Fast Search Strategy for Robots in Dynamic Home Environment

Abstract

In an unstructured home environment, environmental information is mostly disorganized. It is difficult for a service robot to obtain sufficient service information, which significantly hinders task execution. To solve this problem, a new object search strategy is proposed for improving the speed and accuracy of object search in a complex family environment. In this method, a family-environment knowledge graph is constructed using real environmental information and human knowledge, which plays a guiding role in task execution. The home environment is divided into three levels: functional rooms, static objects, and dynamic objects. The co-occurrence probabilities are obtained from open knowledge sources, including the probabilities between static and dynamic objects and between static objects and functional rooms. They are combined with ontology knowledge based on the home to form prior knowledge of a service robot. Inspired by the human search process, a distance function is introduced to calculate the distance between the robot and target objects for optimizing the search strategy. To improve the robustness of robotic services, we designed a probabilistic update model based on the service tasks and knowledge databases. Experimental results indicated that the proposed search strategy can significantly shorten the search time and increase the search accuracy compared with methods without prior knowledge and the distance function.