Multivalued Versus Univalued Reactive Fuzzy Behavior Systems for Navigation Control of Autonomous Ground Vehicles Majura F. Selekwa, Emmanuel G. Collins and Joseph Q. Combey Jr. Department of Mechanical Engineering Florida A& M University - Florida State University College of Engineering Tallahasee, Florida, 32310 Email: majura@eng.fsu.edu; ecollins@eng.fsu.edu; combey@eng.fsu.edu Abstract— Applications of autonomous ground vehicles (AGVs) in field operations have expanded from simple transportation tasks to complicated tasks such as military and rescue missions. The complexity in controlling these vehicles increases with the complexity of the tasks that the vehicles are intended for and the environment in which they are to operate. The behavior robotics approach has been adopted as a paradigm for controlling these systems. Due the uncertainty that surrounds the vehicle dynamics and their environments, fuzzy logic control approaches for navigation control have been developed, hence resulting in fuzzy behavior control systems. Two types of behavior structures have been proposed: the univalued and multivalued behaviors. This paper 1 presents a qualitative and quantitative comparison of the structure and performance of these behavior systems. The quantitative performance comparison is performed by using nu- merical simulation results for the motions of two identical AGVs each controlled by using one of the two types of fuzzy behavior navigation control systems; one vehicle uses a multivalued fuzzy behavior system, and the other uses a univalued fuzzy behavior system. The robots are made to navigate across a closed room with random obstacles. I. I NTRODUCTION Behavior based control systems break the control problem into simple sub-problems each of which is controlled by independent simple units called behaviors or reactive behav- iors [1]. Reactive behavior systems rely entirely on sensory inputs for determining the control command, while the classic deliberative systems use stored information for modelling the environment and determining the appropriate control com- mand. The concept of behavior control was initially seen as a special form decentralized switching control in which each behavior is fully autonomous and when allowed it can control the robot on its own without regard to other behaviors. Under that paradigm, the behaviors are designed to be univalued, i.e., each behavior triggers a single control command that best meets the control responsibilities specific to that behav- ior. Over time, there has been a concerted efforts to make 1 Prepared through collaborative participation in the Robotics Consortium sponsored by the U. S. Army Research Laboratory under the Collaborative Technology Alliance Program, Cooperative Agreement DAAD 19-01-2-0012. The U. S. Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation thereon. behaviors run cooperatively so that the overall robot reaction is generally an amalgamation of the commands from the individual behaviors through some form of command fusion [2], [3], [4], [5], [6]; the classic subsumption approach which picked only one command from the behavior that had highest priority [7] can also be viewed a special form of command fusion. Over time, this univalued structure has been found to have serious flaws [8]. In particular, by treating behaviours as fully autonomous, this structure tends to cause the robot to be indecisive when the behaviors have conflicting interests with nearly equal importance. This observation led to the introduction of multivalued behavior control systems [8], [9], [10], [11]. The first fuzzy implementation of multivalued fuzzy behavior systems was in [9] in which two fuzzy values were used: “allowed” and “disallowed”. A more complex multivalued fuzzy structure was implemented in [11]. The differences between univalued and multivalued fuzzy systems have never been documented. Furthermore, there are no documented results that compare the performance of uni- valued and multivalued fuzzy behavior systems. The purpose of this paper is twofold: to describe the qualitative differences between multivalued and univalued fuzzy behaviors systems, and to compare the performance of the two systems. The paper is divided into five sections. Section II describes the structure of univalued and multivalued fuzzy behavior systems. It goes further to explain the qualitative differences between the two systems. Section III describes two navigation control algorithms: one that uses a univalued behavior system and one that uses a multivalued behavior system. Section IV presents simulation results that compare the two algorithms de- scribed in Section III. The paper presents concluding remarks in Section V II. UNIVALUED AND MULTIVALUED REACTIVE FUZZY BEHAVIOR SYSTEMS This section describes the general structures of univalued and multivalued fuzzy behavior systems. These structures can apply to a variety of control applications.