DiRecT: Safe Diffusion-Based Planning via Receding-Horizon Denoising

2026-06-13

AI Summary

This work addresses the challenge of ensuring trajectory safety in safety-critical planning tasks using diffusion models, where existing methods often compromise sample quality by imposing constraints during intermediate denoising steps. The authors propose DiRecT, a training-free algorithm that, for the first time, integrates terminal constraints with receding-horizon optimization inspired by model predictive control. By constraining only the final trajectory rather than intermediate denoising states, DiRecT decouples the denoising process from constraint satisfaction, enabling plug-and-play compatibility with off-the-shelf optimizers, environmental dynamics priors, and soft reward objectives. This design significantly enhances both flexibility and safety. Experimental results demonstrate that DiRecT outperforms current diffusion-based planning approaches across multiple safety-critical benchmarks, achieving higher task-completion performance while guaranteeing safe deployment.

Abstract

Diffusion models have emerged as powerful tools for planning and control by learning multimodal distributions over actions and trajectories. Yet reliable inference-time safety enforcement remains a key barrier to their deployment in safety-critical tasks. Existing approaches typically project each denoising iterate onto the feasible set, even though constraints are defined only on the final clean trajectory. Enforcing feasibility on noisy intermediate samples can therefore overconstrain the sampling dynamics, substantially degrading sample quality. To address this limitation, we introduce DiRecT (Diffusion-based planning via Receding-horizon denoising with Terminal constraints), a training-free algorithm for constrained sampling from diffusion models via stochastic optimal control (SOC). DiRecT enforces constraints only on the final clean sample, avoiding unnecessary restrictions on the intermediate denoising dynamics. Inspired by model predictive control, we derive a principled receding-horizon surrogate for the otherwise intractable constrained SOC formulation, yielding an efficient algorithm that cleanly separates stochastic denoising from constraint satisfaction, progressively steering samples toward feasible final trajectories without distorting the learned diffusion dynamics. Furthermore, DiRecT is highly flexible: it can leverage off-the-shelf or domain-specific optimizers, incorporate priors over environment dynamics, and optimize additional soft rewards. Extensive experiments on safe planning benchmarks demonstrate that DiRecT substantially improves deployment safety and task performance over existing diffusion-based planning baselines.

Problem

• diffusion models
• safe planning
• constraint enforcement
• trajectory generation
• inference-time safety

Innovation

• diffusion models
• constrained planning
• receding-horizon control
• stochastic optimal control
• safety-critical systems