A Multistrategy Case-Based and Reinforcement Learning Approach to Self-Improving Reactive Control Systems for Autonomous Robotic Navigation

摘要

This paper presents a self-improving reactive
\ncontrol system for autonomous robotic navigation.
\nThe navigation module uses a schema-based
\nreactive control system to perform the
\nnavigation task. The learning module combines
\ncase-based reasoning and reinforcement learning
\nto continuously tune the navigation system
\nthrough experience. The case-based reasoning
\ncomponent perceives and characterizes the
\nsystem’s environment, retrieves an appropriate
\ncase, and uses the recommendations of the case
\nto tune the parameters of the reactive control
\nsystem. The reinforcement learning component
\nrefines the content of the cases based on the current
\nexperience. Together, the learning components
\nperform on-line adaptation, resulting in
\nimproved performance as the reactive control
\nsystem tunes itself to the environment, as well as
\non-line learning, resulting in an improved library
\nof cases that capture environmental regularities
\nnecessary to perform on-line adaptation. The
\nsystem is extensively evaluated through simulation
\nstudies using several performance metrics
\nand system configurations.