Reinforcement Learning for Human-Machine Cooperative Control: Enhancing Stability and Antirollover with Commercial Vehicle

Safety issues in commercial vehicles have consistently been a focal point of research, with yaw stability control and anti-rollover control being particularly critical. To enhance the yaw stability and anti-rollover capabilities of commercial vehicles operated by drivers with varying driving styles, this paper proposes a human-machine cooperative control strategy based on the Deep Deterministic Policy Gradient (DDPG) algorithm within the framework of active front steering (AFS) and differential braking control (DBC) systems. Initially, a model predictive control (MPC) algorithm is utilized to develop an integrated controller for AFS and DBC, aimed at managing yaw stability and preventing rollover. Subsequently, to assess the driving styles of different drivers, driving data from multiple participants were collected and categorized into radical, general, and cautious types based on selected characteristic parameters. The DDPG algorithm is then employed to distribute weights between the driver input and the controller for each of the three driver categories, integrating the constraints of the phase space stability boundaries of sideslip angle, yaw rate, and lateral load transfer rate into the reward function design. Finally, the effectiveness of the proposed cooperative control strategy is confirmed using Matlab/TruckSim co-simulation platform.

I. Introduction

In recent times, driven by the swift progress of Advanced Driver Assistance Systems (ADAS), there has been a notable trend towards the increased intelligence of vehicles [1]. Compared to passenger vehicles, commercial vehicles, which are designed and engineered for the transportation of people and goods, exhibit distinct characteristics such as a higher center of gravity, greater sprung mass, and slower system response [2]. These characteristics of commercial vehicles pose additional challenges for ADAS in enhancing stability and anti-rollover capability. Within human-vehicle closed-loop system, driver's manipulation is an indispensable part, as drivers with different driving styles exhibit significant variations in handling under identical road conditions[3]. Analyzing the driving styles of drivers, along with the characteristics of commercial vehicles, is essential to laying the foundation for improving the yaw stability and anti-rollover capability of commercial vehicles.