Robust Model Predictive Control Design for Autonomous Vehicles with Perception-based Observers

2025-09-05

AI Summary

Deep learning–based perception modules in autonomous driving introduce non-Gaussian, biased, and heavy-tailed noise, leading to inaccurate uncertainty quantification, unbounded estimation errors, and unstable feedback control. Method: This paper proposes a robust set-valued model predictive control (MPC) framework that abandons the conventional zero-mean Gaussian assumption. It characterizes perception uncertainty using constrained zonotopes and constructs a maximal robustly invariant terminal set along with a state-feedback gain via the Minkowski–Lyapunov inequality to ensure closed-loop stability. To improve online computational efficiency, it incorporates slack variables and ellipsoidal approximations. Contribution/Results: Evaluated in simulation and on a ROS2-based omnidirectional mobile robot platform, the method significantly reduces state estimation error, enhances trajectory tracking accuracy, and improves control safety compared to Gaussian-assumption approaches—thereby providing verifiable robustness guarantees for learning-enabled perception–control co-design.

Technology Category

• Intelligent Robots: State Estimation
• Machine Learning: Calibration & Uncertainty Quantification
• Natural Language Processing: Safety and Robustness

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Security and Privacy: Large-scale security measurements

Abstract

This paper presents a robust model predictive control (MPC) framework that explicitly addresses the non-Gaussian noise inherent in deep learning-based perception modules used for state estimation. Recognizing that accurate uncertainty quantification of the perception module is essential for safe feedback control, our approach departs from the conventional assumption of zero-mean noise quantification of the perception error. Instead, it employs set-based state estimation with constrained zonotopes to capture biased, heavy-tailed uncertainties while maintaining bounded estimation errors. To improve computational efficiency, the robust MPC is reformulated as a linear program (LP), using a Minkowski-Lyapunov-based cost function with an added slack variable to prevent degenerate solutions. Closed-loop stability is ensured through Minkowski-Lyapunov inequalities and contractive zonotopic invariant sets. The largest stabilizing terminal set and its corresponding feedback gain are then derived via an ellipsoidal approximation of the zonotopes. The proposed framework is validated through both simulations and hardware experiments on an omnidirectional mobile robot along with a camera and a convolutional neural network-based perception module implemented within a ROS2 framework. The results demonstrate that the perception-aware MPC provides stable and accurate control performance under heavy-tailed noise conditions, significantly outperforming traditional Gaussian-noise-based designs in terms of both state estimation error bounding and overall control performance.

Problem

• Addresses non-Gaussian noise in perception-based state estimation
• Ensures robust control stability under biased heavy-tailed uncertainties
• Improves computational efficiency via linear programming reformulation

Innovation

• Set-based state estimation with constrained zonotopes
• Reformulated robust MPC as linear program
• Minkowski-Lyapunov cost with slack variables