| Summary: | Decision-making is an important operation for any autonomous system. Robots in particular must observe their environment and compute appropriate responses. For solitary robots and centralized multiple-robot systems, decision-making is a relatively straightforward operation, since only a single agent (either the solitary robot or the single central controller) is solely responsible for the operation. The problem is much more complex in a decentralized system, to the point where optimal decision-making is intractable in the general case. Decentralized multiple-robot systems (dec-MRS) are robotic teams in which no robot is in authority over any others. The globally observed behaviour of dec-MRS emerges out of the individual robots’ local interactions with each other. This makes system-level decision-making, an operation in which an entire dec-MRS cooperatively makes a decision, a difficult problem. Social insects have long been a source of inspiration for dec-MRS research, and their example is followed in this work. Honeybees and Temnothorax ants must make group decisions in order to choose a new nest site whenever they relocate their colonies. Like the simple robots that compose typical dec-MRS, the insects utilize local, peer-to-peer behaviours to make good, cooperative decisions. This thesis examines their decision-making strategies in detail and proposes a three-phase framework for system-level decision-making by dec-MRS. Two different styles of decision are described, and experiments in both simulation and with real robots were carried out and presented here to demonstrate the framework’s decision-making ability. Using only local, anonymous communication and emergent behaviour, the proposed collective decision-making framework is able to make good decisions reliably, even in the presence of noisy individual sensing. Social cues such as consensus and quorum testing enables the robots to predicate their behaviour during the decision-making process on the global state of their system. Furthermore, because the operations carried out by the individual robots are so simple, and because their complexity to the individual robots is independent of the population size of a dec-MRS, the proposed decision-making framework will scale well to very large population sizes.
|
|---|
