Isaac Gym: High Performance GPU-Based Physics
Simulation For Robot Learning🤖
Viktor Makoviychuk, Lukasz Wawrzyniak, Yunrong Guo, Michelle Lu, Kier Storey, Miles Macklin, David Hoeller, Nikita Rudin, Arthur Allshire, Ankur Handa, Gavriel State
End-to-end GPU accelerated simulator for robotics and AI research
Abstract
Isaac Gym offers a high performance learning platform to train policies for wide variety of robotics tasks directly on GPU. Both physics simulation and neural network policy training reside on GPU and communicate by directly passing data from physics buffers to PyTorch tensors without ever going through any CPU bottlenecks. This leads to blazing fast training times for complex robotics tasks on a single GPU with 1-2 orders of magnitude improvements compared to conventional RL training that uses a CPU based simulator and GPU for neural networks. Isaac gym can be downloaded at https://developer.nvidia.com/isaac-gym
This work is predated by our previous effort on accelerating physics simulation on GPU: https://sites.google.com/view/accelerated-gpu-simulation/home
Benchmarking Simulation Performance
Ant
Rewards and effective FPS with respect to number of parallel environments for the Ant experiment. Best training time is achieved with 8192 environments and a horizon lengths of 16.
Humanoid
Rewards and effective FPS with respect to number of parallel environments for the Humanoid experiment. Best training time is achieved with 4096 environments and a horizon lengths of 32.
Rewards and effective FPS with respect to number of parallel environments for the Humanoid experiment. Best training time is achieved with both 4096 and 8192 environments and horizon lengths of 64 and 32 respectively.
Shadow Hand
Rewards and effective FPS with respect to number of parallel environments for the Shadow Hand experiment. Best training time is achieved with both 8192 and 16384 environments and horizon lengths of 16 and 8 respectively.
Benchmarking Environment Performance
Ant
Humanoid
Shadow Hand Feed Forward Policy
Shadow Hand LSTM Policy
Allegro Hand Standard
ANYmal Robot
AMP: Adversarial Motion Priors