Replay and check every run

All 330 runs in the dataset, drawn on their arena. Pick a task, agent and trial, scrub through the run, or overlay other agents to compare paths.

Synthetic data. These runs come from simulated agents and show how results are verified, not how real robots or people perform.

How the checks work

This page does not trust the leaderboard. It re-counts every run from the raw positions recorded in its trace (waypoints reached in order, objects that ended inside the goal zone, floor area swept) and compares that count with what the scorer reported. It also checks that time only moves forward, that samples are never more than half a second apart, that no robot moves faster than 3 m/s or leaves the arena, and that every run fits its time limit. A run that fails any check is listed so you can open it.

Anyone can run the same checks on their own data with motionbench viz your-data/ -o explorer.html. Get the toolkit or read how it works.

Prefer a full window? Open the explorer on its own.