Evaluation of RRT* for Quasi Optimal Paths for an Obstacle Field Navigation Benchmark

Abstract

An optimal sampling based algorithm in motion planning called the RRT* is evaluated and tested for producing time optimal trajectories. It is chosen as a potential candidate that can be used in benchmarking. This choice was made because it was expected that the RRT* would be a faster and simpler alterantive to the cell decomposition methods like the motion primitive automaton (MPA) that is used in benchmarking. Our evaluation of the RRT* took three phases. First phase, the RRT* was implemented without dynamic constraints taken into consideration and then it was tested for converging to optimal paths. Second phase, the RRT* was extended to deal with simple dynamic systems like the 3D Dubins vehicle. Third phase, the RRT* was further extended to deal with complex dynamic systems like the ARTIS helicopter. The main contribution of this thesis is that it aided the RRT* algorithm to become a potential alternative for benchmarking, but still several improvements are needed to be an alternative to the MPA method. Also it contributed in trying to combine between motion planning and control, in order to make a more synergy between the system reachable states and motion planning.