Research map / Vision-language and multimodal / Robot action and navigation
IRASim: A Fine-Grained World Model for Robot Manipulation
Fangqi Zhu, Hongtao Wu, Song Guo, Yuxiao Liu, Chilam Cheang, Tao Kong
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 frame-level instead of video-level action conditioning (ablation) | IRASim · RT-1 short | PSNR: 25.446 -> 26.048 | Improved |
| use frame-level instead of video-level action conditioning (ablation) | IRASim · Bridge short | PSNR: 24.733 -> 25.275 | Improved |
| use frame-level instead of video-level action conditioning (ablation) | IRASim · Language-Table short | PSNR: 23.893 -> 28.818 | Improved |
| use frame-level instead of video-level action conditioning (ablation) | IRASim · Language-Table long trajectories | Latent L2: 0.2112 -> 0.1730 | Improved |
| use frame-level instead of video-level action conditioning (ablation) | IRASim · RT-1 short | FID: 4.34 -> 5.60 | Hurt |
| use 2 context frames instead of 1 (ablation) | IRASim · Bridge | PSNR: 25 -> 25 | No clear effect |
| train world model with post-trained rollouts (P=1000 vs P=0) (ablation) | IRASim planning · Push-T K=50 | IoU: 0.418 -> 0.961 | Improved |
| sample more trajectories (K=50 vs K=1) with P=1000 (ablation) | IRASim planning · Push-T | IoU: 0.637 -> 0.961 | Improved |
| sample more trajectories (K=50 vs K=1) without post-trained rollouts (P=0) (ablation) | IRASim planning · Push-T | IoU: 0.637 -> 0.418 | Hurt |
| increase post-trained rollouts from 500 to 1000 at K=5 (ablation) | IRASim planning · Push-T K=5 | IoU: 0.907 -> 0.886 | Hurt |
| use MSE instead of ResNet50 cosine similarity as value function (ablation) | IRASim planning · Real-robot Close Drawer | success rate: 0.60 -> 0.87 | Improved |
| use IRASim planning (MSE) instead of random trajectory selection | IRASim planning · Real-robot Close Drawer | success rate: 0.20 -> 0.87 | Improved |
Closest papers by meaning
- PIVOT-R: Primitive-Driven Waypoint-Aware World Model for Robotic Manipulation (NeurIPS 2024 · similarity 0.86)
- RobotSmith: Generative Robotic Tool Design for Acquisition of Complex Manipulation Skills (NeurIPS 2025 · similarity 0.86)
- AHA: A Vision-Language-Model for Detecting and Reasoning Over Failures in Robotic Manipulation (ICLR 2025 · similarity 0.85)
- HumanoidGen: Data Generation for Bimanual Dexterous Manipulation via LLM Reasoning (NeurIPS 2025 · similarity 0.85)
- RoboCerebra: A Large-scale Benchmark for Long-horizon Robotic Manipulation Evaluation (NeurIPS 2025 · similarity 0.85)
- OmniManip: Towards General Robotic Manipulation via Object-Centric Interaction Primitives as Spatial Constraints (CVPR 2025 · similarity 0.84)
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.
