Isaac Gym: The Next Generation — High-performance Reinforcement Learning in Omniverse

We'll take a deep dive into the evolution of high-performance end-to-end GPU reinforcement learning capabilities within Omniverse, which will replace the Isaac Gym system. We'll provide an overview of the new physics tensor APIs available with Omniverse that power the system, and we'll delve into the new tasking system to demonstrate how to set up massively parallel reinforcement learning training.

Additionally, we will show how multiple GPUs can be used for training more complex RL agents, and provide updates on Sim-to-Real aspects of RL training, showing how policies learned in simulation run on real world robots.