Sensor Planning for 3D Visual Search with Task Constraints

Amir Rasouli, John K. Tsotsos

Abstract: Visual search is a fundamental problem in autonomous robotics. Traditionally, visual search is formulated as an optimization problem in which the sequence of actions ischosen based on immediate efficiency. In this paper we examine the effects of the task constraint in the form of maximum allowable cost on action selection in search. We propose three algorithms, namely Greedy Search with Constraint (GSC),Extended Greedy Search (EGS) and Dynamic Look Ahead Search (DLAS), to investigate which algorithm, whether locally or globally, has the most efficient performance under various conditions with a predefined task constraint. We examine our methods in environments of various sizes and configurations with three cost constraints including time, energy consumption and the distance travelled by the robot. Through extensive experiments on a mobile robot, we show that the environment characteristics as well as the type of constraint applied can alter the performance of the methods significantly. We also show that GSC algorithm, which relies on visual clues in an environment to optimize search, achieves the best and most efficient performance in comparison to EGS and DLAS.