Error Detection in Swarm Robotics: A Focus on Adaptivity to Dynamic Environments

This thesis examines the problem of adaptive error detection in swarm robotics. As part of the challenges for the transition of current swarm robotics research
into the real world implementation, the ability to differentiate between changes to the behaviour due to faulty components and environmental is important. This is a requirement to ensure that robot swarms deployed are fault-tolerant to internal faults as well as external
perturbations. Previous work has investigated this issue from a perspective of a single robot but has largely ignored the aspect of adaptivity to environmental changes. By contrast, this work approaches the problem from a
perspective of a collective and explicitly addresses the issue of adaptive detection. A collective self-detection scheme called the CoDe scheme is proposed and developed. This scheme is demonstrated to work in detecting errors in dynamic environments with the use of various classifiers. This approach has potential to be applied for other domains that share similar characteristics to swarm robotics in which adaptivity to dynamic environments is crucial. Motivated by the potential resource limitations in swarm robotic systems, this thesis also investigates other aspects related to minimising resource usage such as reducing the number of false positives and communication overhead.