nanojaxterity — cartpole, in your browser

A policy trained in Jaxterity (calibrated MJX cartpole → self-contained PPO) stabilising the inverted pendulum and rejecting disturbances — every few seconds the pole gets a random shove and the learned controller catches it. Runs entirely client-side on MuJoCo-WASM (the official google-deepmind build) or a built-in closed-form fallback.

loading…