Real-world domains are notoriously hard to model completely and
correctly. Robotics researchers have developed some learning
capabilities for their systems, but generally these have been limited
to operational parameters or other low-level information. Real-world
systems should adapt to changing situations and absorb information
that will improve their performance. I will be presenting the
complete integrated planning, executing, and learning robot ROGUE,
which analyzes execution experience to detect patterns in the
environment that affect plan quality. ROGUE extracts learning
opportunities from massive, continuous, probabilistic execution
traces. These learning opportunities are then correlated with
environmental features, thus detecting patterns in the form of
situation-dependent rules. I will describe the development and use of
these rules for two planners: the path planner and the task planner,
and present empirical data to show the effectiveness of ROGUE's novel
learning approach. This learning approach is applicable for any
planner operating in any real-world domain. Situation-dependent rules
effectively improve the planner's model of the environment, thus
allowing the planner to predict and avoid failures, to respond to a
changing environment, and to create plans that are tailored to the
real world.
Date: Thurs., March 12; Time: 4:15-5:30PM; Place: Gates 100
Return to seminar schedule.