Decision Making for Automated Vehicles in Merging Situations - using Partially Observable Markov Decision Processes
In this thesis a decision making algorithm for automated cars in lane change situations is presented. The algorithm accounts for situations when all objects detected by the sensor system cannot be classified. The algorithm is partially formulated as a Partially Observable Markov Decision Process which is solved approximately with means of Point Based Value Iteration. The algorithm is implemented in a simulation environment and then tested and analyzed with simulation data. The results show that the decision algorithm is able to choose a proper gap to merge into and it is able to cancel a maneuver if the traffic situation changes so that a merging operation cannot be completed. If the execution time of the implemented Point Based Value Iteration algorithm could be decreased this would further improve the real world applicability of the decision making algorithm.