Fuzzy Logic-based Multi-robot Cooperation for Object-pushing Yifan Cai and Simon X. Yang School of Engineering, University of Guelph Guelph, Ontario, N1G 2W1, Canada {ycai, syang}@uoguelph.ca Abstract— Multi-robot cooperation is an important issue in robotics. Collaboration can improve the productivity and complete some complex tasks effectively. In this paper, a two- stage fuzzy logic-based control scheme is proposed for a team of robots to cooperatively push an object to a target location. A fuzzy logic-based multi-robot cooperation is proposed to effectively generate collision-free paths for the robots, where the number of robots and the environment are uncertain. The simulation results demonstrate the feasibility of the proposed approach. In comparison to other methods, the proposed fuzzy logic-based control is easier to implement and is more effective at resolving problems with uncertainties. Index Terms— Multi-robot cooperation, fuzzy logic control, coordination, membership function. I. I NTRODUCTION The concept of multi-robot cooperation was first intro- duced in the early 1980s [1]. Multi-robot cooperation can deal with many tasks that are difficult to accomplish by an individual robot. A team of robots can provide special redundancy and contribute cooperatively to solve the assigned work “in a more reliable, faster, and cheaper way” [2]. The key point is that each robot just concentrates on one assignment. So it can enhance its advantages and compensate disadvantages by cooperation with others. In a centralized multi-robot system, the controller is a key composition. Design of controllers based on traditional techniques usually involves the assumption of exact knowl- edge about the system. But this assumption is often not valid since the model of a practical system sometimes lacks required precise information due to the uncertainty in the environment. To deal with it, there have been increasing interests in the utilization of novel control strategies, such as neural networks (NNs) [3] and fuzzy logic [4]. These control methods derive their advantages from the fact that they do not use any mathematical model of the system obtained from the first principle [5]. Instead, NNs uses input-output relations while fuzzy logic uses heuristic knowledge about the system [6]. Furthermore, Wang and de Silva [7] implemented a variant including a mechanism of decision based on a Markov process, which is known as Q-Learning. Beside environment model, communication also deserves quite a few attentions from scientists. But sometimes it cannot be regarded as absolutely accurate. Yamada and Saito [8] introduced a fresh idea in 2001, which adaptively selects actions without explicit communication. On the contrary, Newman et al. [9] implemented the client-server schemes to the involved robots to raise the robust character to handle the collisions. Kovac et al. [10] implemented the reinforcement learning algorithm to the solution of communications be- tween the robot members. In this paper, the communications between the robots and the central controller are assumed to be sufficient. The path planning is always a vital step in multi-robot cooperation. Otani and Koshino [11] developed a path plan- ner based on the rapidly exploring random tree (RRT). Wang and Kumar [12] proposed the manipulation by surrounding the object with series of robots. In the control process, the position of the object can be adjusted by the position of each robot that encloses and pushes the object. In this paper, fuzzy logic-based controller is implemented. Dupre and Yang [13] proposed a two-stage fuzzy controller. The outputs from the first stage of the controller are among the inputs of the second stage. The research here adopts this idea and tries to design a two-stage fuzzy logic controller, in which appropriate membership functions and fuzzy rules are implemented. It inputs the previous robot velocities and distances detected from the obstacles, and then outputs proper commands to the robots. Moreover, it is expected to deal with uncertain robot members, which is another key problem in current multi-robot cooperation research. In order to optimize the control rules, there is a need to develop efficient methods to tune the membership functions and obtain optimal rule base. Cai et al. [14] proposed the Step-Forwards scheme. In this paper, this approach is improved and more reasonable outputs are designed. In addition, all the designed strategies will be tested in the simulations. II. THE PROPOSED FUZZY CONTROL SYSTEM In the scheme, several robots pushing one object is studied. It is a basic cooperation between robots. The key problem for the robots is to cooperatively push the object to the target point, while robots can effectively avoid all the potential obstacles along the path. Proceeding of the IEEE International Conference on Information and Automation Shenzhen, China June 2011 978-1-61284-4577-0270-9/11/$26.00 ©2011 IEEE 273