ROBUST STATE ESTIMATION, INFORMATION GATHERING, AND BEHAVIOR FOR AUTONOMOUS SYSTEMS IN COMPLEX UNCERTAIN DOMAINS
Abstract
Our objective is to carry out basic research that underlies an integrated approach to building intelligent systems that interact with the physical world. The systems we want to build are capable of learning from experience, of operating in highly complex environments involving hundreds or thousands of objects, of solving problems with huge variance in the nature of the objective and the character of the solution, of robustly handling partial observability by reasoning about their own uncertainty and explicitly seeking information as necessary, and of operating over very long time-horizons. In this proposal we focus on complex physical domains with partial observability, he ability to act to gain information, large numbers of objects and long time-horizons. Such domains include operating a busy supply depot, disaster site assessment and recovery, and ISR situations in which information gathering might require physical actions beyond sensor positioning. We develop an approach to acting in domains with partial observability that integrate structural assumptions about objects, relations and actions that provide a framework for cumulative, compositional learning. The unifying idea is that of representing and planning over information states, i.e. beliefs . This provides an elegant integration of perception and action, but raises substantial representational and computational challenges. We propose principled approximation strategies designed to render this approach tractable.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 07, 2023
- Source ID
- FA95502210249
Entities
People
- Leslie P. Kaelbling
Organizations
- Air Force Office of Scientific Research
- Massachusetts Institute of Technology
- United States Air Force