This project will advance the understanding of swarms of simple vehicles that can collectively accomplish complex tasks. Prof. Rossi and Shen take their inspiration from social animals such as flocks of birds and schools of fish that communicate using simple rules to coordinate motion and make decisions for the entire group. There can be significant differences between rules found in biology and technological applications. For instance, birds coordinate motion with visual cues whereas networked robots coordinate activities using wireless communication. These two types of communication provide different kinds of information and have different limitations. Profs. Shen and Rossi are particularly interested in autonomous underwater vehicles because they face the challenge of low or irregular communication quality and poor position information.
Profs. Shen and Rossi will combine mathematical modeling and analysis with engineering applications to study specific problems. First, they will explore leadership within swarms to understand how small numbers of individuals within a swarm can shape decision making for the entire group. Second, they will study swarm behavior in the presence of background flows to understand how currents can distort or disassemble swarms that would be effective in calm conditions. Third, they will study threat avoidance strategies to understand how complex motion can protect a swarm when individual evasion is impossible. Finally, they will study coordinated detection of a scalar field such as nutrients or contaminants in the ocean. Profs. Shen and Rossi will create an interdisciplinary research team around these problems involving students from both mathematics and computer science.
The integration of advanced computation, wireless communication, and control technologies has facilitated the creation of swarms of wirelessly networked autonomous vehicles to carry out critical civilian and military tasks. In this project, we focused on the development, rigorous analysis and validation of bio-inspired swarming algorithms to control autonomous underwater vehicle (AUV) swarms. Shortly after the start of the project, the British Petroleum/Gulf Oil Spill provided us with greater motivation to resolve fundamental questions in autonomous swarming and sensing. A swarming algorithm defines a set of rules which a group of autonomous vehicles follow to interact locally with other proximal vehicles without any centralized control. We focused on three-zone swarming algorithms which are very common in biologically-inspired applications. We developed and augmented a continuum approach where one treats the swarm density and velocity as a continuous function of time and space. Results from analysis of the continuous model automatically scale to large numbers of autonomous vehicles. Our model system has been integrated into a wireless communication layer, and our predictions were tested and refined via network simulations and prototyping.Our project has had several important outcomes. Understanding basic swarm behavior: Swarm algorithms (for robots and in the life sciences) have many parameters. To use swarms of vehicles effectively, one must understand how the choice of parameters affects the behavior of the swarm. Examples of swarm behavior include falling apart (not swarming), chaotic swarming in one area, aligned movement and milling in circles. We have determined distinct stability criteria for swarms and these theoretical predictions are supported by realistic wireless simulations. Swarm stability and memory: In biological systems, such as flocks of birds or swarms of fish, swarm interactions are visual and the rate at which an individual can sense changes in the relative positions of nearby individuals is quite high. In a swarm of robots, typically this information must be shared over a wireless channel, so the rate of coordinated movement in the swarm is constrained. Through mathematical analysis, we found ways to optimally store trajectory information in a new swarming algorithm that requires less communication while still remaining coordinated. Swarm interpolation: Many tasks, such as monitoring the extent of the Gulf Oil Spill, trend toward the deployment of large numbers of less complicated but coordinated devices rather than the use of small numbers of more sophisticated devices. In these applications, coordination among nearby autonomous robots is essential. While there have been many efforts dedicated to determining the boundary of a measured field, often investigators need to know the full field. We refer to this challenge as the swarm interpolation problem and in this research activity, we have developed a method to accurately sample the full field using coordinated wireless robots, even in cases where the sensed fields are moving or deforming in time. Covert leadership: One way to get a swarm to do what you want it to do is to inject information into the swarm. A covert leader is an individual who possesses additional information, but do not explicitly identify themselves to others (followers). Additional information can guide the swarm toward a desired location, away from a threat or around a barrier. We have developed a nonlinear leadership model where leaders remain embedded in the swarm, and we have analyzed the stability of swarm structures with embedded leaders and found that the stability qualities are no different than for leader-less swarms. Thus, it is possible to add additional information to a swarm without changing its shape or extrinsic qualities. Theoretical information transfer in swarms: In collaboration with colleagues at Australia's Commonwealth Scientific and Industrial Research Organisation (CSIRO) Information and Communication Technologies (ICT) centre, we have quantified the theoretical information transfer that must occur in swarms. This is, in effect, a lower bound on interactions that must occur for swarming to occur. One must remember that interaction and communication is the limiting resource in robot behavior because communication channels are shared. Communication must be minimized, but most swarming algorithms assume limitless interactions between nearby individuals. Working with the CSIRO team, we have measured theoretical information transfer during swarm interactions which can tell us when it is necessary to interact and when it is not necessary to interact. Human resources: This project has led to the creation of an interdisciplinary Swarm Intelligence Group spanning two departments (Computer & Information Sciences and Mathematical Sciences). In addition to being visible to those working directly on this funded project, it has also attracted personnel who are not directly funded by the project. This group has provided mentoring and training to 3 postdoctoral fellows and 6 graduate students. One graduate student is now teaching at a four year college and continuing her swarming research. Another is a tenure track faculty member at a university and has been mentoring undergraduate research projects on swarming problems.
