Research Interests

Interests

My research interests include: learning and adaptation of intelligent robotics systems, unsupervised learning in continuous domains, intelligent control, and theories of perception and action.

Thesis Proposal

Committee:

Ashwin Ram (Advisor)

Ronald Arkin

Chris Atkeson

Alex Kirlik

Janet Kolodner

Title: Learning Adaptive Reactive Agents

Abstract: This thesis will study the problem of how autonomous agents can learn when and how to adapt to the environment in which they are embedded. An intelligent agent should be able to determine the characteristics of the ongoing situation it is facing and configure itself to take advantage of such characteristics while considering how to accomplish its goals. Moreover, the agent should be able to learn how to adapt to the environment in which it is operating. That is, agents should learn both the relevant situations in which it is desirable to change the system's configuration and the most appropriate configuration to use under such situations. We claim that agents should learn, plan, and react in an incremental, progressive, and continuous manner, and that in order to support these processes, continuous or analogical representations should be used. Additionally, we claim that the learning goal is to detect, store, and use invariants of the interaction between the agent and the environment. An agent that can learn and represent invariants would be able to use them to react and plan appropriately since it would be taking advantage of the specific ``laws'' that regulate its behavior in the environment it is embedded. We discuss a theory of learning adaptive agents and present two implementations of the theory. This research makes several contributions including an algorithm for learning adaptive laws for mobile robots (robotics), an extension to case-based reasoning method to continuous domains (artificial intelligence), and a theory of conceptual change (cognitive science).

Back to Home Page