Research Project Title:
Short Range Recovery Planning for Autonomous Vehicles in Offroad Environments
abstract:Autonomous robotic vehicles navigating in unknown harsh environments are susceptible to a variety of hazards that may exist. When operating in offroad conditions such as desert landscapes or steep rolling hills, obstacles become difficult to detect even with state-of-the art perception systems. This causes robust autonomy stacks to fail and freeze when lethal objects or features suddenly present themselves in front of the vehicle. When analyzing human offroad drivers, we observe the same phenomenon with them reacting in the following way: they detect the obstacle, stop, then reverse to a safe area that will allow them to continue driving. This project is focused on implementing this decision structure on an offroad autonomous vehicle as part of the DARPA RACER program (Robotic Autonomy in Complex Environments with Resiliency) for real time execution with minimal intervention. Through developing a novel perception and uncertainty aware CVaR risk metric to define traversability and by implementing a bi-directional Dubins based recovery planner with efficient run times, a coherent recovery stack is ultimately created.
After working with this group since Spring 2022 and through conducting research with the project over the summer as a JPL intern, I have already seen myself become an integrated part of the team and have been able to learn much about the graduate school application process. I've been able to explore all the various niches that exist within robotics research but also have developed a great idea of how to develop large robotic systems as a whole. By working with my grad student and ultimately communicating with my PI Sertac Karaman and JPL mentor Patrick Spieler, I have become extremely confident of my decision to pursue a PhD and very much credit the SuperUROP experience for giving me the space to explore that.