Events Calendar

CS Colloq: Geoffrey Hollinger

Talk Title: Efficient, Guaranteed Search with Multiple Robots
Speaker: Geoffrey A. Hollinger
Host: Prof. Gaurav SukhatmeAbstract:
Consider the problem of coordinating a team of robots to locate a target in an environment or to authoritatively say that one does not exist. Such a scenario may occur in urban search and rescue, military operation, and even aged care. The search must be robust (deal with robot failures), decentralized (reduce computational and communication bottlenecks), and reactive (make use of any pertinent information that becomes available during search). Prior methods in the literature would force you to make one of two assumptions in this scenario. Do you make the worst-case assumption and choose to treat the target as adversarial? The robots could then utilize graph search algorithms to guarantee finding the target, but the search might take an unnecessarily long time. Or do you decide to trust some non-adversarial model of the target? The robots could then optimize the search with respect to that model, but this approach would eliminate guarantees if the model is inaccurate. In this case, the target may avoid the robots entirely. However, it is possible to do better; how can we strike a balance between risky average-case search and conservative worst-case search?In our recent work we have developed a method that combines the two search paradigms described above to generate plans that clear an environment of a worst-case adversarial target and have good average-case performance considering a non-adversarial motion model.
Our proposed algorithm takes advantage of spanning tree traversal methods along with receding horizon planning to generate a number of candidate search schedules. The resulting architecture is decentralized, scalable, and yields theoretically bounded average-case performance. We validate our algorithm through a number of experiments in simulation and on a team of robot and human searchers in an office building. In addition, I will discuss ongoing work on incorporating communication and connectivity constraints into the search schedules.Bio:
Geoffrey Hollinger is a Ph.D. Candidate at Carnegie Mellon University in the Robotics Institute. He is currently interested in designing scalable and distributed algorithms for estimation and multi-robot coordination in the physical world. He has worked on personal robotics at Intel Research Pittsburgh, multi-robot active estimation at the University of Pennsylvania's GRASP Laboratory, and miniature inspection robots for the Space Shuttle at NASA's Marshall Space Flight Center. He received his M.S. in Robotics from Carnegie Mellon University in 2007 and his B.S. in General Engineering along with his B.A. in Philosophy from Swarthmore College in 2005.

Location: Ronald Tutor Hall of Engineering (RTH) - 406

Audiences: Everyone Is Invited

Contact: CS Front Desk