Trust-Based Shared Control of Human-Vehicle System Using Model Free Adaptive Dynamic Programming

Trust is a critical issue in the field of human-vehicle interaction. In order to improve the efficiency of co-driving system while ensuring the safety and comfort of autonomous vehicles (AVs), a dynamic model of quantitative trust level is proposed and applied to propose a trust-based trajectory tracking control of human-vehicle cooperating system. In this paper, three key contributions have been made: 1) the evolution mechanism of trust in shared steering control is studied and a quantitative trust model is established for the first time by constructing the deviation between actual performance and driver's driving expectations in real-time; 2) an improved model-free adaptive dynamic programming (MF-ADP) method is applied to describe the steering operations of AVs which can adjust its control policy on the basis of evaluated trust value through an adaptive weighing matrix in the cost function; 3) a novel trust-based authority allocation policy is proposed in this work to dynamically assign control proportion between the driver and the autonomous driving system based on real-time trust level, aiming to enhance human-vehicle collaboration. High-fidelity Carsim-Simulink co-simulations are conducted under various velocity conditions to validate the rationality and effectiveness of proposed trust model and shared steering control policy.