Multiple objective action selection in behavior-based control

Abstract

. It is demonstrated that multiple objective decision theory provides a suitable formalism to encompass ideas from behavior-based system synthesis and control, where each behavior is cast as an objective function estimator. Action selection is comprised of generating and then selecting a set of satisficing solutions among a set of solutions that are Pareto-optimal. The basic ideas of the proposed methods are demonstrated through a set of simulated as well as real-world experiments. 1 Introduction In the behavior-based approach the control of a robot is shared between a set of purposive perception-action units, called behaviors. Based on selective sensory information, each behavior produces immediate reactions to control the robot with respect to a particular objective, i.e., a narrow aspect of the robot's overall task such as obstacle avoidance or wall following. Behaviors with different and possibly incommensurable objectives may produce conflicting actions that are seemingly irrecon...