Improved Potential Field-Based Collision Avoidance Control for Autonomous Vehicles

Limiting factors for autonomous vehicle (AV) to be widely used are not only technical, but also psychological. Considering the psychological feelings of drivers during switching manned to unmanned driving status, this paper proposes an algorithm about avoiding collisions combining driver psychological feelings for AVs. The confidence-limit-distance of the driver is experimentally obtained by many real track tests which require the test driver to approach the obstacle as close as possible. The confidence-limit-distance from driver is defined as the distance between the obstacle and the last steering point allowed for the psychological limit of the driver to avoid collisions. Based on Artificial Potential Field (APF), a road potential field is accordingly established to characterize the characteristics and boundary constraints of the real road. To express the different influences of relative speed and direction on the driver's psychological feeling, the confidence potential field is established based on a two-dimensional normal distribution combining von Mises distribution. The second-order Taylor expansions of road potential field and confidence potential field are firstly introduced into the cost functions for model predictive control (MPC), therefore traditional hierarchical structure of ‘path planning + path tracking’ is transformed into unified quadratic programming problem. The corresponding MPC algorithm used here selects front wheel angle as the control variable to be solved. The confidence-limit-distance and the range of vehicle states variable sensed in track tests are taken as the constraints of the MPC. Finally, co-simulations and Hardware-in-the-Loop (HIL) tests are carried out, showing the effectiveness of designed algorithm, which can be useful in the development and design for Advanced Driving Assistant System (ADAS) and AVs.