MERGEABLE NERVOUS SYSTEMS FOR ROBOT SWARMS
Abstract
Swarm intelligence studies natural and artificial systems composed of many individual agents that coordinate their actions using decentralized control and self-organization. In such systems, collective behavior emerges as the individual agents interact locally with each other and with the environment in which they are situated. Examples of this collective behavior can be found both in nature and in man-made systems. Research has shown that by avoiding centralized control (i.e., having a single entity that acts as a decision or command center), swarm intelligence systems have the potential for parallel, flexible, and scalable operation. In the swarm intelligence research community, it is an axiom that centralized control should be avoided and that instead large-scale swarm intelligence systems should be purely self-organized and based on distributed control mechanisms. However, by avoiding all forms of centralized control it becomes di cult to interact with a swarm of agents, not to mention to ensure that the swarm is robust and trustworthy. While self-organizing systems are designed to be scaled up to large numbers of agents, the problem of managing such systems and of ensuring that they are robust and trustworthy, is made significantly more di cult by this increase in the number of agents. This increase in difficulty is due to the highly complex dynamics of self-organizing swarms, whose complexity increases exponentially with the number of agents involved. On the other hand, centralized systems are relatively much easier to understand and control; however, the application of any form of centralized control to a swarm intelligence system tends to immediately negate the benefits of self-organization. Furthermore, and in contrast to self-organizing systems, systems that utilize centralized control are limited in their parallelism, scalability, and flexibility, and by definition have at least one single point of failure. In this proposal, we discuss our plans to investigate how self-organization and centralized forms of control can coexist so that swarm robotics systems can bene t from the advantages of both control paradigms. That is, we want swarm robotics systems that are not only parallel, flexible, and scalable but also manageable, robust, and trustworthy. To this end, we intend to investigate whether it is possible to achieve a coexistence of these forms of control by allowing ad hoc, on they centralized control structures to form as a result of self-organizing behavior. These centralized control structures will be formed to perform specific operations and will be dissolved when they are no longer required. Our main hypothesis in this research is that a system, whose coordination is predominantly self-organized but where centralized control structures are allowed to form on demand, will outperform systems using either strictly centralized or strictly self-organized control mechanisms. A key challenge of this research will be to identify the balance between these two control paradigms. In this research, we intend to study and to gain an understanding of the requirements of the agents in a swarm that will enable the formation of ad hoc, on they centralized control structures as a result of self-organizing behavior. The analysis of these requirements will be done in terms of the communication mechanisms, the amount and type of information exchanged, the frequency of communication, and the computational capabilities used by the agents. Based on this understanding, we intend to identify design principles that will contribute to a framework for the implementation of swarm intelligence systems that are manageable, robust and trustworthy.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 07, 2019
- Source ID
- N629091912024
Entities
People
- Marco Dorigo
Organizations
- Office of Naval Research
- United States Navy