Exploring Model Complexity for Trajectory Planning of Autonomous Vehicles in Critical Driving Scenarios

Trajectory planning is a crucial component of autonomous driving systems. However, using simple vehicle models for trajectory planning may result in unrealistic reference trajectories, especially in critical driving conditions, endangering the safe driving of autonomous vehicles. This study explores the effect of model complexity on the trajectory planning performance of autonomous vehicles in critical driving scenarios. Five trajectory planners of various levels of model complexity, including Planner \({\text {STK}}\) (single-track kinematic model), Planner \({\text {STDL}}\) (single-track dynamic vehicle model with a linear tyre model), Planner \({\text {STD}}\) (single-track dynamic vehicle model with a simplified Pacejka tyre model), Planner \({\text {DTB}}\) (double-track vehicle model with the brush tyre model), and Planner \({\text {DTMlt}}\) (double-track vehicle model with load transfer consideration and the Pacejka tyre model), are designed. The trajectory planners are formulated as optimal control problems, where constraints for obstacle avoidance, yaw stability and the physical limits on vehicle actuators are explicitly considered. These planners are assessed in two severe driving manoeuvres, i.e. the double-lane change and single-lane change manoeuvres. Results indicate that Planner DTMlt outperforms DTB with higher passing velocity as well as smaller peak yaw rate and sideslip angle, and that Planners STD, STDL and STK are not suitable for use in these critical driving scenarios.