |
+
"PushCube-v1": {
|
| 300 |
+
"action_mse_to_best": 0.3595503710948669,
|
| 301 |
+
"expert_success_rate": 0.8198198198198198,
|
| 302 |
+
"num_groups": 111,
|
| 303 |
+
"oracle_success_rate": 1.0,
|
| 304 |
+
"policy_expert_regret": 0.3128753544779511,
|
| 305 |
+
"policy_oracle_regret": 0.33004292161078064,
|
| 306 |
+
"policy_rollout_progress": 0.8411282495603906,
|
| 307 |
+
"policy_rollout_success_rate": 0.8288288288288288,
|
| 308 |
+
"restore_max_error": 2.3795291781425476e-07
|
| 309 |
+
},
|
| 310 |
+
"StackCube-v1": {
|
| 311 |
+
"action_mse_to_best": 0.526873398326583,
|
| 312 |
+
"expert_success_rate": 0.6923076923076923,
|
| 313 |
+
"num_groups": 91,
|
| 314 |
+
"oracle_success_rate": 0.8571428571428571,
|
| 315 |
+
"policy_expert_regret": 1.132382218654339,
|
| 316 |
+
"policy_oracle_regret": 1.2920440658108219,
|
| 317 |
+
"policy_rollout_progress": 0.398561261348672,
|
| 318 |
+
"policy_rollout_success_rate": 0.15384615384615385,
|
| 319 |
+
"restore_max_error": 1.9371509552001953e-07
|
| 320 |
+
}
|
| 321 |
+
}
|
| 322 |
+
},
|
| 323 |
+
{
|
| 324 |
+
"seed": 2,
|
| 325 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_challenger0p01_norm.json",
|
| 326 |
+
"num_groups": 575,
|
| 327 |
+
"selection_mode": "retrieval_residual",
|
| 328 |
+
"num_candidates": 48,
|
| 329 |
+
"candidate_sigma": 0.0,
|
| 330 |
+
"proposal_types": [],
|
| 331 |
+
"proposal_lattice_types": [],
|
| 332 |
+
"selection_margin": 0.2,
|
| 333 |
+
"prepend_policy_candidate": false,
|
| 334 |
+
"field_optim_steps": 0,
|
| 335 |
+
"field_optim_step_size": 0.0,
|
| 336 |
+
"field_optim_trust_radius": 0.0,
|
| 337 |
+
"field_optim_l2_penalty": 0.0,
|
| 338 |
+
"retrieval_neighbors": 4,
|
| 339 |
+
"retrieval_metric": "raw",
|
| 340 |
+
"retrieval_type_min_success": 0.0,
|
| 341 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 342 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 343 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 344 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 345 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 346 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 347 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 348 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 349 |
+
"retrieval_residual_scale": 1.0,
|
| 350 |
+
"retrieval_residual_scales": [
|
| 351 |
+
0.35,
|
| 352 |
+
0.4,
|
| 353 |
+
0.45
|
| 354 |
+
],
|
| 355 |
+
"retrieval_residual_anchor": "expert",
|
| 356 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 357 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 358 |
+
"retrieval_residual_challenger_types": [
|
| 359 |
+
"residual_near_miss",
|
| 360 |
+
"residual_wrong_gripper"
|
| 361 |
+
],
|
| 362 |
+
"retrieval_residual_challenger_scales": [],
|
| 363 |
+
"retrieval_residual_challenger_margin": 0.01,
|
| 364 |
+
"retrieval_residual_challenger_tasks": [],
|
| 365 |
+
"retrieval_residual_challenger_type_tasks": {
|
| 366 |
+
"residual_wrong_gripper": [
|
| 367 |
+
"PickCube-v1",
|
| 368 |
+
"PullCube-v1"
|
| 369 |
+
]
|
| 370 |
+
},
|
| 371 |
+
"lattice_exclude_types": [
|
| 372 |
+
"residual_random_negative",
|
| 373 |
+
"residual_wrong_direction",
|
| 374 |
+
"residual_near_miss+residual_no_op"
|
| 375 |
+
],
|
| 376 |
+
"candidate_type_bonuses": {
|
| 377 |
+
"retrieval_residual_no_op": 0.03
|
| 378 |
+
},
|
| 379 |
+
"candidate_type_bonus_components": false,
|
| 380 |
+
"candidate_oracle_rollouts": 0,
|
| 381 |
+
"candidate_oracle_unique_tolerance": 1e-06,
|
| 382 |
+
"candidate_oracle_success_rate": null,
|
| 383 |
+
"candidate_oracle_progress": null,
|
| 384 |
+
"candidate_oracle_score": null,
|
| 385 |
+
"candidate_oracle_regret": null,
|
| 386 |
+
"candidate_oracle_score_gain_over_selected": null,
|
| 387 |
+
"candidate_oracle_unique_count": null,
|
| 388 |
+
"candidate_oracle_improvement_rate": null,
|
| 389 |
+
"candidate_oracle_selected_branch_success_rate": null,
|
| 390 |
+
"candidate_oracle_selected_branch_progress": null,
|
| 391 |
+
"candidate_oracle_type_counts": {},
|
| 392 |
+
"candidate_oracle_best_branch_rank": null,
|
| 393 |
+
"candidate_oracle_best_branch_rank_counts": {},
|
| 394 |
+
"candidate_oracle_branch_success_rates": [],
|
| 395 |
+
"candidate_oracle_branch_progress": [],
|
| 396 |
+
"candidate_oracle_branch_score_gains_over_selected": [],
|
| 397 |
+
"selected_residual_scale_counts": {
|
| 398 |
+
"0.35": 485,
|
| 399 |
+
"0.45": 70,
|
| 400 |
+
"0.4": 20
|
| 401 |
+
},
|
| 402 |
+
"policy_rollout_success_rate": 0.3773913043478261,
|
| 403 |
+
"policy_rollout_progress": 0.5903768206161002,
|
| 404 |
+
"oracle_success_rate": 0.8765217391304347,
|
| 405 |
+
"action_mse_to_best": 0.4486082898363795,
|
| 406 |
+
"best_policy_val": {
|
| 407 |
+
"bc_loss": 0.11367896075050037,
|
| 408 |
+
"field_effect_loss": 0.009670218582161598,
|
| 409 |
+
"field_potential_loss": 0.2641640139950646,
|
| 410 |
+
"field_preference_loss": 0.5130490180518892,
|
| 411 |
+
"lattice_edges": 3833.3333333333335,
|
| 412 |
+
"progress_mae": 0.2021729110015763,
|
| 413 |
+
"rank_acc": 0.8333857821093665,
|
| 414 |
+
"rank_loss": 0.5130119257503085,
|
| 415 |
+
"regret_mae": 0.3958987047274907,
|
| 416 |
+
"success_accuracy": 0.8680730561415354,
|
| 417 |
+
"total_loss": 1.4394984311527677
|
| 418 |
+
},
|
| 419 |
+
"per_task": {
|
| 420 |
+
"LiftPegUpright-v1": {
|
| 421 |
+
"action_mse_to_best": 0.35602021238203935,
|
| 422 |
+
"expert_success_rate": 0.8229166666666666,
|
| 423 |
+
"num_groups": 96,
|
| 424 |
+
"oracle_success_rate": 0.9270833333333334,
|
| 425 |
+
"policy_expert_regret": 0.8476849501021206,
|
| 426 |
+
"policy_oracle_regret": 0.9539546399998168,
|
| 427 |
+
"policy_rollout_progress": 0.6463205499264101,
|
| 428 |
+
"policy_rollout_success_rate": 0.3333333333333333,
|
| 429 |
+
"restore_max_error": 1.955777406692505e-07
|
| 430 |
+
},
|
| 431 |
+
"PickCube-v1": {
|
| 432 |
+
"action_mse_to_best": 0.39817221942498826,
|
| 433 |
+
"expert_success_rate": 0.9444444444444444,
|
| 434 |
+
"num_groups": 198,
|
| 435 |
+
"oracle_success_rate": 0.9595959595959596,
|
| 436 |
+
"policy_expert_regret": 0.9662310388127361,
|
| 437 |
+
"policy_oracle_regret": 0.9733144893414445,
|
| 438 |
+
"policy_rollout_progress": 0.6401566648197294,
|
| 439 |
+
"policy_rollout_success_rate": 0.3484848484848485,
|
| 440 |
+
"restore_max_error": 2.384185791015625e-07
|
| 441 |
+
},
|
| 442 |
+
"PullCube-v1": {
|
| 443 |
+
"action_mse_to_best": 0.6421529037257035,
|
| 444 |
+
"expert_success_rate": 0.24444444444444444,
|
| 445 |
+
"num_groups": 90,
|
| 446 |
+
"oracle_success_rate": 0.4666666666666667,
|
| 447 |
+
"policy_expert_regret": 0.2687155395539271,
|
| 448 |
+
"policy_oracle_regret": 0.45486517548561095,
|
| 449 |
+
"policy_rollout_progress": 0.36988404384917684,
|
| 450 |
+
"policy_rollout_success_rate": 0.26666666666666666,
|
| 451 |
+
"restore_max_error": 2.384185791015625e-07
|
| 452 |
+
},
|
| 453 |
+
"PushCube-v1": {
|
| 454 |
+
"action_mse_to_best": 0.397049247997232,
|
| 455 |
+
"expert_success_rate": 0.8514851485148515,
|
| 456 |
+
"num_groups": 101,
|
| 457 |
+
"oracle_success_rate": 1.0,
|
| 458 |
+
"policy_expert_regret": 0.3614020819711213,
|
| 459 |
+
"policy_oracle_regret": 0.41630433337523204,
|
| 460 |
+
"policy_rollout_progress": 0.8015174488029858,
|
| 461 |
+
"policy_rollout_success_rate": 0.7821782178217822,
|
| 462 |
+
"restore_max_error": 2.384185791015625e-07
|
| 463 |
+
},
|
| 464 |
+
"StackCube-v1": {
|
| 465 |
+
"action_mse_to_best": 0.5226443493117888,
|
| 466 |
+
"expert_success_rate": 0.7666666666666667,
|
| 467 |
+
"num_groups": 90,
|
| 468 |
+
"oracle_success_rate": 0.9111111111111111,
|
| 469 |
+
"policy_expert_regret": 1.1748778588242001,
|
| 470 |
+
"policy_oracle_regret": 1.3369984101918009,
|
| 471 |
+
"policy_rollout_progress": 0.4047339239054256,
|
| 472 |
+
"policy_rollout_success_rate": 0.14444444444444443,
|
| 473 |
+
"restore_max_error": 2.2351741790771484e-07
|
| 474 |
+
}
|
| 475 |
+
}
|
| 476 |
+
}
|
| 477 |
+
]
|
| 478 |
+
}
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_challenger0p01_norm_summary.md
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# h=16 Best-Policy Checkpoint Rollout
|
| 2 |
+
|
| 3 |
+
Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
|
| 4 |
+
Objective: `near_miss_policy_bc5`
|
| 5 |
+
Result file: `policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmglobal_wgpickpull_challenger0p01_norm.json`
|
| 6 |
+
Completed seeds: 3
|
| 7 |
+
Baseline h=4 policy success: 29.67%
|
| 8 |
+
Baseline h=16 rank-checkpoint success: 29.74%
|
| 9 |
+
|
| 10 |
+
Mean success: 36.06% +/- 1.58%
|
| 11 |
+
Gain vs h=16 rank checkpoint: +6.32%
|
| 12 |
+
Mean progress: 57.41%
|
| 13 |
+
Mean action MSE to best: 0.419
|
| 14 |
+
|
| 15 |
+
| seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
|
| 16 |
+
|---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 17 |
+
| 0 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss,residual_wrong_gripper with residual_wrong_gripper=PickCube-v1,PullCube-v1@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 35.83% | 56.00% | 85.74% | n/a | n/a | 0.396 |
|
| 18 |
+
| 1 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss,residual_wrong_gripper with residual_wrong_gripper=PickCube-v1,PullCube-v1@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 34.61% | 57.19% | 86.96% | n/a | n/a | 0.413 |
|
| 19 |
+
| 2 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss,residual_wrong_gripper with residual_wrong_gripper=PickCube-v1,PullCube-v1@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 37.74% | 59.04% | 87.65% | n/a | n/a | 0.449 |
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_norm_summary.json
ADDED
|
@@ -0,0 +1,472 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"run_root": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs",
|
| 3 |
+
"objective": "near_miss_policy_bc5",
|
| 4 |
+
"out_name": "policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_norm.json",
|
| 5 |
+
"num_completed": 3,
|
| 6 |
+
"baseline_h4_policy_success": 0.2967,
|
| 7 |
+
"baseline_h16_rank_checkpoint_success": 0.29739130434782607,
|
| 8 |
+
"mean_success": 0.36,
|
| 9 |
+
"std_success": 0.01659024698116427,
|
| 10 |
+
"mean_progress": 0.5735963408033935,
|
| 11 |
+
"mean_action_mse_to_best": 0.41794628433532255,
|
| 12 |
+
"gain_vs_h4": 0.06329999999999997,
|
| 13 |
+
"gain_vs_h16_rank_checkpoint": 0.06260869565217392,
|
| 14 |
+
"rows": [
|
| 15 |
+
{
|
| 16 |
+
"seed": 0,
|
| 17 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_0/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_norm.json",
|
| 18 |
+
"num_groups": 575,
|
| 19 |
+
"selection_mode": "retrieval_residual",
|
| 20 |
+
"num_candidates": 48,
|
| 21 |
+
"candidate_sigma": 0.0,
|
| 22 |
+
"proposal_types": [],
|
| 23 |
+
"proposal_lattice_types": [],
|
| 24 |
+
"selection_margin": 0.2,
|
| 25 |
+
"prepend_policy_candidate": false,
|
| 26 |
+
"field_optim_steps": 0,
|
| 27 |
+
"field_optim_step_size": 0.0,
|
| 28 |
+
"field_optim_trust_radius": 0.0,
|
| 29 |
+
"field_optim_l2_penalty": 0.0,
|
| 30 |
+
"retrieval_neighbors": 4,
|
| 31 |
+
"retrieval_metric": "raw",
|
| 32 |
+
"retrieval_type_min_success": 0.0,
|
| 33 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 34 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 35 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 36 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 37 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 38 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 39 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 40 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 41 |
+
"retrieval_residual_scale": 1.0,
|
| 42 |
+
"retrieval_residual_scales": [
|
| 43 |
+
0.35,
|
| 44 |
+
0.4,
|
| 45 |
+
0.45
|
| 46 |
+
],
|
| 47 |
+
"retrieval_residual_anchor": "expert",
|
| 48 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 49 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 50 |
+
"retrieval_residual_challenger_types": [
|
| 51 |
+
"residual_near_miss",
|
| 52 |
+
"residual_wrong_gripper"
|
| 53 |
+
],
|
| 54 |
+
"retrieval_residual_challenger_scales": [],
|
| 55 |
+
"retrieval_residual_challenger_margin": 0.01,
|
| 56 |
+
"retrieval_residual_challenger_tasks": [
|
| 57 |
+
"PickCube-v1",
|
| 58 |
+
"PullCube-v1"
|
| 59 |
+
],
|
| 60 |
+
"retrieval_residual_challenger_type_tasks": {},
|
| 61 |
+
"lattice_exclude_types": [
|
| 62 |
+
"residual_random_negative",
|
| 63 |
+
"residual_wrong_direction",
|
| 64 |
+
"residual_near_miss+residual_no_op"
|
| 65 |
+
],
|
| 66 |
+
"candidate_type_bonuses": {
|
| 67 |
+
"retrieval_residual_no_op": 0.03
|
| 68 |
+
},
|
| 69 |
+
"candidate_type_bonus_components": false,
|
| 70 |
+
"candidate_oracle_rollouts": 0,
|
| 71 |
+
"candidate_oracle_unique_tolerance": 1e-06,
|
| 72 |
+
"candidate_oracle_success_rate": null,
|
| 73 |
+
"candidate_oracle_progress": null,
|
| 74 |
+
"candidate_oracle_score": null,
|
| 75 |
+
"candidate_oracle_regret": null,
|
| 76 |
+
"candidate_oracle_score_gain_over_selected": null,
|
| 77 |
+
"candidate_oracle_unique_count": null,
|
| 78 |
+
"candidate_oracle_improvement_rate": null,
|
| 79 |
+
"candidate_oracle_selected_branch_success_rate": null,
|
| 80 |
+
"candidate_oracle_selected_branch_progress": null,
|
| 81 |
+
"candidate_oracle_type_counts": {},
|
| 82 |
+
"candidate_oracle_best_branch_rank": null,
|
| 83 |
+
"candidate_oracle_best_branch_rank_counts": {},
|
| 84 |
+
"candidate_oracle_branch_success_rates": [],
|
| 85 |
+
"candidate_oracle_branch_progress": [],
|
| 86 |
+
"candidate_oracle_branch_score_gains_over_selected": [],
|
| 87 |
+
"selected_residual_scale_counts": {
|
| 88 |
+
"0.35": 512,
|
| 89 |
+
"0.45": 56,
|
| 90 |
+
"0.4": 7
|
| 91 |
+
},
|
| 92 |
+
"policy_rollout_success_rate": 0.3582608695652174,
|
| 93 |
+
"policy_rollout_progress": 0.559967816375007,
|
| 94 |
+
"oracle_success_rate": 0.8573913043478261,
|
| 95 |
+
"action_mse_to_best": 0.3939786813845453,
|
| 96 |
+
"best_policy_val": {
|
| 97 |
+
"bc_loss": 0.13721593966086706,
|
| 98 |
+
"field_effect_loss": 0.009290305380192068,
|
| 99 |
+
"field_potential_loss": 0.2666468388504452,
|
| 100 |
+
"field_preference_loss": 0.5130573478009965,
|
| 101 |
+
"lattice_edges": 3833.3333333333335,
|
| 102 |
+
"progress_mae": 0.1933159919248687,
|
| 103 |
+
"rank_acc": 0.8265031774838766,
|
| 104 |
+
"rank_loss": 0.5130523675017886,
|
| 105 |
+
"regret_mae": 0.3756548762321472,
|
| 106 |
+
"success_accuracy": 0.8773836526605818,
|
| 107 |
+
"total_loss": 1.5581054819954767
|
| 108 |
+
},
|
| 109 |
+
"per_task": {
|
| 110 |
+
"LiftPegUpright-v1": {
|
| 111 |
+
"action_mse_to_best": 0.3206862742465335,
|
| 112 |
+
"expert_success_rate": 0.8865979381443299,
|
| 113 |
+
"num_groups": 97,
|
| 114 |
+
"oracle_success_rate": 0.9278350515463918,
|
| 115 |
+
"policy_expert_regret": 1.0783334703175063,
|
| 116 |
+
"policy_oracle_regret": 1.1047489480259491,
|
| 117 |
+
"policy_rollout_progress": 0.5714634232914325,
|
| 118 |
+
"policy_rollout_success_rate": 0.2268041237113402,
|
| 119 |
+
"restore_max_error": 2.384185791015625e-07
|
| 120 |
+
},
|
| 121 |
+
"PickCube-v1": {
|
| 122 |
+
"action_mse_to_best": 0.3064204964142006,
|
| 123 |
+
"expert_success_rate": 0.9375,
|
| 124 |
+
"num_groups": 208,
|
| 125 |
+
"oracle_success_rate": 0.9471153846153846,
|
| 126 |
+
"policy_expert_regret": 0.9335763099818275,
|
| 127 |
+
"policy_oracle_regret": 0.9469991774083331,
|
| 128 |
+
"policy_rollout_progress": 0.6244595646141813,
|
| 129 |
+
"policy_rollout_success_rate": 0.3605769230769231,
|
| 130 |
+
"restore_max_error": 2.384185791015625e-07
|
| 131 |
+
},
|
| 132 |
+
"PullCube-v1": {
|
| 133 |
+
"action_mse_to_best": 0.6037605747387007,
|
| 134 |
+
"expert_success_rate": 0.19480519480519481,
|
| 135 |
+
"num_groups": 77,
|
| 136 |
+
"oracle_success_rate": 0.36363636363636365,
|
| 137 |
+
"policy_expert_regret": 0.24754985438728278,
|
| 138 |
+
"policy_oracle_regret": 0.4211023997069932,
|
| 139 |
+
"policy_rollout_progress": 0.31856763445232034,
|
| 140 |
+
"policy_rollout_success_rate": 0.22077922077922077,
|
| 141 |
+
"restore_max_error": 2.3096799850463867e-07
|
| 142 |
+
},
|
| 143 |
+
"PushCube-v1": {
|
| 144 |
+
"action_mse_to_best": 0.3675119048578942,
|
| 145 |
+
"expert_success_rate": 0.8617021276595744,
|
| 146 |
+
"num_groups": 94,
|
| 147 |
+
"oracle_success_rate": 0.9893617021276596,
|
| 148 |
+
"policy_expert_regret": 0.5508309629686335,
|
| 149 |
+
"policy_oracle_regret": 0.628381609441118,
|
| 150 |
+
"policy_rollout_progress": 0.7014056246014352,
|
| 151 |
+
"policy_rollout_success_rate": 0.6702127659574468,
|
| 152 |
+
"restore_max_error": 2.384185791015625e-07
|
| 153 |
+
},
|
| 154 |
+
"StackCube-v1": {
|
| 155 |
+
"action_mse_to_best": 0.5117174406921623,
|
| 156 |
+
"expert_success_rate": 0.6767676767676768,
|
| 157 |
+
"num_groups": 99,
|
| 158 |
+
"oracle_success_rate": 0.8585858585858586,
|
| 159 |
+
"policy_expert_regret": 0.925188622390381,
|
| 160 |
+
"policy_oracle_regret": 1.0728688810509865,
|
| 161 |
+
"policy_rollout_progress": 0.4666678419317862,
|
| 162 |
+
"policy_rollout_success_rate": 0.29292929292929293,
|
| 163 |
+
"restore_max_error": 2.384185791015625e-07
|
| 164 |
+
}
|
| 165 |
+
}
|
| 166 |
+
},
|
| 167 |
+
{
|
| 168 |
+
"seed": 1,
|
| 169 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_1/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_norm.json",
|
| 170 |
+
"num_groups": 575,
|
| 171 |
+
"selection_mode": "retrieval_residual",
|
| 172 |
+
"num_candidates": 48,
|
| 173 |
+
"candidate_sigma": 0.0,
|
| 174 |
+
"proposal_types": [],
|
| 175 |
+
"proposal_lattice_types": [],
|
| 176 |
+
"selection_margin": 0.2,
|
| 177 |
+
"prepend_policy_candidate": false,
|
| 178 |
+
"field_optim_steps": 0,
|
| 179 |
+
"field_optim_step_size": 0.0,
|
| 180 |
+
"field_optim_trust_radius": 0.0,
|
| 181 |
+
"field_optim_l2_penalty": 0.0,
|
| 182 |
+
"retrieval_neighbors": 4,
|
| 183 |
+
"retrieval_metric": "raw",
|
| 184 |
+
"retrieval_type_min_success": 0.0,
|
| 185 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 186 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 187 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 188 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 189 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 190 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 191 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 192 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 193 |
+
"retrieval_residual_scale": 1.0,
|
| 194 |
+
"retrieval_residual_scales": [
|
| 195 |
+
0.35,
|
| 196 |
+
0.4,
|
| 197 |
+
0.45
|
| 198 |
+
],
|
| 199 |
+
"retrieval_residual_anchor": "expert",
|
| 200 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 201 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 202 |
+
"retrieval_residual_challenger_types": [
|
| 203 |
+
"residual_near_miss",
|
| 204 |
+
"residual_wrong_gripper"
|
| 205 |
+
],
|
| 206 |
+
"retrieval_residual_challenger_scales": [],
|
| 207 |
+
"retrieval_residual_challenger_margin": 0.01,
|
| 208 |
+
"retrieval_residual_challenger_tasks": [
|
| 209 |
+
"PickCube-v1",
|
| 210 |
+
"PullCube-v1"
|
| 211 |
+
],
|
| 212 |
+
"retrieval_residual_challenger_type_tasks": {},
|
| 213 |
+
"lattice_exclude_types": [
|
| 214 |
+
"residual_random_negative",
|
| 215 |
+
"residual_wrong_direction",
|
| 216 |
+
"residual_near_miss+residual_no_op"
|
| 217 |
+
],
|
| 218 |
+
"candidate_type_bonuses": {
|
| 219 |
+
"retrieval_residual_no_op": 0.03
|
| 220 |
+
},
|
| 221 |
+
"candidate_type_bonus_components": false,
|
| 222 |
+
"candidate_oracle_rollouts": 0,
|
| 223 |
+
"candidate_oracle_unique_tolerance": 1e-06,
|
| 224 |
+
"candidate_oracle_success_rate": null,
|
| 225 |
+
"candidate_oracle_progress": null,
|
| 226 |
+
"candidate_oracle_score": null,
|
| 227 |
+
"candidate_oracle_regret": null,
|
| 228 |
+
"candidate_oracle_score_gain_over_selected": null,
|
| 229 |
+
"candidate_oracle_unique_count": null,
|
| 230 |
+
"candidate_oracle_improvement_rate": null,
|
| 231 |
+
"candidate_oracle_selected_branch_success_rate": null,
|
| 232 |
+
"candidate_oracle_selected_branch_progress": null,
|
| 233 |
+
"candidate_oracle_type_counts": {},
|
| 234 |
+
"candidate_oracle_best_branch_rank": null,
|
| 235 |
+
"candidate_oracle_best_branch_rank_counts": {},
|
| 236 |
+
"candidate_oracle_branch_success_rates": [],
|
| 237 |
+
"candidate_oracle_branch_progress": [],
|
| 238 |
+
"candidate_oracle_branch_score_gains_over_selected": [],
|
| 239 |
+
"selected_residual_scale_counts": {
|
| 240 |
+
"0.35": 519,
|
| 241 |
+
"0.45": 48,
|
| 242 |
+
"0.4": 8
|
| 243 |
+
},
|
| 244 |
+
"policy_rollout_success_rate": 0.3443478260869565,
|
| 245 |
+
"policy_rollout_progress": 0.5704443854190733,
|
| 246 |
+
"oracle_success_rate": 0.8695652173913043,
|
| 247 |
+
"action_mse_to_best": 0.41151866878018434,
|
| 248 |
+
"best_policy_val": {
|
| 249 |
+
"bc_loss": 0.1315989200439718,
|
| 250 |
+
"field_effect_loss": 0.010110370814800262,
|
| 251 |
+
"field_potential_loss": 0.31546536915832096,
|
| 252 |
+
"field_preference_loss": 0.4960637440284093,
|
| 253 |
+
"lattice_edges": 3833.3333333333335,
|
| 254 |
+
"progress_mae": 0.19023226698239645,
|
| 255 |
+
"rank_acc": 0.8255469501018524,
|
| 256 |
+
"rank_loss": 0.4960531195004781,
|
| 257 |
+
"regret_mae": 0.42156532737943864,
|
| 258 |
+
"success_accuracy": 0.8807473679383596,
|
| 259 |
+
"total_loss": 1.561984618504842
|
| 260 |
+
},
|
| 261 |
+
"per_task": {
|
| 262 |
+
"LiftPegUpright-v1": {
|
| 263 |
+
"action_mse_to_best": 0.3255710795406351,
|
| 264 |
+
"expert_success_rate": 0.8584070796460177,
|
| 265 |
+
"num_groups": 113,
|
| 266 |
+
"oracle_success_rate": 0.9380530973451328,
|
| 267 |
+
"policy_expert_regret": 0.9873926745051831,
|
| 268 |
+
"policy_oracle_regret": 1.0665947045900126,
|
| 269 |
+
"policy_rollout_progress": 0.6355170211960784,
|
| 270 |
+
"policy_rollout_success_rate": 0.23008849557522124,
|
| 271 |
+
"restore_max_error": 2.384185791015625e-07
|
| 272 |
+
},
|
| 273 |
+
"PickCube-v1": {
|
| 274 |
+
"action_mse_to_best": 0.36477837357265147,
|
| 275 |
+
"expert_success_rate": 0.9402173913043478,
|
| 276 |
+
"num_groups": 184,
|
| 277 |
+
"oracle_success_rate": 0.9456521739130435,
|
| 278 |
+
"policy_expert_regret": 1.08854141729155,
|
| 279 |
+
"policy_oracle_regret": 1.08856542176648,
|
| 280 |
+
"policy_rollout_progress": 0.5675299971506161,
|
| 281 |
+
"policy_rollout_success_rate": 0.2826086956521739,
|
| 282 |
+
"restore_max_error": 2.384185791015625e-07
|
| 283 |
+
},
|
| 284 |
+
"PullCube-v1": {
|
| 285 |
+
"action_mse_to_best": 0.5960272875075278,
|
| 286 |
+
"expert_success_rate": 0.25,
|
| 287 |
+
"num_groups": 76,
|
| 288 |
+
"oracle_success_rate": 0.40789473684210525,
|
| 289 |
+
"policy_expert_regret": 0.27355212223176895,
|
| 290 |
+
"policy_oracle_regret": 0.4974974064940685,
|
| 291 |
+
"policy_rollout_progress": 0.30258720147570495,
|
| 292 |
+
"policy_rollout_success_rate": 0.19736842105263158,
|
| 293 |
+
"restore_max_error": 2.384185791015625e-07
|
| 294 |
+
},
|
| 295 |
+
"PushCube-v1": {
|
| 296 |
+
"action_mse_to_best": 0.3571035131677851,
|
| 297 |
+
"expert_success_rate": 0.8198198198198198,
|
| 298 |
+
"num_groups": 111,
|
| 299 |
+
"oracle_success_rate": 1.0,
|
| 300 |
+
"policy_expert_regret": 0.32961789887767656,
|
| 301 |
+
"policy_oracle_regret": 0.3467854660105061,
|
| 302 |
+
"policy_rollout_progress": 0.8333947141696741,
|
| 303 |
+
"policy_rollout_success_rate": 0.8198198198198198,
|
| 304 |
+
"restore_max_error": 2.3795291781425476e-07
|
| 305 |
+
},
|
| 306 |
+
"StackCube-v1": {
|
| 307 |
+
"action_mse_to_best": 0.5250320660543966,
|
| 308 |
+
"expert_success_rate": 0.6923076923076923,
|
| 309 |
+
"num_groups": 91,
|
| 310 |
+
"oracle_success_rate": 0.8571428571428571,
|
| 311 |
+
"policy_expert_regret": 1.1324474562655438,
|
| 312 |
+
"policy_oracle_regret": 1.2921093034220266,
|
| 313 |
+
"policy_rollout_progress": 0.39849602373746723,
|
| 314 |
+
"policy_rollout_success_rate": 0.15384615384615385,
|
| 315 |
+
"restore_max_error": 1.9371509552001953e-07
|
| 316 |
+
}
|
| 317 |
+
}
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"seed": 2,
|
| 321 |
+
"path": "/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs/near_miss_policy_bc5/seed_2/policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_norm.json",
|
| 322 |
+
"num_groups": 575,
|
| 323 |
+
"selection_mode": "retrieval_residual",
|
| 324 |
+
"num_candidates": 48,
|
| 325 |
+
"candidate_sigma": 0.0,
|
| 326 |
+
"proposal_types": [],
|
| 327 |
+
"proposal_lattice_types": [],
|
| 328 |
+
"selection_margin": 0.2,
|
| 329 |
+
"prepend_policy_candidate": false,
|
| 330 |
+
"field_optim_steps": 0,
|
| 331 |
+
"field_optim_step_size": 0.0,
|
| 332 |
+
"field_optim_trust_radius": 0.0,
|
| 333 |
+
"field_optim_l2_penalty": 0.0,
|
| 334 |
+
"retrieval_neighbors": 4,
|
| 335 |
+
"retrieval_metric": "raw",
|
| 336 |
+
"retrieval_type_min_success": 0.0,
|
| 337 |
+
"retrieval_type_success_bonus_scale": 0.0,
|
| 338 |
+
"retrieval_residual_consensus_penalty_scale": 0.0,
|
| 339 |
+
"retrieval_residual_min_source_progress": 0.0,
|
| 340 |
+
"retrieval_residual_min_source_advantage": -1000000000.0,
|
| 341 |
+
"retrieval_residual_source_progress_bonus_scale": 0.0,
|
| 342 |
+
"retrieval_residual_source_score_bonus_scale": 0.0,
|
| 343 |
+
"retrieval_residual_source_advantage_bonus_scale": 0.0,
|
| 344 |
+
"retrieval_residual_composite_l2_penalty_scale": 0.0,
|
| 345 |
+
"retrieval_residual_scale": 1.0,
|
| 346 |
+
"retrieval_residual_scales": [
|
| 347 |
+
0.35,
|
| 348 |
+
0.4,
|
| 349 |
+
0.45
|
| 350 |
+
],
|
| 351 |
+
"retrieval_residual_anchor": "expert",
|
| 352 |
+
"retrieval_residual_direction": "candidate_minus_anchor",
|
| 353 |
+
"retrieval_residual_reduce": "compose_mean_by_type",
|
| 354 |
+
"retrieval_residual_challenger_types": [
|
| 355 |
+
"residual_near_miss",
|
| 356 |
+
"residual_wrong_gripper"
|
| 357 |
+
],
|
| 358 |
+
"retrieval_residual_challenger_scales": [],
|
| 359 |
+
"retrieval_residual_challenger_margin": 0.01,
|
| 360 |
+
"retrieval_residual_challenger_tasks": [
|
| 361 |
+
"PickCube-v1",
|
| 362 |
+
"PullCube-v1"
|
| 363 |
+
],
|
| 364 |
+
"retrieval_residual_challenger_type_tasks": {},
|
| 365 |
+
"lattice_exclude_types": [
|
| 366 |
+
"residual_random_negative",
|
| 367 |
+
"residual_wrong_direction",
|
| 368 |
+
"residual_near_miss+residual_no_op"
|
| 369 |
+
],
|
| 370 |
+
"candidate_type_bonuses": {
|
| 371 |
+
"retrieval_residual_no_op": 0.03
|
| 372 |
+
},
|
| 373 |
+
"candidate_type_bonus_components": false,
|
| 374 |
+
"candidate_oracle_rollouts": 0,
|
| 375 |
+
"candidate_oracle_unique_tolerance": 1e-06,
|
| 376 |
+
"candidate_oracle_success_rate": null,
|
| 377 |
+
"candidate_oracle_progress": null,
|
| 378 |
+
"candidate_oracle_score": null,
|
| 379 |
+
"candidate_oracle_regret": null,
|
| 380 |
+
"candidate_oracle_score_gain_over_selected": null,
|
| 381 |
+
"candidate_oracle_unique_count": null,
|
| 382 |
+
"candidate_oracle_improvement_rate": null,
|
| 383 |
+
"candidate_oracle_selected_branch_success_rate": null,
|
| 384 |
+
"candidate_oracle_selected_branch_progress": null,
|
| 385 |
+
"candidate_oracle_type_counts": {},
|
| 386 |
+
"candidate_oracle_best_branch_rank": null,
|
| 387 |
+
"candidate_oracle_best_branch_rank_counts": {},
|
| 388 |
+
"candidate_oracle_branch_success_rates": [],
|
| 389 |
+
"candidate_oracle_branch_progress": [],
|
| 390 |
+
"candidate_oracle_branch_score_gains_over_selected": [],
|
| 391 |
+
"selected_residual_scale_counts": {
|
| 392 |
+
"0.35": 485,
|
| 393 |
+
"0.45": 70,
|
| 394 |
+
"0.4": 20
|
| 395 |
+
},
|
| 396 |
+
"policy_rollout_success_rate": 0.3773913043478261,
|
| 397 |
+
"policy_rollout_progress": 0.5903768206161002,
|
| 398 |
+
"oracle_success_rate": 0.8765217391304347,
|
| 399 |
+
"action_mse_to_best": 0.4483415028412381,
|
| 400 |
+
"best_policy_val": {
|
| 401 |
+
"bc_loss": 0.11367896075050037,
|
| 402 |
+
"field_effect_loss": 0.009670218582161598,
|
| 403 |
+
"field_potential_loss": 0.2641640139950646,
|
| 404 |
+
"field_preference_loss": 0.5130490180518892,
|
| 405 |
+
"lattice_edges": 3833.3333333333335,
|
| 406 |
+
"progress_mae": 0.2021729110015763,
|
| 407 |
+
"rank_acc": 0.8333857821093665,
|
| 408 |
+
"rank_loss": 0.5130119257503085,
|
| 409 |
+
"regret_mae": 0.3958987047274907,
|
| 410 |
+
"success_accuracy": 0.8680730561415354,
|
| 411 |
+
"total_loss": 1.4394984311527677
|
| 412 |
+
},
|
| 413 |
+
"per_task": {
|
| 414 |
+
"LiftPegUpright-v1": {
|
| 415 |
+
"action_mse_to_best": 0.35442226944239036,
|
| 416 |
+
"expert_success_rate": 0.8229166666666666,
|
| 417 |
+
"num_groups": 96,
|
| 418 |
+
"oracle_success_rate": 0.9270833333333334,
|
| 419 |
+
"policy_expert_regret": 0.8476849501021206,
|
| 420 |
+
"policy_oracle_regret": 0.9539546399998168,
|
| 421 |
+
"policy_rollout_progress": 0.6463205499264101,
|
| 422 |
+
"policy_rollout_success_rate": 0.3333333333333333,
|
| 423 |
+
"restore_max_error": 1.955777406692505e-07
|
| 424 |
+
},
|
| 425 |
+
"PickCube-v1": {
|
| 426 |
+
"action_mse_to_best": 0.39817221942498826,
|
| 427 |
+
"expert_success_rate": 0.9444444444444444,
|
| 428 |
+
"num_groups": 198,
|
| 429 |
+
"oracle_success_rate": 0.9595959595959596,
|
| 430 |
+
"policy_expert_regret": 0.9662310388127361,
|
| 431 |
+
"policy_oracle_regret": 0.9733144893414445,
|
| 432 |
+
"policy_rollout_progress": 0.6401566648197294,
|
| 433 |
+
"policy_rollout_success_rate": 0.3484848484848485,
|
| 434 |
+
"restore_max_error": 2.384185791015625e-07
|
| 435 |
+
},
|
| 436 |
+
"PullCube-v1": {
|
| 437 |
+
"action_mse_to_best": 0.6421529037257035,
|
| 438 |
+
"expert_success_rate": 0.24444444444444444,
|
| 439 |
+
"num_groups": 90,
|
| 440 |
+
"oracle_success_rate": 0.4666666666666667,
|
| 441 |
+
"policy_expert_regret": 0.2687155395539271,
|
| 442 |
+
"policy_oracle_regret": 0.45486517548561095,
|
| 443 |
+
"policy_rollout_progress": 0.36988404384917684,
|
| 444 |
+
"policy_rollout_success_rate": 0.26666666666666666,
|
| 445 |
+
"restore_max_error": 2.384185791015625e-07
|
| 446 |
+
},
|
| 447 |
+
"PushCube-v1": {
|
| 448 |
+
"action_mse_to_best": 0.397049247997232,
|
| 449 |
+
"expert_success_rate": 0.8514851485148515,
|
| 450 |
+
"num_groups": 101,
|
| 451 |
+
"oracle_success_rate": 1.0,
|
| 452 |
+
"policy_expert_regret": 0.3614020819711213,
|
| 453 |
+
"policy_oracle_regret": 0.41630433337523204,
|
| 454 |
+
"policy_rollout_progress": 0.8015174488029858,
|
| 455 |
+
"policy_rollout_success_rate": 0.7821782178217822,
|
| 456 |
+
"restore_max_error": 2.384185791015625e-07
|
| 457 |
+
},
|
| 458 |
+
"StackCube-v1": {
|
| 459 |
+
"action_mse_to_best": 0.5226443493117888,
|
| 460 |
+
"expert_success_rate": 0.7666666666666667,
|
| 461 |
+
"num_groups": 90,
|
| 462 |
+
"oracle_success_rate": 0.9111111111111111,
|
| 463 |
+
"policy_expert_regret": 1.1748778588242001,
|
| 464 |
+
"policy_oracle_regret": 1.3369984101918009,
|
| 465 |
+
"policy_rollout_progress": 0.4047339239054256,
|
| 466 |
+
"policy_rollout_success_rate": 0.14444444444444443,
|
| 467 |
+
"restore_max_error": 2.2351741790771484e-07
|
| 468 |
+
}
|
| 469 |
+
}
|
| 470 |
+
}
|
| 471 |
+
]
|
| 472 |
+
}
|
results/h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_norm_summary.md
ADDED
|
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# h=16 Best-Policy Checkpoint Rollout
|
| 2 |
+
|
| 3 |
+
Run root: `/scratch/knguy52/dovla/experiments/dovla_h16_policy_ckpt_runs`
|
| 4 |
+
Objective: `near_miss_policy_bc5`
|
| 5 |
+
Result file: `policy_rollout_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmwgchallenger0p01_pickpull_norm.json`
|
| 6 |
+
Completed seeds: 3
|
| 7 |
+
Baseline h=4 policy success: 29.67%
|
| 8 |
+
Baseline h=16 rank-checkpoint success: 29.74%
|
| 9 |
+
|
| 10 |
+
Mean success: 36.00% +/- 1.66%
|
| 11 |
+
Gain vs h=16 rank checkpoint: +6.26%
|
| 12 |
+
Mean progress: 57.36%
|
| 13 |
+
Mean action MSE to best: 0.418
|
| 14 |
+
|
| 15 |
+
| seed | mode | k | policy cand | retrieval K | retrieval metric | residual anchor | residual direction | residual reduce | min type success | type success bonus | consensus penalty | min source progress | source progress bonus | source score bonus | residual scale | residual scales | margin | sigma | opt steps | trust | success | progress | oracle | candidate oracle | oracle gain | action MSE |
|
| 16 |
+
|---:|---|---:|---|---:|---|---|---|---|---:|---:|---:|---:|---:|---:|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---:|
|
| 17 |
+
| 0 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss,residual_wrong_gripper on PickCube-v1,PullCube-v1@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 35.83% | 56.00% | 85.74% | n/a | n/a | 0.394 |
|
| 18 |
+
| 1 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss,residual_wrong_gripper on PickCube-v1,PullCube-v1@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 34.43% | 57.04% | 86.96% | n/a | n/a | 0.412 |
|
| 19 |
+
| 2 | retrieval_residual | 48 | no | 4 | raw | expert | candidate_minus_anchor | compose_mean_by_type + challenger residual_near_miss,residual_wrong_gripper on PickCube-v1,PullCube-v1@0.01 | 0.00 | 0.000 | 0.000 | 0.00 | 0.000 | 0.000 | 1.00 | 0.35,0.40,0.45 | 0.200 | 0.00 | 0 | 0.00 | 37.74% | 59.04% | 87.65% | n/a | n/a | 0.448 |
|