Experiments in high-level robot control using Congolog: reactivity, failure handling, and knowledge-based search

Abstract

Getting robots to do complex tasks is very difficult. The robot must not only be able to navigate through the world and execute various primitive tasks such as "go forwards by 5 cm" and "take a picture", but also support the specification and execution of complex behaviors, which requires reasoning about goals and actions. The occurrence of exogenous events and runtime failures, and having incomplete information of the environment have long been two major problems encountered by most robots operating in the real world. In this thesis, we develop an approach to building robust high-level control modules using the ConGolog language and prioritized interrupts to handle exogenous events and runtime failures. Such a module is implemented and tested on a mobile robot in a real environment for a mail delivery task. We also develop a knowledge-based approach to guide search for people in the presence of incomplete knowledge.