Warning:
This page is provided for historical and archival purposes only. While the seminar dates are correct, we offer no guarantee of informational accuracy or link validity. Contact information for the speakers, hosts and seminar committee are certainly out of date.

RI SEMINAR -- Maja J Mataric

• WHEN: Friday, April 22, 1994; 3:30 pm - 5:00 pm Refreshments will be served starting at 3:15 pm
• WHERE: ADAMSON WING Auditorium in Baker Hall
• SPEAKER: Maja J Mataric

          Artificial Intelligence Lab
          Massachusetts Institute of Technology
         

• TITLE: Group Behavior and Learning in Autonomous Systems

ABSTRACT

Traditional AI has focused on complex agents in well-behaved worlds, while reactive and behavior-based approaches have dealt with simple agents in noisy and uncertain environments. Many interesting domains are left untreated, in particular those requiring intelligent interaction and learning in multi-agent, complex and uncertain environments.

The first part of the talk will describe the complexities intrinsic in synthesizing and analyzing situated group behavior. Basic behaviors will be introduced as an approach for generating adaptive, goal-driven group behaviors from simple local interactions between agents. Two methods for combining basic behaviors into higher-level aggregates will be described. The theory will be demonstrated by defining a set of basic behaviors for the spatial domain and applying both combination methods to it. A video illustrating the experimental design and the results with up to ten physical robots will be shown.

In the second part of the talk the basic behavior approach will be extended to include learning to coordinate behaviors, i.e. to automatically combine them into higher-level aggregates. The weaknesses of existing approaches to learning in situated, non--Markovian domains with noisy and hidden state, multiple goals, and inconsistent reinforcement will be discussed. An approach that enables and accelerates learning in such domains will be described, analyzed, and demonstrated on a group of four robots learning a foraging task. Experimental results will be discussed and a video will be shown.

The conclusion of the talk will address methods for applying the described approaches to learning more complex behaviors and generalizing them to other situated, interactive domains.

Host:           Yangsheng Xu (xu@cs.cmu.edu)
Appointment:    Mike Erdmann (me@cs.cmu.edu)