MJLab Nightly Benchmark

Updated: 2025-12-08 16:30
What is this?

Every night, we train a policy to imitate a reference motion on the Unitree G1 using the latest MJLab commit. The trained policy is then evaluated across 1024 trials, with the results logged here to catch regressions over time.

All Runs
Date Commit Run Success Rate MPKPE (m) EE Pos Error (m)
What is this?

Measures simulation throughput (steps per second) for different tasks. Physics FPS is raw MuJoCo stepping speed (just mj_step). Env FPS is the full environment step including observations, rewards, terminations, and resets. Overhead is the percentage slowdown from environment logic on top of physics.

Note: These benchmarks use zero actions, which causes frequent terminations and resets. This represents a worst-case scenario for overhead since resets are expensive.

All Runs
Date Commit Task Physics FPS Env FPS Overhead