A DYNAMICALLY DISTRIBUTED CONTROL FRAMEWORK FOR ROBOT TEAMS

Adrian Martin and M. Reza Emami

Keywords

Robot teams, distributed systems, control system architecture, fault tolerance

Abstract

This paper discusses an implementation of a versatile architecture, labelled as HAA (Host, Avatar, Agent), using provably correct distributed algorithms. The HAA architecture lays a foundation on which virtually any type of control system can be built, while providing features such as dynamically distributed processing, distributed database, and failure recovery. The algorithms for the key components of the architecture are specified, including host membership service, agent allocation, agent transfer, and agent recovery. The first component controls the processing network, while the second component oversees the dynamic distribution of processes. The third component ensures that processes are transparently transferred without any loss of data, and the fourth component facilitates the recovery of failed processes. The real-time performance of the algorithms was measured in experiments for mapping, exploration, and foraging and proved to be effective for the system sizes tested.

Important Links:

Go Back