MetaLab

Engine-agnostic robot RL simulator — define a task once as a single Python contract and train/evaluate it identically on Newton and Genesis.

MetaLab is an engine-agnostic robot reinforcement-learning simulator framework. A task is written once as a single Python contract; the same environment then trains and evaluates on two physics engines, Newton and Genesis, through a unified API, on a single local GPU.

  • One contract file per task; engine-specific code lives in per-engine backends.
  • Pinned engine sources and per-engine uv environments from one setup script.
  • Web launchpad and CLI for training, evaluation, standalone simulation, and engine-parity comparison.