Reinforcement Learning based Parking Space Egress for Autonomous Driving Scenarios

Automated parking systems for cars have become the need of the hour in all nations. They have gained wide acceptance from customers, and hence OEMs (Original Equipment Manufacturers) are working towards achieving precise/accurate automated parking. Various algorithms are being developed to plan the trajectory of the vehicle to be moved in/out of the desired parking slot. Most of these algorithms consider a static environment and don’t account for dynamic objects in parking-out scenarios. In this paper, we intend to develop an algorithm which considers traffic objects while moving effectively along the planned trajectory by giving the 'right of way' to the traffic object. In this study, we specifically propose a novel approach for generating linear and angular velocity profiles using reinforcement learning (RL) in conjunction with Hybrid A* path planning for autonomous vehicles (AVs) navigating parking maneuvers. The aim is to address challenges faced by traditional Model Predictive Control (MPC) methods in trajectory planning, such as the lack of consideration for the right of way for the ego vehicle. Our proposed RL-based velocity profile generation algorithm employs a reward function that considers safety, path efficiency, and the ego vehicle’s right of way. We compare the performance of our method with traditional MPC approaches in various parking-out scenarios, including parallel, perpendicular, and diagonal parking, by analyzing the generated trajectories and evaluating their adherence to the desired criteria. We employ the MATLAB Autonomous Driving Toolbox to simulate and test our approach in a range of parking scenarios. The results indicate that the RL-based method outperforms traditional MPC in terms of safety and efficiency, while adhering to the right-of-way rules. This novel approach demonstrates the potential benefits of using RL for trajectory planning in autonomous driving applications, specifically in complex parking-out situations.