Layered Parking Control of Self-driving Buses based on Active Disturbance Rejection and Nonliner Model Predictive Control

As a convenient means of transportation, self-driving bus can effectively improve the traffic efficiency and safety, for which arriving and stopping smoothly and precisely at the platform affect the traffic efficiency and the user experiences. In this paper, a layered framework of self-driving bus is proposed, consisting of parking speed planning and longitudinal speed control, based on a combination of nonlinear model predictive control(NMPC) and active disturbance rejection control(ADRC). First, the target speed is planned by NMPC in real-time by minimizing the weighted sum of distance error, ride comfort and travel time. Then, the tracking of the target speed is realized by ADRC, where the extended state observer (ESO) is used to observe the uncertainties in the vehicle longitudinal dynamic, road-tire friction, etc. For improved robustness against the vehicle weight variations, the recursive least square(RLS) algorithm is deployed for online vehicle weight estimation, based on the total disturbance observed by ESO. Experimental results show that using the proposed solution the stopping distance error is within 0.03m, the average speed tracking error is less than 0.04km/h, the error of vehicle weight estimation is about 4.8% with the initial vehicle speed ranging from 10km/h to 25 km/h.