gymnasium_rs

gymnasium_rs implements the Gymnasium environment interface natively in Rust, with Python bindings to the same environments. Its live demo compiles the environments to WebAssembly, so every simulation on the page runs on your own machine, with no server or recorded video.

The gymnasium_rs demo: a wall of live classic-control environments behind the title “Reinforcement-learning environments, running natively in your browser”

The live demo

  • Catalog: six native environments, each driven by a small hand-written controller that you can switch to random actions.
  • Swarm lab: up to 16,384 copies of one environment, each with its own randomized physics, under one controller or your own input.
  • Robustness map: a grid of physics parameters that shows where a controller succeeds and where it breaks.
  • Training: Oniro, a Rust reinforcement-learning trainer built on Burn, trains a PPO agent on 16 CartPoles in the browser. Its playground also offers SAC, TD-MPC2 and a Dreamer-style world model, and a WebGPU tier for browsers with a real GPU.

Status

gymnasium_rs is in pre-release development. Its source, crates and Python package are not public yet; the demo ships compiled WebAssembly only.