Game theoretic interactions between pedestrians and autonomous vehicles

One of the key areas that demands robust solutions is the deployment of autonomous vehicles in the real world is self-driving during negotiations for public space between pedestrians and autonomous vehicles. University of Lincoln has recently designed, built, and published an open-source hardware and software autonomous vehicle, OpenPodcar, for testing pedestrian interaction models. This thesis extends this design to OpenPodcar2, upgrading it to a ROS2 stack and lowering cost by replacing lidar detection and mapping with a depth camera. It upgrades the vehicle’s SLAM from a 2D to 3D method, RTAB-Map, and pedestrian detection and tracking to YOLOv8. The upgraded OpenPodcar2 is utilized in a new iteration of Lincoln’s sequential chicken game theory experiments. This approach employs game theory to control the vehicle and evaluate pedestrian behavior in terms of game theoretic parameters. Traditional game theory models suggest that, similar to human drivers, autonomous vehicles can mitigate collision risks by accepting minor yielding delays. Recent research proposes that these collisions could be replaced by frequent, less harmful negative utilities, such as encroaching on pedestrians’ personal space. This thesis presents the first real-world demonstration and evaluation of this approach using an autonomous vehicle and human subjects.