Title: Autonomously learning mobility limits using the learning reference governor approach
Abstract: Modern mobility systems have demanding performance requirements that may push systems to their limits. These limits may change with manufacturing variability, degradation, or damage. A common practice is to operate conservatively and avoid violation of these limits in the worst-case scenario. Such conservative operation ensures safety but can unnecessarily limit vehicle performance and mobility, especially in emergency situations. This talk focuses on a novel, emerging approach to address the performance-robustness tradeoff, that relies on the integration of prediction and learning/adaptation. In safety-critical cases, violation of these mobility limits can cause catastrophic consequences. To this end, the learning reference governor (LRG) is developed such that it will initially operate systems conservatively and improve systems’ performance as it learns more about mobility limits and maneuvers that approach these limits. In this talk, LRG will be described along with simulation studies on rollover avoidance of trucks. LRG is effective in learning lateral load transfer limits safely without constraint violation and helping the autonomous steering algorithm drive the vehicle at these limits without rollover.