CCNY Lecture Series

Computer Vision, Robotics and Human-Computer Interaction


Title: Constraint Maintenance in Adaptive Behavior-Based Robot Systems

Roderic A. Grupen
Professor, Computer Science Department
Director, Laboratory for Perceptual Robotics
University of Massachusetts at Amherst
Date: Friday, May 13, 2005

Time: 10:30 AM - 11:30 AM
Room: Steinman Hall, Room T-512

ABSTRACT

This presentation describes techniques and representations that facilitate the design of distributed and sensor-based, reactive robot control policies. Distributed controllers introduce issues regarding the use of multiple internal models, the management of control interactions, process synchronization, and communication, but they also promise to provide practical controllers for complex control tasks including the coordination of many individuals.

We view these kinds of systems as redundant spatial mechanisms and construct multi-objective, concurrent controllers by using a generalization of null-space (pseudoinverse) control. As in many related approaches, behavior in our approach is expressed by sequences of control actions. However, in the UMass work, control actions consist of combinations of control primitives and resource designations. The basis controllers introduce a finite set of system equilibria, thus permitting the end-to-end behavior of the system to be modeled as a discrete event system, which allows powerful formal techniques to be applied to further structure the behavior of the system. By explicitly exploring system resources, we acquire control policies that are robust and responsive to a wide range of operational contexts. Moreover, stochastic exploration techniques for adaptive optimal control design can be guaranteed not to violate design constraints.

This perspective has been applied successfully to many kinds of robotic platforms and as a computational model of sensorimotor development in human infants during the first year of life. I will illustrate our approach with several examples drawn from these studies including: manipulation systems; coordinated policies for mobile robots; and distributed sensor networks.

Biography

Professor Grupen is a Professor of Computer Science at the University of Massachusetts. His primary research interests include:  robot learning and adaptation, sensor-based modeling and control, dexterous robots, multimodal sensory interpretation, resource allocation, and architectures for real-time autonomous systems. Grupen received a B.A. in Physics from Franklin and Marshall College, a M.S. in Mechanical Engineering from Penn State University, and a Ph.D. in Computer Science from the University of Utah in 1988. Professor Grupen is the Director of the Laboratory for Perceptual Robotics at UMass, an editor for AIEDAM (AI in Engineering Design and Manufacture), and the Robotics and Autonomous Systems Journal, and serves on several program committees.

The lecture series is supported by CCNY School of Engineering, and a planning grant from NSF Minority Institutional Infrastructure program.