A proposed measure of environmental complexity for robotic applications

This paper proposes a quantitative method to describe how complex an environment looks to an individual robot. The assumption is that robots are designed to perform a task in a specific setting or group of settings. The structure of the environment and knowledge the robot has about the environment determine how sophisticated the robot hardware and control algorithms need to be to perform a given task Information about the environment can be 1) obtained from a robot sensors; 2) downloaded or programmed into a robot; or 3) placed in the environment for a robot to read, as needed. Extensive simulations of a search task indicate that a combination of two factors can describe environmental complexity: a measure of entropy and the compressibility of the environment as seen by a robot´s sensors. These combined measures can be used as a yardstick by engineers in determining how to design a robot-environment system. They allow the designer to determine how to divide his/her resources between robot hardware/software capabilities and information/structure in an environment. The complexity measure also allows the designer to determine quantitatively what information is most important for improving a robot´s performance.