Wheeler Ruml, Ph.D.
Embedded Reasoning Area
Palo Alto Research Center (PARC)
Visit photos

Biographical information: Dr. Wheeler Ruml is Manager of the Embedded Reasoning Area at the Palo Alto Research Center (PARC). His research interests include time-bounded optimization, temporal planning, and real-time control. He received his Ph.D. in Computer Science from Harvard University in 2002. His publications include papers at AAAI, IJCAI, ICAPS (winner of the 2005 Best Applications Paper award), NIPS, and SIGGRAPH.

A CSE Colloquium

Date: Thursday, Nov. 10, 2005.
Refreshments 3:30 p.m., 348 Avery Hall
Talk at 4:00 p.m., 115 Avery Hall

On-line Planning for High-speed Manufacturing

Even in a small manufacturing system, there can be dozens of jobs in process simultaneously. For each job, there can be many possible sequences of actions that will produce it. Can we find the plan that minimizes the overall time for producing all the jobs? I'll discuss the approach we've taken at PARC to a real-life instance of this problem, including complications such as on-line arrival of jobs, modeling errors, and execution failures.

Paper on which talk is based: On-line Planning for High-speed Manufacturing, W. Ruml, M.B. Do, and M.P.J. Formherz. Proceedings of the Fifteenth International Conference on Automated Planning and Scheduling (ICAPS-05), 2005.

A lecture at CSCE 421/821, Fall 2005

Thursday, Nov. 10, 2005, 111 Avery Hall.

Learning to Search Trees

Most backtracking algorithms for CSPs are variations on depth-first search. This is silly. One would expect value selection heuristics to be least accurate at the top of the search tree, when the entire problem remains to be solved, and most accurate at the bottom, when the remaining subproblem is tiny. This is exactly the opposite backtracking order used by DFS. Inspired by discrepancy search algorithms, I've done some work on learning good backtracking orders, which I'll talk about at whatever length the class desires.

Slides of lecture.

Required reading for class: Heuristic Search in Bounded-depth Trees: Best-Leaf-First Search Wheeler Ruml. Working Notes of the AAAI-02 Workshop on Probabilistic Approaches in Search, 2002.

Informal dinner: Thursday, November 10, 2005, with students right after Colloquium.
Acknowledgments: This visit is sponsored by Palo Alto Research Center, the Department of Computer science and Engineering (UNL), and the Constraint Systems Laboratory (UNL).