Supervisory Control of Learning and Adaptive Systems

This was a workshop at AAAI 2004. Here are my opinions of the best talks and research:

Ken Goldberg and Dana Plautz, from UC Berkeley and Intel Research spoke on "The Tele-Actor and Collaborative Tele-Experience"
First, Dana gave an intro in Intel Research, which I, of course, knew about

Then Ken spoke about his work which has progressed from a tele-robot that enabled remote users to move a robot arm in a sandbox and blow air and watch the results, to a robotic gardener which did a similar thing, but in a garden with a seed depositor and water. (I'm actually quoted in his book "A Robot in the Garden") Then he replaced the remote robot with a tele-actor, a person who acted out what a group of network observers wanted. This led to a graphical voting process that allows remote-observers to choose options in tele-actor environments with implications for online education. Then he has done some work on how to resolve several individuals votes on a graphical interface. Really fun, imaginitive ideas.

Patrick Riley from Carnegie Mellon University spoke on "An Agent Coaching Agents": The talk wasn't as interesting as talking to him offline. Why do you want a separate coach agent? Different perspective, different world model, differing computational limitations. How does the coach interact with the agents? Through a shared model, through a shared language etc.

Gregory Kuhlmann from University of Texas in Austin spoke on "Guiding a Reinforcement Learner with Natural Language Advice: Initial results in RoboCup Soccer" An interesting project on adding advice values to an MDP's Q-function.

Tessa Lau from IBM TJ Watson spoke on "Programming Shell Scripts by Demonstration": This was an interesting talk on Programming By Demonstration but it didn't really get into the depth that I wanted to know about and that I know that Tessa knows about.

Keynote - Paul Schenker from Jet Propulsion Lab Cal-Tech NASA is working on a lot of different points along the human-robotic interaction boundary. My opinion is that the state of the art is getting heterogenous robots to work with a human team cooperatively and getting the engineering of the robots to improve. Relevant buzzwords: Supervisory control, tele-operation, tele-presence.