Instruction understanding for intelligent robots in nuclear facilities

As a first step to realize an autonomous mobile robot for plant maintenance, where the robot is capable to understand instructions written in natural languages, we have developed a prototype of instruction understanding system which makes the robot construct its motion sequences to approach instrumentations and inspect them from input sentences written in Japanese. In the prototype system, the instruction understanding and planning capabilities are integrated by an inference engine which consists of a cyclic operation of three processings, i.e., sensing, decision, and execution. Based on environmental data and current states of the robot, a proper process such as natural language processing is triggered by the decision part of the inference engine to accomplish the input instructions. The multiple- and dynamic-planning capabilities, which are necessary to cope with dynamic changes of environments surrounding the robot, are achieved by utilizing the cyclic inference engine together with a set of the inference packets which keep intermediate results of natural language processing and planning for respective input instructions. (orig.)