Constrained Diffusers for Safe Planning and Control

2025-06-14

AI Summary

To address the challenge of enforcing safety constraints in planning and control tasks using diffusion models—without requiring retraining or architectural modifications—this paper proposes a general-purpose constraint-embedding framework. The method systematically integrates constrained optimization principles into the reverse diffusion process for the first time, synergistically combining projection-based methods, primal-dual optimization, and the augmented Lagrangian method, augmented with discrete control barrier functions to ensure online safety guarantees. Leveraging constrained Langevin sampling and zero-shot fine-tuning, the approach achieves near 100% constraint satisfaction across Maze2D navigation, biomimetic locomotion, and PyBullet sphere-rolling benchmarks. It reduces computational overhead by over 35% while matching or surpassing the performance of state-of-the-art safe planning methods.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Constraint Satisfaction and Optimization: Constraint Optimization
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Diffusion models have shown remarkable potential in planning and control tasks due to their ability to represent multimodal distributions over actions and trajectories. However, ensuring safety under constraints remains a critical challenge for diffusion models. This paper proposes Constrained Diffusers, a novel framework that incorporates constraints into pre-trained diffusion models without retraining or architectural modifications. Inspired by constrained optimization, we apply a constrained Langevin sampling mechanism for the reverse diffusion process that jointly optimizes the trajectory and realizes constraint satisfaction through three iterative algorithms: projected method, primal-dual method and augmented Lagrangian approaches. In addition, we incorporate discrete control barrier functions as constraints for constrained diffusers to guarantee safety in online implementation. Experiments in Maze2D, locomotion, and pybullet ball running tasks demonstrate that our proposed methods achieve constraint satisfaction with less computation time, and are competitive to existing methods in environments with static and time-varying constraints.

Problem

• Ensuring safety under constraints for diffusion models
• Incorporating constraints into pre-trained diffusion models
• Guaranteeing safety in online implementation using barrier functions

Innovation

• Constrained Langevin sampling for reverse diffusion
• Discrete control barrier functions for safety
• Three iterative optimization algorithms for constraints