Behavior Retrieval for Robot Control in a Unified CBR Hybrid Planner

摘要

This research examines a hybrid planner for a real-world mo-bile robot delivering messages in an office environment. The overall project, RUPART, uses "unified CBR " to combine behavior-based control with high-level case-based planning, using a single similarity metric to retrieve both behavior cases and plan cases. This paper focuses on case-based reasoning for behavior cases. Each behavior case describes a set of be-haviors targeted for a particular environment. We explore fea-tures used to index behavior cases, as well as adaptation and learning methods. This project is incomplete at this time, and thorough evaluation, while underway, is not complete.