Research map / Reinforcement learning / Visual and representation RL
PhyPlan: Learning to Plan Tasks with Generalizable and Rapid Physical Reasoning for Embodied Manipulation
Ankit Kanwar, Hartej Soin, Abhinav Barnawal, Mudit Chopra, Harshil Vagadia, Tamajit Banerjee, Shreshth Tuli, Rohan Paul, Souvik Chakraborty
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 PhyPlan joint tool and control selection instead of MCTS-based Tool+Control Selector | PhyPlan · Bridge Task | average regret: 0.42 -> 0.15 | Improved |
| use PhyPlan instead of best baseline DQN | PhyPlan · Average over tasks | regret: 45% lower regret | Improved |
| use PhyPlan instead of MBPO | PhyPlan · Average over tasks | speed and regret: 3× faster, 70% improvement in regret | Improved |
| use an LLM planner instead of random | LLM · Bounce Task | average regret: 0.42 -> 0.41 | No clear effect |
| use LLM planner on Ricochet instead of random | LLM · Ricochet Task | average regret: 1.53 -> 1.90 | Hurt |
| add physics to data-driven dynamics predictor (physics-informed) (ablation) | PhyPlan dynamics predictor · Sliding/Swinging/Throwing trajectory prediction | relative L2 error | Improved |
| add GP-based online reward adaptation (ablation) | PhyPlan · Bounce Task | convergence toward goal | Improved |
| use soft-adaptive MCTS instead of conventional MCTS (ablation) | PhyPlan · Ricochet Task | regret and iterations | Improved |
| replace GFlowNet tool selector with MCTS (ablation) | PhyPlan · Launch Task | average regret: 0.05 -> 0.17 | Hurt |
Closest papers by meaning
- I-PHYRE: Interactive Physical Reasoning (ICLR 2024 · similarity 0.88)
- Multi-Stage Manipulation with Demonstration-Augmented Reward, Policy, and World Model Learning (ICML 2025 · similarity 0.87)
- UniDomain: Pretraining a Unified PDDL Domain from Real-World Demonstrations for Generalizable Robot Task Planning (NeurIPS 2025 · similarity 0.86)
- Embodied-R1.5: Evolving Physical Intelligence via Embodied Foundation Models (NeurIPS 2026 · similarity 0.86)
- Kinetix: Investigating the Training of General Agents through Open-Ended Physics-Based Control Tasks (ICLR 2025 · similarity 0.86)
- LaST-R1: Reinforcing Robotic Manipulation via Adaptive Physical Latent Reasoning (NeurIPS 2026 · similarity 0.86)
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.
