Distributed consensus control of robot swarm with obstacle and collision avoidance

Summary form only given. Research on multi-robot system is receiving a great deal of attention in recent years. Multi-robot system has many advantages over a single robot in certain missions, such as reducing complexity, availability of redundancy, and reconfiguration capabilities. Using coordination scheme in multiple mobile robots allows them to complete tasks with higher complexity. The ability of each robot does not have to be very complete, since each robot can focus on a particular task. Therefore, in some cases, multiple robots working together to complete a certain mission can be relatively cheaper and easier to implement than a single robot. One of the main concerns in the discussion of multi-robot system is formation control. Having a group of robots moving in formation allows user to control the entire group of robots without the need to specify the commands for each robot. This is very useful in many realworld applications, such as search and rescue missions, surveillance, security patrols, military missions, and transportation. Several approaches have been proposed to solve the formation control problem : behavior-based approach to formation control, leader-follower approach, and virtual structure approach. However, in most of these approaches, there exists one group leader (either actual robot or virtual leader) that acts as the central coordination of movement for the whole group. This introduces a single point massive failure problem to the system, i.e. when the leader is disturbed, the whole group will be severely affected. Therefore, we need a distributed approach to overcome this problem. One of the most popular algorithms in distributed control community is consensus algorithm. The algorithm is originally inspired by swarming behavior found in living systems, such as flocking behavior of group of birds. Swarm behavior specifies that in a multi agent system in which each agent acts locally by following simple rules, intelligent collective behavior could emerge, unknown to each agent. Based on this idea, consensus algorithm directs each agent's state towards agreement on a certain value. In general, there are two strategies for formation control using consensus; distance-based control, and displacement-based control. In distance-based control, each robot must keep certain distance with its neighbors. This method requires rigid graph topology for the networked robots. In displacement-based control, each robot tracks its desired position, i.e. the desired deviation from a pre-defined formation center. This method requires all agents to know the formation center. In most cases, multi-robot systems should also be able to work in a dynamically changing environment. Therefore, we also consider obstacle and/or collision avoidance as additional objective while keeping the robots in formation. Two major approaches are employed to this issue : 1) the obstacle avoidance control and distance-based formation control is added (blended) to get the overall control signal, producing smooth movement of the robots; 2) Hybrid model, in which each robot is guaranteed to be in a certain state at one time, either forming formation or avoiding obstacle. In this talk, consensus formation control with obstacle avoidance is addressed using regulated hybrid model as well as blended model. Some theoretical framework concerning consensus, formation and related graph theory for a networked of agents are briefly discussed. We then describe layered architecture of the distributed formation control system that is useful for implementation in a networked of robots. In the blended model approach, a novel scheme is presented, which combines Social Force Model (SFM) for crowd modeling and consensus algorithm. Some simulation studies are presented to show the effectiveness of these approaches. Experimental studies of formation control of swarm of robots are described. The first experiment employs a group of mobile robots connected by wireless communication. The second experiment employs a group of humanoid robots using visual localization. Finally, future research issues are briefly discussed.