Onboard Planning for
Geological Investigations using a Rover Team
|
|||||||||
Abstract This paper describes an
integrated system for coordinating multiple rover behavior with the
overall goal of collecting planetary surface data. The MISUS system
combines techniques from planning and scheduling with machine learning
to perform autonomous scientific exploration with cooperating rovers. A
distributed planning and scheduling approach is used to generate
efficient, multi-rover coordination plans, monitor plan execution, and
perform re-planning when necessary. A machine learning clustering
component is used to deduce geological relationships among collected
data and select new science activities. A key concept promoted by this
system is the use of goal interdependency information to perform plan
optimization and increase the value of collected science data. We
discuss how we represent and reason about goal dependency and utility
information in our planning system and explain how this information can
change dynamically during system use. We show through experimental
results that our approach significantly increases overall plan quality
versus a standard approach that treats goal utilities independently.
|
|||||||||
pdf file ![]() |