Iain Stobie
Research Interests
The goal of my research is to improve the navigation capabilities of autonomous agents.
This work is part of a large-scale distributed agent research effort,
Soar/IFOR project,
using the Soar integrated architecture.
The problem of using diverse sources of knowledge to plan routes according to several
criteria and constraints is being investigated in pursuit-evasion
scenarios, with agents that demonstrate a close integration
of reaction, planning, execution, interruption, replanning, learning, search, use of heuristic
knowledge, navigation, and abstract spatial reasoning. A fundamental question is how can agents
flexibly plan routes for new situations and acquire new strategies for movement.
Existing systems that learn to improve their route planning as a function of their environment
tend to use low-level non-symbolic representations of their environment and route, and
to use experimental techniques such as reinforcement learning that tune parameters as a function
of successes and failures in the environment. In contrast, we are looking at
higher-level symbolic representations and analytical (rather than
experimental) learning techniques, grounded in an ontology and theory of commonsense
planar geometry. We hope to demonstrate that this approach helps agents
to not only plan for themselves but also to understand and use existing specialized tools
to generate, adapt and explain their plans.
Abbreviated Research Biography
Graduate student in the Computer Science Department
at the University of Southern California
(USC), a research assistant at the USC
Information Sciences Institute (ISI) working with Dr. Paul S. Rosenbloom.
M.A. in Mathematics from the University of Edinburgh (1976).
Contact Information
Information Sciences Institute
University of Southern California
4676 Admiralty Way
Marina del Rey CA 90292
Iain Stobie (stobie@isi.edu)