DualGuard MPPI: Safe and Performant Optimal Control by Combining Sampling-Based MPC and Hamilton-Jacobi Reachability
Javier Borquez, Luke Raus, Yusuf Umut Ciftci, Somil Bansal
AI summary
Problem
Classical sampling-based MPC methods like MPPI struggle to enforce hard safety constraints, often relying on penalty functions that fail to guarantee safety or wasting computation on invalid samples. Post-optimization safety filters can also degrade performance and ignore long-term control effects.
Approach
The method integrates Hamilton-Jacobi reachability analysis directly into the MPPI sampling loop to filter control perturbations and generate provably safe trajectories, followed by an output least-restrictive filter to prevent unsafe multimodal combinations before execution.
Key results
- Eliminates safety penalty tuning by guaranteeing all sampled rollouts are safe
- Reduces sampling variance and improves performance optimization
- Prevents unsafe multimodal control combinations via an output least-restrictive filter
- Demonstrates superior safety and performance in simulations and RC car hardware experiments
Why it matters
Enables reliable, high-performance control for safety-critical autonomous systems without compromising safety guarantees or requiring manual penalty tuning.
Abstract
Designing controllers that are both safe and per- formant is inherently challenging. This co-optimization can be formulated as a constrained optimal control problem, where the cost function represents the performance criterion and safety is specified as a constraint. While sampling-based methods, such as Model Predictive Path Integral (MPPI) control, have shown great promise in tackling complex optimal control problems, they often struggle to enforce safety constraints. To address this limitation, we propose DualGuard-MPPI, a novel framework for solving safety- constrained optimal control problems. Our approach integrates Hamilton-Jacobi reachability analysis within the MPPI sampling process to ensure that all generated samples are provably safe for the system. On the one hand, this integration allows DualGuard- MPPI to enforce strict safety constraints; at the same time, it facilitates a more effective exploration of the environment with the same number of samples, reducing the effective sampling variance Received 31 January 2025; accepted 29 April 2025. Date of publication 9 May 2025; date of current version 2 June 2025. This article was recommended for publication by Associate Editor F. Shi and Editor O. Stasse upon evaluation of the reviewers’ comments. This work was supported in part by the University of Santiago de Chile, in part by NSF CAREER Program under Award 2240163 and in part by DARPA ANSR Program. (corresponding author: Javier Borquez.) Javier Borquez and Yusuf Umut Ciftci are with the University of Southern California, Los Angeles, CA 90089 USA (e-mail: javierbo@usc.edu). Luke Raus is with the Olin College of Engineering, Needham, MA 02492 USA. Somil Bansal is with Stanford University, Stanford, CA 94305 USA (e-mail: somil@stanford.edu). Digital Object Identifier 10.1109/LRA.2025.3568686 and leading to better performance optimization. Through several simulations and hardware experiments, we demonstrate that the proposed approach achieves much higher performance compared to existing MPPI methods, without compromising safety.