Ground truthEstimated poseOccupied / unknown cellsGoal
Explore a warehouse mission from LiDAR mapping to crate interaction. Compare a learned Q-table controller with the heuristic controller using the same offline simulation seed.
Play the mapping and crate-pushing sequence. Toggle the occupancy estimate to reveal the true warehouse layout. Switch controllers and inspect the crate position, localization error and final outcome.
The original Python mission uses scan matching, log-odds mapping and navigation before local crate pushing. Both controller runs are generated from the repository’s offline simulator, using seed 4 and a 0.16 m map.
These are offline 2D surrogate runs, not Isaac Sim video or a live Isaac session. A single matched seed is a demonstration, not a general success-rate benchmark. Camera detections are simulated from geometry. The saved Q-table is reused without retraining. The project source also contains the separate Isaac integration.