crowd-bottleneck-56

Crowd evacuation · solved 10 of 10 agent attempts pass

The task. Open at this gate and the hardest June crowd scenario (three 20-proposal resumes failed before a 10-proposal run passed on 5 seeds).

That controller (candidate_0010) scores 0.775 on 40 seeds and is shipped as a non-passing reference.

Gate. >= 95 % evacuated and zero collision/wall/obstacle violations in >= 90 % of 40 verifier-only seeds

Axes: n_agents:56, observation:local, exit:random-narrow, obstacles:two-pillar-bottleneck, safety:zero-collision

Each video re-runs the attempt's final controller through the task's own evaluator on a public seed (never the grader's held-out ones). The table under each video is the official grade on the held-out seeds.

Best attemptSol (cursor-cli (pre-clock), 7 min) · obstacle_bottleneck_v3

Graded on the held-out seeds: passes every gate on the grader's held-out seeds.

criterion	graded	needed
base:success	1.000	≥ 0.900

How the agent got there: 7 min · 53 tool calls · 13 edit-tool edits · 28 runs of its own Python (tests, experiments, rewrites) · no tuner runs · official verifier 1× (success rates in order: 1.00).

The agent's final message

Implemented `/app/work/controller.py`. Validation: - Official seeds 0–4: **100% success** - Additional seeds 5–34: **100% success** - Zero collisions, wall violations, and obstacle violations - At least 54/56 evacuated on every tested seed

Controller family: social force / potential field

Decentralized controller that applies a velocity-tracking force toward the exit and sums repulsive terms from neighbors, pillars, and walls, with centerline attraction into the gap and a final acceleration clamp and stopping guard at the exit jamb.

Shown: public seed 0 (obstacle_bottleneck_v3); this episode passes the default rung. This episode: evacuation_fraction 1, collision_count 0, wall_violation_count 0, obstacle_violation_count 0, total_reward 1088.

controller.py · trial crowd-bottleneck-56__nrbseYi