Adaptive Behavior Generation for Conversational Robot in Human-Robot Negotiation Environment

Abstract

This study addresses human-robot interactions in a controlled negotiation environment. The aim is to prove that a robot, given its limitations, can win a non-equilibrium based negotiation against a human by convincing him/her. To do so, a behavioral model based on decision trees is proposed, which chooses behavior and action of the robot adaptively depending on the circumstances, robot's intention and human's past response. An experiment under two conditions was conducted:one where the robot was set to play the Desert Survival Situation negotiation game against 10 humans; and one where the robot was compared to other system with the same knowledge about the game but without the behavioral and action generator model. The extracted conclusions were that the robot could win the game in most of the cases, convincing the human. The results also show that its performance is significantly better than the human's and that the other system's robot.