TARSphereBench
Back to the project

The authored benchmark

Everyday worlds.
Changing conditions.

30 scenes. Two task families. Six paired conditions.
Explore the places, requests and causal steps behind all 360 episodes.

360 authored episodes

These are authored tasks, not measured robot runs. Reference routes are planned on each scene’s official occupancy grid with 0.25 m G1 clearance. Maps load as they enter view; episode definitions are available as JSON in the details.

To targetReturn / deliver

Causal DAG

Solid · outcome   Dashed · event

The initial request

Events & dialogue conditions

Authored episodeEpisode JSONCompare six conditions

Reference routes are planned on static occupancy, not recorded model trajectories. The map does not simulate moving people, perception changes or skill outages. Every DAG edge and live guard comes from the authored definition. Events gate dependent outcomes without adding score.