Massively parallel rigidbody physics simulation on accelerator hardware

BRAX

Brax is a differentiable physics engine that simulates environments made up of rigid bodies, joints, and actuators. It’s also a suite of learning algorithms to train agents to operate in these environments (PPO, SAC, evolutionary strategy, and direct trajectory optimization are implemented).

Brax is written in JAX and is designed for use on acceleration hardware. It is both efficient for single-core training, and scalable to massively parallel simulation, without the need for pesky datacenters.

ant

fetch

grasp

halfcheetah

humanoid

Some policies trained via Brax. Brax simulates these environments at millions of physics steps per second on TPU.

Colab Notebooks

Explore Brax easily and quickly through a series of colab notebooks: