Robust Path Planning for Space Exploration Rovers

Motion planning is an active area of research, primarily due to itsnecessity in many robotics applications. Research can be focused on improving the efficiency of existing algorithms, expand motion planning capabilities to new robotic domains, tailoring motion planning to a common domain, etc. In the context of rover navigation, motion planning was an afterthought to the science mission, or is seen as a secondary constraint. For this reason, the common motion planning solutions are based on search-based methods, i.e. A* variants. This inherently limits the rover's capabilities instead of leveraging them. A class of motion planning methods that leverage the robot's inherent capabilities and dynamics are sampling-based methods. Methods such as RRT can be applied to robotic systems that have kinematic constraints or significant dynamics, in contrast to the searchbased methods that are commonly used for rovers. The drawback of using a samplingbased method previously was the path quality the motion planner returns was provably suboptimal. One of the first contributions from this fellowship was a sampling-based planner that can provide path quality guarantees for a dynamical rover model. After this initial work, the motion planning method has been adapted to improve its efficiency with heuristics and preprocessing. While some of this is commonly applied domain knowledge, one of the novel inclusions in this work involves the use of small local maneuvers and mechanisms for incorporating them seamlessly in sampling-based planners. This ends up creating a novel hybrid method that aims to get the best parts of a search method and the best parts of sampling-based methods. One of the surprising research directions from the fellowship is the advent of the tensegrity based rover prototypes at Ames Research Center. As these prototypes were being developed, the motion planning methods I was developing were maturing, and enabled motion planning for these complex and dynamic rover reimagining. The commonly used motion planning methods may have been applicable to tensegrity rover planning, but the method I developed has a more natural integration for this rover.