The Robotics Institute

I'm interested in artificial intelligence under uncertainty; active sensing and learning; and multi-robot systems, both cooperative and competitive. The last point includes human/robot teams, as I believe a human is simply an organic machine.

Furthermore, I view sensing, learning, and acting as strictly multi-agent tasks:

1. any robot with multiple sensors is in fact sharing information between self-contained modules (usually a higher-level digest of the raw sensor data)
2. supervised learning is merely a description of the interaction between a human and a software agent
3. any robot with multiple actuators must coordinate them and distribute sensory information amongst them
4. any system with only a single sensor, no communication with humans or other robots, and a single actuator is a degenerate case of a multi-agent system, typically neither particularly interesting nor particularly useful

Finally, it is my belief that all the world is a multi-agent game, and we are just players in it. Similarly, just as a society is the interaction of a collection of individual agents, a single human is the interaction of individual cells. To consider a human being an "individual" is really somewhat arbitrary - every cell in your body is just as much an individual as you are, as is every species, every nation.

Currently, I'm working on active sensing and active learning in POMDPs (Partially Observable Markov Decision Processes). I've introduced the concept of an "oracle," an abstract entity that, for some cost, provides information about a system's state, dynamics, or observability. In the real world, the oracle could be a human, another robot, or a sensor.