Fast Trajectory Planning and Robust Trajectory Tracking for Pedestrian Avoidance

Abstract

This paper presents an integrated design method for pedestrian avoidance by considering the interaction between trajectory planning and trajectory tracking. This method aims to reduce the need for control calibration by properly considering plant uncertainties and tire force limits at the design stage. Two phases of pedestrian avoidance-trajectory planning and trajectory tracking-are designed in an integrated manner. The available tire force is distributed to the feedforward part, which is used to generate the nominal trajectory in trajectory planning phase, and to the feedback part, which is used for trajectory tracking. The trajectory planning problem is solved not by searching through a continuous spectrum of steering/braking actions, but by examining a limited set of “motion primitives,” or motion templates that can be adopted in sequence to avoid the pedestrian. An emergency rapid random tree (RRT) methodology is proposed to quickly identify a feasible solution. Subsequently, in order to guarantee accuracy and provide safety margin in trajectory tracking with presence of model uncertainties and exogenous disturbance, a simplified LQR-based funnel algorithm is proposed. Simulation results provide insight into how pedestrian collisions can be avoided under given initial vehicle and pedestrian states.

Keywords

• Vehicle safety
• motion planning
• motion control
• rapid random tree
• LQR funnel