behavior triggers can easily be mimicked in the robotic world to obtain very useful results.. In the firefighting example, a robot that noticed the water getting low could self-invoke a behavior to get more water, even if it was in the vicinity of the group. Many researchers have focused their efforts toward the study of ant colonies.. Kube and Bonabeau study ants that cooperate to move large prey. The ants systematically rearrange their positions until the prey can be lifted and transported back to the bed. This behavior is mimicked in a group of box-pushing robots. Behaviors need not be complex; as Kube and Zhang observe, many behaviors can be realized in combinational circuits. Examples of 15 simple behaviors include wandering and obstacle avoidance. In a typical robot system, behaviors are embedded in the control architecture and are intended as building blocks for achieving higher-level goals. In multi-robot systems, the simple and complex behaviors of each robot are combined to form a group behavior that is both new and desirable. To make this point clear, consider two robots that have the task of pushing a box from point A to point B. One robot is a pusher and the other is a steered. Notice that this is a heterogeneous, intentional cooperation approach. On a single robot level, the pusher will have a contact behavior that will keep it in contact with the box and a drive behavior that will keep it in forward motion. Taken individually, these behaviors accomplish nothing, but when combined the robot will push a box. Likewise, the steered will have a homing behavior to get to point B and some behavior to keep the box between itself and point B. On the multi-robot level, the pushing and steering behaviors of each robot are useless when acting alone; but when combined, the task is accomplished.
Behavior-based robotics started in the late 1980’s after Rodney Brooks proposed a new method, known as sub assumption architecture, for controlling autonomous mobile robots.
Kube propose a set of four behaviors necessary for a box-pushing group of robots: find, slow, goal, avoid. His approach is a selfishly-motivated swarm approach in which each robot is attempting to push the box by itself, as opposed to the intentionally cooperative push-steer method mentioned earlier. The find behavior keeps the robot in motion and avoid keeps it from colliding with other robots. The Slow behavior allows the robot to push the target or box without causing damage to itself. The 18 Goal behavior keeps the robot moving the box toward a lighted goal. The transition in these behaviors is controlled through environmental cues.
Parker has developed the ALLIANCE architecture based on a strict sub assumption-style approach. Her architecture is non task-specific and is designed for maximum fault tolerance. She adds two novel mathematically-modeled motivations called impatience and acquiescence. Impatience is a parameter associated with a particular behavior set that initializes to zero and grows as long as a particular task is not being accomplished. When it reaches a threshold, a robot may decide to take over that particular task. While impatience reflects the fact that other robots may fail, acquiescence reflects the fact that a robot itself may fail. When this parameter reaches a threshold, a robot may decide to give the task over to another robot. These parameters work together to solve problems such as malfunctioning hardware and the sudden need for additional tasks to be performed.
A second generation robot platform is currently in the design phase. It will have multiple sonar sensors and a Gumstix™/Robostix™ embedded computer that will run the Player device server under Linux. The use of Player as an interface between the control programs and the robot will allow for faster development of code that is more 28 29 reusable. For example, in the event that a device or sensor is changed or upgraded, only the device driver will need to be rewritten—the control program will remain unchanged. Also, any control programs written for the robot will be tested and debugged in the Stage simulator first. This second generation robot platform is a major step toward the hardware implementation of many complex cooperative algorithms such as SARA-1. In conjunction with the robotic software and hardware development, the CRR team is also investigating the use of sensor “motes” (wireless communication boards) on mobile robot platforms. The team currently has eight Moteiv Tmote Sky™ sensor motes. The motes are equipped with several sensors such as temperature, humidity, and light sensors as well as onboard radio communication capabilities. These sensor motes are typically used in a wireless network to share environmental information between units and a base station. The CRR team hopes to incorporate these sensor motes with future generations of mobile robot platforms to allow inter-robot communication and possibly robot-human communication.
Experiments from the literature
Number of Robots Experiments Finding an appropriate number of robotic agents for a particular task and environment is a classic research scenario in the cooperative robotics literature .Typically, the positive effect of cooperation is weighed against the negative effect of cooperation by products such as inter-robot interference, competition for resources, or communication and computational overhead. There is usually some threshold in which the negative byproducts of cooperation begin to outweigh the positive effects—at this point the task achievement begins to deteriorate with the addition of more robots. Researching that threshold, identifying the reasons for it, and finding out what makes it change remains an open area of research in cooperative robotics. Halme experiment with the number of robots in a swarm approach to a stone-collecting task. In this set of experiments, it is found that, as the number of robots increases, the time per stone decreases while the number of collisions increases. This illustrates the positive effect of cooperation versus a negative effect: inter-robot interference.
Schneider-Fontán and Matrix point out that the main cause of inter-robot interference is the competition for space. They explore easing this competition for space by setting up non-overlapping territories for each robot in a puck-gathering task. This spatial isolation minimizes interference; however, task performance still degrades when the number of robots is increased above a threshold. In contrast, the results in this research report show that increased cooperation reduces inter robot interference; but there is still a threshold in the number of robots that causes the time for task completion to deteriorate.
A search-and-rescue algorithm for multiple robots cooperating through wireless communication has been presented. The algorithm, referred to as SARA-1, was shown to be robust, adaptable, and scalable both in theory and during experimental runs. It was also shown to respond well to the breakdown of communication and to the malfunctioning of other robots. It is applicable to both indoor and outdoor environments. Several experiments were run in simulation using SARA-1. Three parameters were tested: the number of robots, the communication interval, and the complexity of the environment. In each set of experiments, the number of robots was varied and it was seen that there is typically an optimal number of robots which will minimize the time for task completion. This is due to a tradeoff between the benefits of cooperation and the burdens of communication overhead and inter-robot.
Obstruction gestations for Future Work
§ Install Player/Stage on multiple machines to enable parallel simulation runs.
§ Conduct further experiments in more complex environments.
§ Use Player’s “subscribing” ability instead of the MCom push/pop/read stack for inter-robot communication. This would require some alterations of SARA-1 but would eliminate the need for the sensor motes. It would also have an effect on the time for task completion.
§ Modify the code to enable the handling of “newcomers”—robots added in during simulation.
Contribution to the Field of Cooperative Robotics This research report has presented, to the author’s knowledge, There have been few experiments presented in the literature involving a cooperative search-and-rescue task and Robotic platforms that are capable of running the SARA-1 algorithm are currently in the design phase.
C. R. Kube and H. Zhang, “Collective Robotic Intelligence,” in Second International conference on Simulation of Adaptive Behavior, MIT Press, 1992, pp. 460-468. 10 C. R. Kube, Collective Robotic Intelligence Project (CRIP) Home Page: http://www.cs.ualberta.ca/~kube/research.html 11 C. R. Kube, E. Bonabeau, “Cooperative transport by ants and robots,” in Robotics and Autonomous Systems, vol. 30, 2000, pp. 85-101.
23 J. Liu and J. Wu. Multi-Agent Robotic Systems. CRC Press: London, 2001.