Safe Robot Swarms

: Methods to Prevent, Detect, and Mitigate Faults

Abstract

Robot swarms have great potential to be used for real-world applications. As with any system implemented in the real-world, swarms are susceptible to failure - both at the level of an individual robot and the swarm as a collective. For swarms to be successfully deployed in real-world applications, it will be necessary to develop methods for fault detection, diagnosis and recovery (FDDR). There is a large body of work for FDDR for individual robots but methods specific to robot swarms are still underexplored. In this thesis, I consider the key characteristics of a swarm to be its distributed and decentralised operation, where individuals act according to locally sensed variables. These characteristics inform my approach to FDDR for swarms. In particular, inter-robot interactions and robot-environment interactions give rise to emergent behaviour at the level of the collective. Emergence arising from complexity makes it difficult to model the propagation of faults. As such, a major theme of this thesis is a focus on data-driven methods which offer a practical and effective means of working with emergent swarm behaviour. Additionally, I focus on the local information and actions available to individual robots in order to self-detect faults and mitigate their impact on overall swarm performance.

Date of Award	30 Sept 2025
Original language	English
Awarding Institution	University of Bristol
Supervisor	Sabine Hauert (Supervisor)