PhysisForcing: Physics Reinforced World Simulator for Robotic Manipulation

2026-06-26

AI Summary

Existing video generation models often produce physically implausible motions—such as discontinuous trajectories or inconsistent object interactions—when applied to robotic simulation, undermining their reliability as world simulators. This work proposes PhysisForcing, a framework that enforces multi-level physical constraints without altering the backbone generative model. It jointly applies pixel-level trajectory alignment loss and semantic-level relational alignment loss, leveraging DiT feature supervision, reference point trajectory alignment, and a frozen video understanding encoder to extract spatial relationships. Evaluated on benchmarks including R-Bench, the method substantially outperforms strong baselines, improving closed-loop task success rates from 16.0% to 24.0% and enhancing downstream policy performance.

Abstract

Video generation models have emerged as a promising paradigm for embodied world simulation. However, both general-domain video generators and robot-specific data fine-tuned models can still produce physically implausible manipulations, including discontinuous motion trajectories and inconsistent robot-object interactions, which limits their reliability as world simulators. Through extensive experiments, we find that such physical instability mainly arises from two factors: deformation of moving objects and implausible spatio-temporal correlations among interacting entities, particularly during contact. Building on this observation, we propose PhysisForcing, a scalable training framework that strengthens physical consistency by focusing supervision on physics-informative regions through joint optimization of pixel-level and semantic-level features. The framework consists of a pixel-level trajectory alignment loss, which supervises DiT features using reference point trajectories, and a semantic-level relational alignment loss, which aligns DiT features with inter-region relations extracted from a frozen video understanding encoder. Extensive experiments on R-Bench, PAI-Bench, and EZS-Bench show that PhysisForcing consistently improves embodied video generation over strong baselines, improving the Wan2.2-I2V-A14B and Cosmos3-Nano base models on R-Bench by 22.3\% and 9.2\% (7.1\% and 3.7\% over vanilla finetuning), with the Cosmos3-Nano variant attaining the best overall score. Beyond generation, as a world model under the WorldArena action-planner protocol it raises the closed-loop success rate from 16.0\% to 24.0\% and further improves downstream policy success, indicating that physically aligned video models yield stronger representations for robotic manipulation.

Problem

• physical plausibility
• robotic manipulation
• video generation
• embodied simulation
• spatio-temporal consistency

Innovation

• PhysisForcing
• physical consistency
• video generation
• robotic manipulation
• trajectory alignment