An Adaptive Markov Process for Robot Deception.

Abstract

Deception has a long history in the study of intelligent systems. Animals and humans both gain several advantages from deception, hence researchers started developing different ways to introduce deception in robots and in two-player interactive games. This paper investigates deception in the context of motion using a simulated mobile robot. To our knowledge, there have been no foundational mathematical underpinnings developed for robot deception, but some researchers have worked in the past on specific robot deception applications. We first analyze some of the previously designed deceptive strategies on a mobile robot simulator. Then, we present a novel approach to randomly choose target-oriented deceptive trajectories in an adaptive manner to deceive humans in the long run. Additionally, we propose a new metric to evaluate deception in the data collected from the users when interacting with the mobile robot simulator. We performed three different user studies to test effectiveness of different deceptive strategies and our adaptive algorithm in the long run. The statistical evaluation of these studies showed that the proposed adaptive deceptive algorithm did deceive humans in the long run and it is more effective than a random choice of deceptive strategies.