Population Coding : A New Design Paradigm for Embodied Distributed Systems

Abstract

Designing embodied distributed systems, such as multi-robot systems, is challenging especially if the individual components have limited capabilities due to hardware restrictions. In self-organizing systems each component has only limited information and a global, organized system behavior (macro-level) has to emerge from local interactions only (micro-level). A general, structured design approach to self-organizing distributed systems is still lacking. We develop a general approach based on behaviorally heterogeneous systems. Inspired by the concept of population coding from neuroscience, we show in two case studies how designing an embodied distributed system is reduced to picking the right components from a predefined set of controller types. In this way, the design challenge is reduced to an optimization problem that can be solved by a variety of optimization techniques. Our approach is applicable to scenarios that allow for representing the component behavior as (probabilistic) finite state machine. We anticipate the paradigm of population coding to be applicable to a wide range of distributed systems.