Effects of Dynamic Modeling on the Path Planning of Simulated Autonomous Vehicles

Abstract: The purpose of the thesis was to evaluate the robustness of the Hybrid A' algorithm in regards to the levels of dynamic vehicle modeling, vehicle parameters and driving velocity. The original scope of the thesis project was to alter vehicle and tire models but this was discovered to be out of the time limits of the thesis project. The new scope was to alter the parameters with close connection to the vehicle dynamics, such as tire friction, vehicle mass and driving velocity. The thesis project was performed in collaboration with Syntronic and their internal project Autodrive.  Due to the complexity of the system, tests were first performed to evaluate if the path planner and path following were deterministic. Variation of tire friction, vehicle mass, reference velocity and maximum steering was then performed during three different driving missions designed to simulate different complexity levels of realistic driving scenarios. The dependence of the varied vehicle parameters on the deterministic behaviour was also examined. Lastly, the impact of altering parameters used for smoothing of the Hybrid A' solution and parameters of the controller were explored and analysed. The position of the vehicle, the velocity profile, slip angles and angular acceleration were used in order to draw conclusions regarding the performance of Hybrid A', the motion control and the limitations of vehicle and tire models.  The results showed that the implementation used of Hybrid A' is deterministic but fail to produce a consistent route when faced with obstacles and re-planning due to the vehicle control not being deterministic. Variation in vehicle mass and reference velocity as well as the aggressiveness of the course were concluded to impact the determinism of the path following. The best values for the smoothing parameters are dependent on the desired vehicle behaviour, but can be used to reduce the difference between the planned and followed path.  The Hybrid A' planner was concluded to be robust in regards to variations in vehicle mass, tire friction and maximum steering angles but very sensitive to a change in velocity. The slip angles were generally small, but the kinematic single-track model theoretically only hold when the slip angles are zero. The results did, however, indicate that the implementation of the Hybrid A' algorithm was able to produce a drivable and realistic path for all cases except the most extreme parameter choices. For larger velocities and unrealistically low tire friction, the dynamic single-track model is more suitable.  The choice of vehicle model is not critical except when using high driving velocities. It is important to have knowledge of the desired driving velocity when choosing a dynamic model, but the solution of the implemented path planner and path following is otherwise robust enough to handle a less advanced model. The current implementation of the Hybrid A' algorithm, using the kinematic single-track model, can therefore be concluded to be sufficient in planning a realistic path for a simulated autonomous vehicle for low driving velocities, while a more advanced model is needed when increasing the driving velocity.