David's World

August 12, 2005

Ant roboting: Lots of little guys trying to become a smart mob

I haven't written much lately here. The Silence of the Lambchops. "Boy, you better have damn good reasons for that!" Yes, Sir, I have reasons.
a) slightly different priorities (writing a research proposal) and b) attending conferences. IJCAI for example.

IJCAI is one of the largest conferences in Artificial Intelligence research, and it was fun meeting people in the field and finding out what other researchers are up to. People in "Swarm Robotics" build miniature robots that can team up to achieve simple tasks. If a few of them break down: no problem! Among the questions of interest are: how do they communicate in the network, when they can only transmit to their neighbors (lacking a strong battery)? Just like ants or bees or the Internet, they can pass on messages in a redundant information network. How do the agents assume roles and commit to their tasks? How does a distributed system integrate redundant information and reason under conditions of uncertainty?

These are quite practical problems: it's not that trivial to actually instruct them to do a specific task as a team, such as delivering drugs right into a tumor somewhere in the human body. An excellent tutorial about collaborative multiagent systems, given by Barbara Grosz (from Harvard, she's highly respected for her work on human dialogue and discourse coherence), Charles Ortiz (SRI) and Milind Tambe (USC CompSci) helped me understand better what the principal issues are. Too bad that the practical Swarm Robotics workshop could only show somewhat simplistic applications and systems. I wonder if the robotics researchers would be better off simulating little robots in software form...