Go1 Joystick

Task goal

An agent is placed in an offline sandbox with the paper and the MJX playground environment. It must implement the paper's method or otherwise optimize the Episode reward metric, then be evaluated on hidden seeds by an independent evaluator over gym-over-gRPC.

Scoring

The evaluator computes the authoritative Episode reward. The harness normalizes it against the paper target:

score = clamp(measured / paper_target, 0, 1.5)

Higher is better. Beating the paper scores above 1.0, capped at 1.5. The release aggregate uses macro_mean.

Submission & evaluator

Runs are evaluated in Batch mode. The agent writes a submission to a shared volume; the evaluator runs its native pipeline.

# request a challenge, run locally, then submit the encrypted bundle
ale-bench run --task go1_joystick
ale-bench submit go1_joystick.ale-submission.tar.zst.age