embodybench

benchmark platform

HomeNew runRunsAPI docs

API documentation

Protocol references for the EmbodyBench evaluation API.

Draft / temporary documentation

These specs are still evolving. The shapes here match the deployed backend today, but treat them as a snapshot — they may change between platform releases. Canonical source lives at github.com/markli1hoshipu/hoshipu_website/docs.

Policy server guide

user-facing

For model implementers. Defines the three endpoints your server exposes (/episode/init, /step, /finalize), the action and state vectors for each action_space, image format, and includes a 40-line minimum-viable example.

Open

Simulator-side protocol spec

internal

For worker implementers. Documents what the simulator sends on the wire, observation serialization, lifecycle, retry policy, the failure-reason taxonomy used for auto-requeue, and the converters embodybench_remote uses for each action_space.

Open