Localization over the Dual Quaternions
Abstract: Demands placed on autonomous systems are growing. Delivery robots, self-driving cars, automated warehouses, and more are all expected to execute their functions quickly and safely. In these and other applications, robots' performance depends in part on precise knowledge of where they are. In other words, robots must quickly and accurately solve various localization problems to execute their other tasks. In this talk, I will cover my group's recent developments in enabling improved localization for autonomous mobile robots. This talk formulates localization problems over the manifold of unit dual quaternions, which enables the development of richer, more accurate models of uncertainty than in prior work. Then it develops a manifold optimization algorithm that allows for optimal localization over the unit dual quaternions. This algorithm is provably convergent, and simulations on standard benchmarks show a substantial improvement in accuracy over the state of the art.
Biography: Matthew Hale is an Associate Professor of Electrical and Computer Engineering at Georgia Tech. His research is broadly on new approaches to autonomy, with an emphasis on modeling how information is computed and shared in multi-agent systems. Recent research interests have included localization problems in robotics, privacy and deception in control, and compositional models of multi-agent coordination. He has received the NSF CAREER, ONR YIP, and AFOSR YIP awards.