Research map / Reinforcement learning / Visual and representation RL
Isaac Gym: High Performance GPU Based Physics Simulation For Robot Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| increase parallel environments from 256 to 8192 (ablation) | Isaac Gym PPO on Ant · Ant, time to 7000 reward | training time: 1000 seconds -> 100 seconds | Helped |
| increase parallel environments from 8192 to 16384 (ablation) | Isaac Gym PPO on Ant · Ant | training time | No clear effect |
| increase parallel environments from 256 to 4096 (ablation) | Isaac Gym PPO on Humanoid · Humanoid, time to 7000 reward | training time: 10^4 seconds -> 10^3 seconds | Helped |
| increase parallel environments beyond 4096 with short horizon (ablation) | Isaac Gym PPO on Humanoid · Humanoid | training time and gait quality | Hurt |
| double horizon length when using 8192 and 16384 environments (ablation) | Isaac Gym PPO on Humanoid · Humanoid | walking success | Helped |
| increase parallel environments from 256 to 16384 (ablation) | Isaac Gym PPO on Shadow Hand · Shadow Hand | training time: 5x10^4 seconds -> 3x10^3 seconds | Helped |
| switch from LSTM-free feed-forward to LSTM network with OpenAI observations | Shadow Hand OpenAI · Shadow Hand consecutive successes | consecutive successes: 20 -> 37 consecutive successes | Helped |
| fixed wrist and longer fingers (Allegro) versus movable wrist (Shadow Hand) | in-hand cube reorientation · consecutive successes | consecutive successes | Hurt |
| replace CPU simulation with GPU-based Isaac Gym | adversarial motion priors humanoid · Humanoid AMP imitation | training time: 30 hours -> 6 minutes | Helped |
Closest papers by meaning
- Large Batch Simulation for Deep Reinforcement Learning (ICLR 2021 · similarity 0.80)
- Efficient Learning of Mesh-Based Physical Simulation with Bi-Stride Multi-Scale Graph Neural Network (ICML 2023 · similarity 0.78)
- Staggered Environment Resets Improve Massively Parallel On-Policy Reinforcement Learning (NeurIPS 2025 · similarity 0.78)
- StitchCUDA: An Automated Multi-Agents End-to-End GPU Programing Framework with Rubric-based Agentic Reinforcement Learning (ICML 2026 · similarity 0.77)
- Accelerating Visual-Policy Learning through Parallel Differentiable Simulation (NeurIPS 2025 · similarity 0.77)
- DOGE-Train: Discrete Optimization on GPU with End-to-End Training (AAAI 2024 · similarity 0.76)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.