The talk will present three main thrusts of my research: how to act in the presence of uncertainty that arises during execution, how to act in the presence of deadlines or in high risk situations, and how to act if the agent's initial state is not completely known. To address the problem of acting in the presence of uncertainty, I will introduce a complete agent architecture built around Partially Observable Markov Decision Process models and demonstrate that this architecture allows agents to act, plan, and learn despite the uncertainty that results from actuator and sensor noise and missing information about their environment. I will show how to combine planning methods from Artificial Intelligence with non-linear utility functions to plan efficiently either for immediate soft deadlines or in high stake domains. Finally, I will explain how to use agent-centered search methods to search non-deterministic domains efficiently by interleaving acting and planning.
Throughout the talk, I will use navigation problems faced by office delivery robots to illustrate the algorithms, and will present formal analyses, experimental results, and a real-world implementation on an autonomous mobile robot controlled over the World Wide Web.
|
Modified: Jun 24, 1997|
Current events|
Your comments
and inquiries are welcome.