Video World Models · 2026-11-05
ReactWorld: Structuring Interaction Responses for Embodied Video Generation
Hanwei Liu, Huijia Zhang, Hanzhang Wang, Guangyu Liu, Hanbei Zhan, Linjie Hu, Bin Xiao, Xuan Zhou, Dong Zhang
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
ReactWorld is an embodied video generation model that anticipates structured interaction responses in the generation backbone’s feature space, predicting participant roles, future motion response intensity, and interval motion. It uses a response feedback adapter that forms feature increments via differences from a zero-dynamics reference and converts predicted displacements into positional biases for temporal attention. The authors report improved task completion and physical plausibility on RBench and EZS-Bench.
Why it matters
The authors address a gap where visually plausible embodied video can still produce object responses that are inconsistent with task intent and contact dynamics. By explicitly structuring where/when/how objects should respond and feeding predicted local dynamics back into generation, ReactWorld aims to produce more reliable interaction generation in spatially local, time-critical regions.
Method
- ReactWorld predicts an interaction response field (participant roles, future motion response intensity, and interval motion) from intermediate backbone features within each denoising forward pass.
- A response feedback adapter updates object features using the difference between object outputs under predicted dynamics and a zero-dynamics reference.
- Predicted displacements are converted into positional biases for cross-frame temporal attention within a fixed local window.
Limitation
ReactWorld currently anticipates interactions using a 2D response representation, leaving room for explicit modeling of 3D interaction relations.
Abstract (from arXiv)
Embodied video world models predict scene changes caused by robot actions, yet generated object responses may remain inconsistent with task intent and contact dynamics. Critical responses in embodied scenes often concentrate in local regions and specific time intervals. Inspired by chain-of-thought reasoning, we introduce ReactWorld, an embodied video generation model guided by structured interaction anticipation. In the generation backbone's feature space, an interaction response field predictor anticipates how the interaction will unfold, explicitly describing participant roles, future motion response intensity, and local motion changes. A response feedback adapter forms increments to object features from the difference between the object branch's outputs with predicted and zeroed local dynamics inputs. It also converts predicted displacements into positional biases for temporal attention. These two mechanisms guide object feature updates and information retrieval across frames, respectively. Through interaction anticipation and generative feedback, ReactWorld achieves an average score of 77.51 on RBench, outperforming PhysisForcing by 3.29 points with improvements in all nine categories. On EZS-Bench, it achieves a Domain score of 86.98 and an average score of 81.87.
Related papers
- Video2World: Benchmarking Coding Agents for Interactive World Modeling from Embodied Videos
- Seeing and Imagining as One: Unified Video Generation and 4D Perception for Embodied World Modeling
- EmbodiedSmith: Scaling Embodied Data through Recursive Self-Improvement Flywheel in Simulation
- BoxWorld: Let World Engines Control and Video Generators Render
- BWM-Forge: Agentic World Programming for Embodied Intelligence
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-11-04 · Indexed by Pub-AI: 2026-11-05 UTC
ReactWorld: Structuring Interaction Responses for Embodied Video Generation
Preprint posted: 2026-11-04 · Original record ↗
- arxiv
- 2611.06120
Explicit source versions (1)
- v1 · 2026-11-04
- Preprint available
Source date: 2026-11-04 · First detected: 2026-11-05 UTC · Historical baseline
Source evidence (2)
- Official evidence ↗ · RSS guid=oai:arXiv.org:2611.06120v1; arxiv:announce_type=new; item pubDate (official metadata)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Embodied video world models predict scene changes caused by robot actions, yet generated object responses may remain inconsistent with task intent and contact dynamics. Critical responses in embodied scenes often concentrate in local regions and specific time intervals. Inspired by chain-of-thought reasoning, we introduce ReactWorld, an embodied video generation model guided by structured interaction anticipation. In the generation backbone's feature space, an interaction response field predictor anticipates how the interaction will unfold, explicitly describing participant roles, future motion response intensity, and local motion changes. A response feedback adapter forms increments to object features from the difference between the object branch's outputs with predicted and zeroed local dynamics inputs. It also converts predicted displacements into positional biases for temporal attention. These two mechanisms guide object feature updates and information retrieval across frames, respectively. Through interaction anticipation and generative feedback, ReactWorld achieves an average score of 77.51 on RBench, outperforming PhysisForcing by 3.29 points with improvements in all nine categories. On EZS-Bench, it achieves a Domain score of 86.98 and an average score of 81.87.
