Evaluating Driving Performance of a Novel Behavior Planning Model on Connected Autonomous Vehicles

Many current algorithms and approaches in autonomous driving attempt to solve the trajectory generation or trajectory following” problems: given a target behavior (e.g. stay in the current lane at the speed limit or change lane), what trajectory should the vehicle follow, and what inputs should the driving agent apply to the throttle and brake to achieve this trajectory? In this work, we instead focus on the “behavior planning” problem—specifically, should an autonomous vehicle change lane or keep lane given the current state of the system? In addition, current theory mainly focuses on single-vehicle systems, where vehicles do not communicate with their surroundings. However, other works have showed opportunities for data sharing between vehicles and other vehicles (V2V) or road infrastructure (V2I) to improve driving algorithms. For example, a vehicle’s lane changing decisions on a highway might be improved by using traffic density data from other cars further ahead, beyond the first vehicle’s own field of vision. In this work, we use the CARLA vehicle simulator to implement a simple behavior planning algorithm inspired by previous work from collaborators at UConn. The algorithm uses shared V2V information about position and velocity of nearby neighbors to improve lane changing decisions while still maintaining vehicle safety. We observe the driving performance of the fleet of connected vehicles under various traffic conditions, focusing on speed and passenger driving comfort. The goals of this project are to demonstrate the performance improvements that result from V2V information sharing, highlight safety challenges and solutions during lane-changing, and lay the groundwork to develop other behavior planning algorithms in the future. Our lab plans to use this work to continue future autonomous driving research