Towards Fault Diagnosis in Robot Swarms: An Online Behaviour Characterisation Approach

Abstract

Abstract. Although robustness has been cited as an inherent advantage of swarm robotics systems, it has been been shown that this is not always the case. Fault diagnosis will be critical for future swarm robotics systems if they are to retain their advantages (robustness, flexibility and scalability). In this paper, existing work on fault detection is used as a foundation to propose a novel approach for fault diagnosis in swarms based on a behavioural feature vector approach. Initial results show that behavioural feature vectors can be used to reliably diagnose common electro-mechanical fault types in most cases tested.

Adaptive Online Fault Diagnosis in Autonomous Robot Swarms

“…Features are updated at each control-step with threshold values for F 3 , F 4 and F 5 set to 80% of v max (Equation 3), 20% of v max (Equation 4) and 40% of ω max (Equation 5), respectively. These thresholds allow for clear distinctions in individual robot behavior to be represented in the robot's BFVs whilst also being tolerant of simulated electro-mechanical noise, as we demonstrated previously in O'Keeffe et al ( 2017a ).…”

“…One limitation of such approaches is the lack of capability to learn or adapt to dynamic fault signatures, and rely on comprehensive prior modeling in order to be effective. To retain the advantages of BFVs, we would argue that it is more appropriate for an autonomous fault diagnosis mechanism to establish models of faulty behavior online, in which case the resulting system will bear a closer resemblance to Learning Classifier Systems (Shafi and Abbass, 2017 ) than the supervised learning methods described by Daigle et al ( 2007 ) and Carrasco et al ( 2011 ), and used in our earlier work O'Keeffe et al ( 2017a ). Faults can also be diagnosed through more explicit assessment.…”

“…The suitability of BFVs for fault detection in robot swarms has been demonstrated by Tarapore et al ( 2017 ). In our previous work (O'Keeffe et al, 2017a ), we demonstrated that we could design BFVs for a swarm robotic system that could then be used to classify different types of fault using a decision tree Quinlan ( 1986 ).…”

“…The existing literature on active fault tolerance in swarms examines fault detection alone (Tarapore et al, 2017 ) or fault detection and recovery (Khadidos et al, 2015 ), where recovery is only achieved by completely removing the faulty robot from the swarm. To the best of our knowledge, no research has yet been conducted with explicit regard to fault diagnosis in swarm robotic systems, as defined in Şahin ( 2004 ) (with the exception of our own previous work, O'Keeffe et al ( 2017a ) and O'Keeffe et al ( 2017b ). We propose that fault diagnosis is a necessary feature of active fault tolerance in order to ensure long-term autonomy.…”

Mobile Robot Simulation Framework

“…It was designed to cover the area of large-scale heterogeneous robot swarms simulation which was not provided by already existed simulators [2]. ARGoS simulation of 10 marXbot robots performing a dispersion behavior can be found in [12].…”