Research map / Reinforcement learning / Visual and representation RL
Large Batch Simulation for Deep Reinforcement 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 |
|---|---|---|---|
| Use batch simulation and rendering rather than independent simulator workers (ablation) | BPS for PointGoalNav Depth · RTX 3090 | FPS: 2800 -> 19900 | Helped |
| Use batch simulation and rendering rather than independent simulator workers (ablation) | BPS for PointGoalNav RGB · RTX 3090 | FPS: 990 -> 13300 | Helped |
| Use the SE-ResNet9 64x64 policy instead of ResNet50 128x128 (ablation) | BPS for PointGoalNav Depth · RTX 3090 | FPS: 2300 -> 19900 | Helped |
| Use the SE-ResNet9 64x64 policy instead of ResNet50 128x128 (ablation) | BPS for PointGoalNav RGB · RTX 3090 | FPS: 2000 -> 13300 | Helped |
| Reduce encoder depth and input resolution (ablation) | BPS Depth agent · Gibson validation and test | SPL: 94.4 -> 95.6; 91.5 -> 94.4 | Hurt |
| Use 64x64 RGB input rather than 128x128 (ablation) | BPS RGB agent · Gibson validation | SPL: 87.8{ $\pm$0.7} -> 88.4{ $[\pm$0.9} | Mixed |
| Use 64x64 RGB input rather than 128x128 (ablation) | BPS RGB agent · Gibson test | SPL: 85.6 -> 83.7 | Hurt |
| Increase simulation batch size from N=256 to N=1024 (ablation) | BPS Depth agent · PointGoalNav training | SPL | No clear effect |
| Increase simulation batch size to N=4096 rather than N=1024 (ablation) | BPS Depth agent · PointGoalNav training | SPL: 20% | Hurt |
| Use Lamb instead of Adam (ablation) | BPS Depth agent · PointGoalNav training after 1 billion samples | SPL: 1.5% | Helped |
| Use Lamb instead of Adam (ablation) | BPS Depth agent · PointGoalNav training after 2.5 billion frames | SPL: 0.6% | Helped |
Closest papers by meaning
- Accelerating Visual-Policy Learning through Parallel Differentiable Simulation (NeurIPS 2025 · similarity 0.87)
- Staggered Environment Resets Improve Massively Parallel On-Policy Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- NAVIX: Scaling MiniGrid Environments with JAX (NeurIPS 2025 · similarity 0.86)
- GPUDrive: Data-driven, multi-agent driving simulation at 1 million FPS (ICLR 2025 · similarity 0.85)
- Mind the GAP! The Challenges of Scale in Pixel-based Deep Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
- Memo: Training Memory-Efficient Embodied Agents with Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
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.