Grounding Generative Policies in Physics: Optimization-Guided Diffusion for Robot Control

2026-06-23

AI Summary

This work addresses the frequent failure of zero-shot deployment in generative policies due to violations of physical constraints—such as reachability, collision avoidance, and closed-loop executability—which often result in infeasible actions. To bridge the gap between generated behaviors and real-world execution, the authors propose an optimization-guided mechanism that replaces the standard denoising perturbations in diffusion model inference with constraint-satisfying optimization corrections. This approach enables hard or soft enforcement of physical constraints during sampling without requiring model retraining. Notably, it formulates diffusion guidance as a constrained optimization problem for the first time. Evaluated on dexterous grasping and visuomotor manipulation tasks, the method achieves up to 20% and 23% higher task success rates, respectively, significantly outperforming baseline methods in both action feasibility and quality.

Abstract

Diffusion models sample effectively from high-dimensional, multimodal distributions, but their outputs may violate deployment constraints. For task-space robot policies, generated grasps, waypoints, or trajectories can be distributionally valid yet infeasible, violating reachability, collision-avoidance, or closed-loop executability requirements. This embodiment gap limits zero-shot deployment across robots, even when the task-space behavior itself is transferable. We propose an inference-time optimization framework that couples the behavior generation to physical feasibility by formulating diffusion guidance as a constrained optimization problem. Our key insight is to replace the sampling perturbation in the backward process with an optimized correction, allowing hard constraints or soft penalties to be imposed during sampling without the need to retrain the diffusion model, while keeping samples close to the learned prior. We evaluate the method on dexterous grasp synthesis with reachability and collision-avoidance constraints, and dynamic manipulation with controller-level trackability constraints. Across settings and robot embodiments, optimization-guided denoising matches the feasibility of projection- and gradient-guidance baselines while better preserving grasp quality, and improving controller-level executability and task success, with task success improving by up to 20pp. on dexterous grasping and 23pp. on visuomotor manipulation over the best baseline.

Problem

• diffusion models
• robot control
• physical feasibility
• task-space policies
• embodiment gap

Innovation

• optimization-guided diffusion
• constrained sampling
• robot policy grounding
• embodiment-aware generation
• feasibility-preserving denoising