RI: Small: Exploiting Symmetries of Decision-Theoretic Planning for Autonomous Vehicles

点击翻译按钮获取中文摘要

The project aims at making breakthroughs in AI planning theories that directly affect the accuracy and efficiency of autonomous robots' basic motion and coordination. These advances will play an important role for enhancing the systems' resilience to extreme disturbances in unstructured environments. Specifically, an important focus of this research is to gain a deeper understanding of the interrelationship between model "symmetries" and various performance characteristics in decision-theoretic planning. Traditionally, study of model symmetries belongs to the field of group theory. This research is based on the intuition that autonomous systems acting in real-world environments follow physics laws, and many systems studied in physics show some form of symmetry. Model symmetries in planning, decision-making, and multi-agent coordination domains can be immensely beneficial for designing efficient solving mechanisms to tackle problems with large computational and communication complexities. The results of this project will enhance the autonomy and intelligence of many mobile robot platforms that perform challenging outdoor missions such as environmental monitoring and surveillance, search and rescue, and other applications of societal importance. This research re-thinks the modeling methods of decision-theoretic planning by examining an important class of group-theoretic mechanisms and the possibilities for utilizing them to boost existing solutions to many planning, decision making and coordination problems. Identifying the symmetric properties during the modeling process will help reduce search space and minimize the redundancy that often leads to drastic inefficiencies in computation and communication. By carefully exploiting the symmetric structures of not only the continuous vehicle motion transformation, but also the discrete vehicle state transition, principled frameworks that synthesize various elegant symmetries will be designed, and new models, theoretical analyses, as well as implementation algorithms will be developed. The project will seek improvements related to algorithm convergence, solution quality/accuracy, uncertainty tractability, and system distributability (for multi-robot systems).This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.