Safe Model Predictive Diffusion with Shielding
Taekyung Kim, Keyvan Majd, Hideki Okamoto, Bardh Hoxha, Dimitra Panagou, Georgios Fainekos
AI summary
Problem
Diffusion-based trajectory planners suffer from severe sampling inefficiency and lack formal safety guarantees when applied to constrained kinodynamic systems, often relying on computationally expensive or infeasible post-processing corrections.
Approach
Safe MPD inserts a shielded rollout mechanism into every denoising step, transforming potentially unsafe candidate trajectories into provably safe and feasible ones before scoring and averaging them.
Key results
- Guarantees kinodynamic feasibility and safety by construction during denoising
- Dramatically improves sample efficiency by eliminating zero-weight constraint violations
- Achieves sub-second planning times via parallelized GPU implementation
- Outperforms existing safety strategies in success rate and safety on non-convex tractor-trailer planning tasks
Why it matters
Enables reliable, real-time trajectory planning for safety-critical robotic systems operating in complex, constrained environments.
Abstract
Generating safe, kinodynamically feasible, and optimal trajectories for complex robotic systems is a central challenge in robotics. This paper presents Safe Model Predictive Diffusion (Safe MPD), a training-free diffusion planner that unifies a model-based diffusion framework with a safety shield to generate trajectories that are both kinodynamically feasible and safe by construction. By enforcing feasibility and safety on all samples during the denoising process, our method avoids the common pitfalls of post-processing corrections, such as computational intractability and loss of feasibility. We validate our approach on challenging non-convex planning problems, including kinematic and acceleration-controlled tractor-trailer systems. The results show that it substantially outperforms existing safety strategies in success rate and safety, while achieving sub-second computation times. [Project Page]1 [Code] [Video]