Reinforcement learning in path planning and obstacle avoidance for autonomous vehicles

Abstract

Path planning and trajectory planning is an important aspect of navigation in the field of robotics and automation. It involves studying the environment space, evaluating the obstacle positions or the potential areas of danger, computing the cost and then eventually planning a route from one point to another point. During the planning of routes, the cost is aimed to be kept minimal in terms of saving time, avoiding obstacles and fewer casualties. Most literature reviews and experiments that used this approach have applied these to mobile robots so as to measure the accuracy, reliability and efficiency. This has shown great progress but with enormous research, there is another potential problem that arises. The uncertainty that lies in a real-time environment due to changes in the map, the addition of objects and changes in the orientations results in the inaccuracy of the routes planned. This aspect can be addressed through the application of reinforcement learning techniques that allows the robots to learn by itself. Therefore, the objective of this project is to test path planning algorithms and implement reinforcement learning in a simulated environment.