Instruction Parsing and Action Sequence Generation System for Home Service Task

Abstract

Service robots have broad application prospects in home service tasks, but there are problems such as difficulty in task analysis, poor environmental adaptability, and system integration. To more accurately understand user intentions and plan service steps in line with the current environment, a family service task-oriented robot instruction parsing and action sequence generation system is proposed. The system is connected by a knowledge-base-based human-robot interaction question answering system. First, an improved gate and prior knowledge vector are proposed for instruction parsing. The relationship between task types and keywords can be better extracted, and prior knowledge can be used to improve recognition accuracy. Then a service strategy is proposed as an intermediate state of the instruction to action sequence. The task steps are generated in the form of sequence text using keyword guidance to facilitate the generation of action sequences. At the same time, it performs visual inspection and knowledge base query for environmental objects. Robots fill in the missing through human-robot interaction to generate the most suitable service strategy for the current environment. Finally, domain and problem descriptions that are more in line with family service tasks are defined. Robots can generate executable sequences of actions through a task planner. The experimental results show that the system can improve the accuracy of instruction parsing and generate logical and reasonable action sequences.