Find Optimal Suspension Kinematics Targets for Vehicle Dynamics Using Reinforcement Learning

Setting up suspension kinematics targets has been a challenging task for vehicle engineers. The challenges involve a high-dimensional search space, nonlinear relationships between the suspension kinematics and vehicle dynamics, exploration and exploitation trade-offs, and the need for domain-specific knowledge. Traditional multi-objective optimization methods are time-consuming, sensitive to initial conditions, and rarely converge to the global optimum in high-dimensional spaces. This article explores how reinforcement learning can be used to automate the design of suspension kinematics targets, addressing a longstanding challenge in vehicle dynamics design: the inverse problem of satisfying high-level handling objectives through low-level subsystem parameters. The method is based on the accumulation of knowledge through the interaction between an intelligent agent and a simulation environment. The agent optimizes suspension kinematics targets by receiving rewards tied to vehicle dynamics performance. The agent, employing a Gaussian policy and σ-based sensitivity analysis, enables the identification of critical and non-critical design parameters. The results show that the proposed method can find optimal suspension kinematics targets with the help of accumulated knowledge. The knowledge-guided learning process demonstrates a novel approach to solving high-dimensional optimization problems, offering good convergence time and valuable results. The proposed method contributes to the field by using reinforcement learning to set up suspension kinematics targets in the automotive industry.