Purpose

In many applications of robotics, it is difficult or impossible for a single robot to complete complex tasks where, for example, a search of a large area or a detailed examination of an object is required. For this reason, systems have been proposed where multiple robots can work together to complete a task ^{1 2 4 5 6 9 10 12 14 16 20 25 26}.

Existing systems, however, are lacking in that they generally use a fixed program, involving simple algorithms ^{5 9 12 16 20 25 26} to complete the task. This limitation prevents the robots from being useful in many applications; they are unable to learn from and adapt to their environment to achieve better performance and learn new tasks.

Most systems which make use of neural networks typically do so to perform a simple function, such as to define a state ², to avoid an obstacle ¹⁰, or for locomotion ^{4 6}. Even systems using neural memory and cognition ¹⁴ do not make full use of the huge potential of the adaptive colony available in a distributed robotics system.

In this project, a customizable design is proposed for a neural-based distributed robotics system, capable of learning to complete advanced tasks. Intelligence in the form of programmable-logic based neural networks and input modules (including miniature cameras, accelerometers, GPS sensors, and microphones) corresponding to various abilities are distributed throughout the group. An algorithm for neurally-adaptive communication to allow group feedback, as well as a simple form of societal development, has been designed, allowing each robot to receive feedback from themselves (self-feedback), each other (group-feedback), and an “expert system” (human teacher or ideal algorithm) when necessary.