The interaction between vehicles or with human agents is a fundamental problem for planning and control. Game theory is the most useful tool for making decisions in such situations. In this project, a motion planning algorithm based on game theory is presented. First, candidate paths are generated for ego car and opponent using a conformal lattice planner, and then the utility of each path is calculated for ego and opponent, and a path is selected using Nash equilibrium. Finally, using two longitudinal and lateral PID controllers, the car is controlled in the selected path. We defined a race between two vehicles in the CARLA simulator, the results of which indicate the proper performance of the motion planner.
Overtaking and blocking maneuver using game theoretic motion planner in CARLA
This paper presents a new algorithm for optimum path selection of autonomous cars in a car race. The algorithm uses a bidirectional LSTM to predict the driving strategy of the opponent car and prescribes the best move based on the notion of Nash equilibrium.