Extended State Nonlinear Model Predictive Stopping Control for Autonomous Buses with Braking System Characteristics Self-learning

Stopping control, which involves arriving and stopping smoothly and precisely at the platform, is a crucial performance measure for autonomous buses. However, the effectiveness of stopping control is significantly impacted by variations in braking system characteristics resulting from vehicle weight changes, over-heating, and aging. In this paper, we propose an extended state nonlinear model predictive stopping controller (ES-NMPPC) with self-learning braking system characteristics for autonomous buses. Our approach begins with developing a simplified first-principal model for the bus speed and location, which includes a sub-model for the controlled-by-wired braking system (input: target deceleration, output: braking torque). The proposed model's discrepancy from the real bus is referred to as an extended state named “total disturbance,” which is observed by an extended state observer (ESO). A nonlinear predictive controller (NMPC) is designed based on the extended state model to control the relative distance of the bus from the platform. To reduce the observation burden of ESO, we introduce a recursive least square (RLS) algorithm to estimate the braking system model parameters used in NMPC. Our simulation results on a highly-fidelity model demonstrate that the stopping distance error is within 0.03 m under nominal conditions and increases up to 1.2 m with 40% variation in bus weight when ESO and RLS are not enabled. By enabling ESO, the stopping distance error can be reduced to 0.6 m, which can be further reduced to 0.05 m when both ESO and RLS are enabled.