Robot Evidence Based Search for a Dynamic User in an Indoor Environment

Abstract

In this paper we present the development of an evidence-based search planner for a mobile assistive robot to autonomously search for a dynamic person in a multi-room home environment in order to provide assistance. We solve the dynamic person search problem by uniquely considering evidence of household objects along with a user spatial-temporal model to increase the probability of finding the user. Our planner utilizes a Partially Observable Markov Decision Process (POMDP) to plan optimal robot search paths in the environment as the user and evidence locations are partially observable. Extensive simulated experiments in a home environment were conducted to compare our proposed evidence-based search approach with 1) a search technique without prior user information, and 2) a search technique that only uses a user model. The results show that our proposed search technique has higher success rates for finding the user and is more robust to the dynamic behaviors of the user.