Research map / Multi-agent systems and games / Human-AI and goal-conditioned
IPR-1: Interactive Physical Reasoner
Mingyu Zhang, Lifeng Zhuo, Tianxi Tan, Guocan Xie, Xian Nie, Yan Li, Renjie Zhao, Zizhu He, Ziyu Wang, Jiting Cai, Yong-Lu Li
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 |
|---|---|---|---|
| add IPR (world-model prediction plus GRPO) to Qwen3-VL-8B (ablation) | Qwen3-VL-8B · game benchmark survival | normalized mean: 0.105 -> 0.252 | Improved |
| add IPR to Qwen3-VL-8B (ablation) | Qwen3-VL-8B · game benchmark curiosity | normalized mean: 0.325 -> 1.173 | Improved |
| add IPR to Qwen3-VL-8B (ablation) | Qwen3-VL-8B · game benchmark utility | normalized mean: 0.176 -> 0.493 | Improved |
| add IPR to Qwen3-VL-8B (ablation) | Qwen3-VL-8B · game benchmark overall | average rank: 18.2 -> 4.9 | Improved |
| add behavior cloning to pretrained VLM (ablation) | Qwen3-VL-8B · survival | score: 0.62 -> 0.63 | No clear effect |
| add behavior cloning to pretrained VLM (ablation) | Qwen3-VL-8B · curiosity | score: 2.14 -> 1.88 | Hurt |
| add PPO to pretrained VLM (ablation) | Qwen3-VL-8B · survival | score: 0.62 -> 1.00 | Improved |
| add PPO to pretrained VLM (ablation) | Qwen3-VL-8B · curiosity | score: 2.14 -> 1.79 | Hurt |
| add BC on top of PPO (ablation) | Qwen3-VL-8B · survival | score: 1.00 -> 0.57 | Hurt |
| add GRPO instead of PPO (ablation) | Qwen3-VL-8B · survival | score: 1.00 -> 0.95 | No clear effect |
| add world-model prediction with GRPO (IPR) over VLM + GRPO (ablation) | Qwen3-VL-8B · curiosity | score: 1.78 -> 2.77 | Improved |
| add world-model prediction with GRPO (IPR) over VLM + PPO (ablation) | Qwen3-VL-8B · survival | score: 1.00 -> 0.76 | Hurt |
Closest papers by meaning
- I-PHYRE: Interactive Physical Reasoning (ICLR 2024 · similarity 0.91)
- PhysVLM-AVR: Active Visual Reasoning for Multimodal Large Language Models in Physical Environments (NeurIPS 2025 · similarity 0.88)
- Playing ZendoWorld: Challenging AI Agents on Active Visual Concept Induction (NeurIPS 2026 · similarity 0.88)
- BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games (ICLR 2025 · similarity 0.87)
- PhysBench: Benchmarking and Enhancing Vision-Language Models for Physical World Understanding (ICLR 2025 · similarity 0.87)
- Causal-PIK: Causality-based Physical Reasoning with a Physics-Informed Kernel (ICML 2025 · similarity 0.87)
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.
