Towards Resilient Tracking in Autonomous Vehicles: A Distributionally Robust Input and State Estimation Approach

TL;DR

The paper addresses reliable input and state estimation for autonomous vehicles under model uncertainties and measurement outliers by introducing Distributionally Robust Input and State Estimation (DRISE). It leverages moment-based ambiguity sets for both state and measurement-noise distributions and derives a robust recursive estimator that also estimates unknown inputs, reducing to traditional ISE when uncertainties vanish. Key contributions include a novel estimator formula, worst-case error considerations, and a validation in the CARLA simulator showing improved state and unknown-input estimation and trajectory tracking compared to Kalman filter, DRE, and ISE. The work offers practical robustness guarantees and a computationally tractable approach for autonomous driving in uncertain and adversarial environments.

Abstract

This paper proposes a novel framework for the distributionally robust input and state estimation (DRISE) for autonomous vehicles operating under model uncertainties and measurement outliers. The proposed framework improves the input and state estimation (ISE) approach by integrating distributional robustness, enhancing the estimator's resilience and robustness to adversarial inputs and unmodeled dynamics. Moment-based ambiguity sets capture probabilistic uncertainties in both system dynamics and measurement noise, offering analytical tractability and efficiently handling uncertainties in mean and covariance. In particular, the proposed framework minimizes the worst-case estimation error, ensuring robustness against deviations from nominal distributions. The effectiveness of the proposed approach is validated through simulations conducted in the CARLA autonomous driving simulator, demonstrating improved performance in state estimation accuracy and robustness in dynamic and uncertain environments.

Figures (5)

• Figure 1: Distributionally Robust Input and State Estimation (DRISE) framework diagram. The estimation process follows the steps outlined in Algorithm \ref{['alg_DRISE']}.
• Figure 2: Comparison of state estimation errors.
• Figure 3: Trajectory tracking performance in X-Y plane.
• Figure 4: Comparison of unknown input estimation errors of DRISE and ISE.
• Figure 5: Trace of the unknown input estimation covariances.