PureJaxRL
A reinforcement learning library based on JAX that uses jax.vmap and jax.lax.scan to express the entire RL training loop as pure functions, executing thousands of environment instances in parallel on the GPU.
Timeline (last 90 days)
CARL was inspired by PureJaxRL, which had implemented a similar massively parallel environment scheme on JAX.
Massively parallel environments are an important direction in current reinforcement learning research, and work such as PureJaxRL and Isaac Gym is driving this trend.
PureJaxRL is a reinforcement learning library based on JAX that uses primitives such as jax.vmap and jax.lax.scan to express the entire RL training loop as pure functions, thereby executing thousands of environment instances in parallel on the GPU.
All Facts (3)
CARL was inspired by PureJaxRL, which had implemented a similar massively parallel environment scheme on JAX.
50%UnverifiedMassively parallel environments are an important direction in current reinforcement learning research, and work such as PureJaxRL and Isaac Gym is driving this trend.
50%UnverifiedPureJaxRL is a reinforcement learning library based on JAX that uses primitives such as jax.vmap and jax.lax.scan to express the entire RL training loop as pure functions, thereby executing thousands of environment instances in parallel on the GPU.
50%