Differentiable Separating-Axis Clearance Barrier / bench_report.json

Mechanism confirmed, baseline not beaten

Raw ⬇ ZIP
  1{
  2  "bench_version": 1,
  3  "track": "obb_clearance_trajectory",
  4  "model": "mlp_tiny",
  5  "metric_direction": "lower is better",
  6  "n_seeds": 8,
  7  "baseline": {
  8    "best_cfg": {
  9      "lr": 0.01,
 10      "epochs": 24
 11    },
 12    "sweep": [
 13      {
 14        "cfg": {
 15          "lr": 0.001,
 16          "epochs": 24
 17        },
 18        "mean": 2.5900412797927856
 19      },
 20      {
 21        "cfg": {
 22          "lr": 0.003,
 23          "epochs": 24
 24        },
 25        "mean": 1.0762107968330383
 26      },
 27      {
 28        "cfg": {
 29          "lr": 0.01,
 30          "epochs": 24
 31        },
 32        "mean": 0.08821548242121935
 33      }
 34    ],
 35    "full": {
 36      "mean": 0.11757882172241807,
 37      "std": 0.04988984373475929,
 38      "per_seed": [
 39        0.11706195771694183,
 40        0.1304021030664444,
 41        0.03879903629422188,
 42        0.06659883260726929,
 43        0.1332305371761322,
 44        0.21891747415065765,
 45        0.13242962956428528,
 46        0.10319100320339203
 47      ],
 48      "n": 8
 49    }
 50  },
 51  "idea": {
 52    "mean": 0.1973940273746848,
 53    "std": 0.1311616228648052,
 54    "per_seed": [
 55      0.126096710562706,
 56      0.13761737942695618,
 57      0.15668733417987823,
 58      0.12452193349599838,
 59      0.203719824552536,
 60      0.5375758409500122,
 61      0.12353026866912842,
 62      0.16940292716026306
 63    ],
 64    "n": 8,
 65    "signature_values": [
 66      [
 67        0.22611111402511597,
 68        -0.5267328023910522,
 69        0.3687500059604645
 70      ],
 71      [
 72        0.14370578527450562,
 73        -0.4949674904346466,
 74        0.37312498688697815
 75      ],
 76      [
 77        0.3007834553718567,
 78        -0.5578157901763916,
 79        0.27250000834465027
 80      ],
 81      [
 82        0.20459963381290436,
 83        -0.4704010784626007,
 84        0.43437498807907104
 85      ],
 86      [
 87        0.22297126054763794,
 88        -0.26175621151924133,
 89        0.17687499523162842
 90      ],
 91      [
 92        0.6471734642982483,
 93        -0.03403729200363159,
 94        0.0012499999720603228
 95      ],
 96      [
 97        0.1714925915002823,
 98        -0.5904415249824524,
 99        0.4350000023841858
100      ],
101      [
102        0.2024228423833847,
103        -0.5173735618591309,
104        0.35874998569488525
105      ]
106    ]
107  },
108  "comparison": {
109    "delta_mean": 0.07981520565226674,
110    "idea_wins": 1,
111    "n_pairs": 8,
112    "per_seed_diffs": [
113      0.00903475284576416,
114      0.00721527636051178,
115      0.11788829788565636,
116      0.057923100888729095,
117      0.07048928737640381,
118      0.31865836679935455,
119      -0.00889936089515686,
120      0.06621192395687103
121    ],
122    "p_value": 0.02415,
123    "mde": 0.08792278503854178,
124    "mde_rel_pct": 74.77773952022693,
125    "verdict": "idea worse (significant)",
126    "system_worked": false
127  },
128  "mechanism_signature": {
129    "math_check": {
130      "predicted_exact_zero_m": 1.1,
131      "observed_exact_zero_m": 1.100000023841858,
132      "observed_smooth_zero_m": 1.0299999713897705,
133      "smoothing_bound": [
134        {
135          "tau": 0.2,
136          "error": 0.003949284553527832,
137          "bound": 0.2772588722239781
138        },
139        {
140          "tau": 0.1,
141          "error": 3.3736228942871094e-05,
142          "bound": 0.13862943611198905
143        },
144        {
145          "tau": 0.05,
146          "error": 0.0,
147          "bound": 0.06931471805599453
148        }
149      ],
150      "active_axis_gradient_dx": -0.0,
151      "passed": true
152    },
153    "idea_sweep": [
154      {
155        "cfg": {
156          "lr": 0.001,
157          "epochs": 24,
158          "lambda_barrier": 0.08,
159          "tau": 0.1,
160          "m0": 0.05,
161          "beta": 0.05
162        },
163        "mean": 2.590525209903717
164      },
165      {
166        "cfg": {
167          "lr": 0.003,
168          "epochs": 24,
169          "lambda_barrier": 0.08,
170          "tau": 0.1,
171          "m0": 0.05,
172          "beta": 0.05
173        },
174        "mean": 1.1768128275871277
175      },
176      {
177        "cfg": {
178          "lr": 0.01,
179          "epochs": 24,
180          "lambda_barrier": 0.08,
181          "tau": 0.1,
182          "m0": 0.05,
183          "beta": 0.05
184        },
185        "mean": 0.1362308394163847
186      }
187    ],
188    "predicted_vs_observed": {
189      "predicted_smoothed_margin_mean": 0.264907518401742,
190      "observed_exact_sat_margin_mean": -0.4316907189786434,
191      "predicted_collision_fraction": 0.30257812156924047
192    },
193    "prediction": "barrier should reduce collision fraction and move margins upward",
194    "confirmed": true
195  },
196  "custom_track": {
197    "name": "obb_clearance_trajectory",
198    "file": "obb_track.py",
199    "domain": "geometric_collision_avoidance"
200  }
201}