mjswan: Real-Time MuJoCo Physics Simulation on WebAssembly

Primary Source: https://github.com/ttktjmt/mjswan
Author: ttktjmt · Year: 2026 · Publisher: ttktjmt

Summary

A WebAssembly port of DeepMind’s MuJoCo physics engine integrated with in-browser neural networks for real-time interactive reinforcement learning and robotics simulation directly inside web browsers.

Key Points

• Runs complex articulated rigid-body contact dynamics at 60 FPS in browser client threads.
• Direct ONNX / WebGPU neural policy inference driving simulated bipedal and quadrupedal robots.
• Interactive parameter perturbation, gravity manipulation, and joint damping in real time.

Mentioned Pages

Entities

• ttktjmt

Concepts

• WebAssembly Physics Simulation
• In-Browser Reinforcement Learning