Planar point-agent simulation
Solver reliability
Command feasibility and physical audits
Baseline retained. Uncertified steps fell from 29 to 0; three physical-audit failures remained.
- Simulation episodes
- 896
- 448 old / revised pairs
- Uncertified steps
- 29 → 0
- Old projected → revised; study totals
- Physical-audit failures
- 0 → 3
- Old projected → revised; study totals
- Paired seeds per case
- 16
- 28 condition / policy cases
Recorded evidence
Results
448 matched controller pairs across seven policies and four conditions.
Case 1 of 28nominal · adaptive_lloyd_bandit
| Metric | Old projected | Revised | Revised − old | 95% interval lower | 95% interval upper |
|---|---|---|---|---|---|
| Uncertified steps / episode | 0 | 0 | 0 | 0 | 0 |
| Physical-audit failures / episode | 0 | 0 | 0 | 0 | 0 |
| Optimizer failures / episode | 0 | 0 | 0 | 0 | 0 |
| Recovered commands / episode | 0 | 0 | 0 | 0 | 0 |
| Observed area fraction | 0.419617 | 0.419617 | 0 | 0 | 0 |
| Map RMSE | 0.362636 | 0.362636 | 0 | 0 | 0 |
| Squared-speed energy proxy | 3.25919 | 3.25919 | 0 | 0 | 0 |
Case notes & acceptance
Safety gate: True. Progress/cost gate: True.
null = missing or not applicable, never zero. Intervals and sample sizes are reported in each table. Scroll tables horizontally to see every column.
Methods & interpretation
Zero collisions is not a pass: three tiny actuator-acceleration excesses remain physical-audit failures, and frontier progress/cost gates fail. The primary baseline is old projected, not repulsion. Hardening v2 rechecked saved finite commands; it did not rerun these episodes or measure new runtime. Descriptive Student-t intervals within this condition/policy; no multiplicity correction or optimality proof.