Constraint-Aware Diffusion Models for Trajectory Optimization

TL;DR

This paper introduces a constraint-aware diffusion model for trajectory optimization that reduces constraint violations in generated solutions, demonstrated on manipulation and reach-avoid tasks.

Contribution

The paper proposes a hybrid loss function for diffusion models that minimizes constraint violations while maintaining data distribution fidelity.

Findings

Outperforms traditional diffusion models in constraint violation minimization

Effective on tabletop manipulation and two-car reach-avoid problems

Generates solutions close to local optima

Abstract

The diffusion model has shown success in generating high-quality and diverse solutions to trajectory optimization problems. However, diffusion models with neural networks inevitably make prediction errors, which leads to constraint violations such as unmet goals or collisions. This paper presents a novel constraint-aware diffusion model for trajectory optimization. We introduce a novel hybrid loss function for training that minimizes the constraint violation of diffusion samples compared to the groundtruth while recovering the original data distribution. Our model is demonstrated on tabletop manipulation and two-car reach-avoid problems, outperforming traditional diffusion models in minimizing constraint violations while generating samples close to locally optimal solutions.