Efficient Object Search With Belief Road Map Using Mobile Robot

Abstract

This letter describes a pipeline for autonomous object search using a mobile robot. The robot is required to efficiently find an object in an unknown environment. In this letter, we formulate the object-search problem as a Partially Observable Markov Decision Process (POMDP). The semantic information of the room types and the object is utilized for training the belief distribution in the POMDP. We use Gaussian Mixed Model (GMM) to model the distribution of belief states. This letter introduces a novel scheme for the belief propagation when the robot is searching for an object. Both the belief distribution trained by the prior knowledge and the searching experience of the robot are taken into consideration when a new room turns up. The objective function in our formulation concerns both the path cost to the target and the information gain in the target area. We propose a novel graph structure called belief road map (BRM) that is built along with the searching process. The robot can efficiently query a path on the generated BRM instead of on the whole occupancy grid map. To reduce the belief states in the POMDP, the node in BRM that has the most information is selected to represent a room. The simulation and the real experimental studies demonstrate the efficiency and efficacy of our proposed searching approach.