AI Path Prediction & Planning for Autonomous Vehicles

Abstract

Autonomous navigation of vehicles at various places can tremendously improve daily traversal, especially while driving in unfamiliar territory. Path-planning systems have been continuously improved to cope with increasingly difficult issues and advances in computing technology. One example is the rise of sampling-based planners, which made difficult planning issues easier to solve. As a result, problems that were previously thought difficult can now be addressed, and evendynamical constraints can be overcome. The ongoing modifications and study in path planning are aimed at achieving robotic systems' fullautonomy. By relying solely on path planning approaches, this goal has not been fully realized. The majority of path-planning algorithms work well in simulations and laboratory robots, but real- world robots operate in unknown surroundings and interact with other agents whose motion is often unpredictable and whose conduct can be uncooperative. Real robots have inherent kinematical and dynamical (kinodynamical) constraints that make autonomous collision-free motion impossible in particular circumstances. This research demonstrates the implementation, testing, and comparison of two path-planning algorithms for autonomous navigation of a vehicle/rover, namely rapidly-exploring random trees (RRT) and second probabilistic roadmap (PRM). Using random samples from the search space, an RRT generates a tree rooted at the beginning configuration. As each sample is drawn, a link is made between it and the state closest to it in the tree. This results in the adding of the new state to the tree if the link is viable (goes wholly acrossfree space and obeys all constraints). The main idea underlying PRM, on the other hand, is to pick random samples from the robot's configuration space, test them to see whether they are in the free space, and then use a local planner to connect these configurations to other adjacent configurations.The beginning and goal configurations are brought in, and the resulting graph is subjected to a graph search algorithm in order to find a path between them. To assess the performance of the recommended approaches, simulations and experiments are carried out.