Motion Planning and Control – Ackermann Robots : Path Finding and Trajectory Tracking with Nonlinear Dynamics

Sammanfattning

This thesis is about the development of a complete framework for path planning and automatic motion control for an ackermann steered robot. The purpose of the thesis was to let a robot automatically navigate its way through an obstacle course, from a given start position to a set goal, whilst adhering to the robot's physical constraints and dynamics. The work was divided into three main parts: path finding, trajectory planning, and motion control. For path finding, we build an algorithm based on Steven M. LaValle's Rapidly Exploring Trees (RRT), which sequentially generates a tree like structure made up of nodes. The algorithm was chosen based on its reliability and ability to navigate difficult environments without the need to first discretize the world space. Two different methods to track the generated path were implemented: a waypoint method, and a polynomial fit function method. Both methods were tested using the same control system which we also developed. The control system was made by linearizing the robot's ackermann dynamics, which made it possible to use a PID-like control strategy where positional error, error in velocity and error in acceleration could be used to calculate appropriate values for the control signals, jerk and rateof steering angle. The entire framework was simulated using MATLAB, although it was never realized on a real world robot. The results were good. Both methods were able to produce reliable and consistent collision avoidance and satisfactory trajectory tracking. Upon testing it became apparent that the waypoint method was best suited for labyrinth-like obstacle courses, and the polynomial fit function method was best suited for more open environments. Even though the simulation results were good, there is still further development required beforeany real world application can be made, such as testing the framework on a real robot, and implementing more features which take physical constraints into account, such as sensor noise.