MOTION PLANNING AND COLLISION AVOIDANCE IN AUTONOMOUS VEHICLES USING REACHSET MPC

Abstract

Autonomous vehicles are the future of technology, and there has been a significant amount of research in the field. Many car manufacturers have commercially provided various semi-autonomous features in their vehicles. They are progressively working to create a fully autonomous vehicle capable of operating in a real-time environment. Automated lane keeping and collision avoidance is an inescapable requirement for this goal. Keeping this in mind, various techniques to achieve collision avoidance considering stationary obstacles were explored, and their merits and demerits were considered. Model Predictive Control (MPC) employing Reachability Analysis was the ideal candidate for the scenario under consideration due to its effortlessness of implementation and its ability to consider constraints and adhere to a safety guarantee in an atmosphere riddled with disturbances and measurement noise. Since safety is the primary consideration while designing collision avoidance mechanisms amongst various MPC configurations, it was observed that the drawbacks of conservatism and not catering for the computation time could also be resolved by combining reachability analysis with MPC. Post-implementation of the mentioned approach on a kinematic bicycle model of a car, trajectories are plotted in MATLAB, and results are analyzed to bring out the same.