Virtual Validation and Uncertainty Quantification of an Adaptive Model Predictive Controller-Based Motion Planner for Autonomous Driving Systems

Abstract

In the context of increasing research on algorithms for different modules of the autonomous driving stack, the development and evaluation of these algorithms for deployment onboard vehicles is the next critical step. In the development and verification phases, simulations play a pivotal role in achieving this aim. The uncertainty quantification of Autonomous Vehicle (AV) systems could be used to enhance safety assurance and define the error-handling capabilities of autonomous driving systems (ADSs). In this paper, a virtual validation methodology for the control module of an autonomous driving stack is proposed. The methodology is applied to a rule-defined Model Predictive Controller (MPC)-based motion planner, where uncertainty quantification (UQ) is performed across various scenarios, based on the intended functionality within the algorithm’s operational design domain (ODD). The framework is designed to assess the performance of the algorithm under localization uncertainties, while performing obstacle vehicle-overtaking, vehicle-following, and safe-stopping maneuvers.

Keywords

• autonomous vehicle
• model predictive control
• optimal control problem
• control barrier functions
• obstacle avoidance
• uncertainty analysis