Fault Detection in a Swarm of Physical Robots Based on Behavioral Outlier Detection

• Published in: IEEE transactions on robotics Vol. 35; no. 6; pp. 1516 - 1522
• Main Authors: Tarapore, Danesh, Timmis, Jon, Christensen, Anders Lyhne
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.12.2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Actuators; Anomaly detection; Collective behavior; Data analysis; Fault detection; Mobile robots; Multi-robot systems; multirobot systems; Outliers (statistics); Reliability; Robot sensing systems; robot swarms; Robotics; Robots
• DOI: 10.1109/TRO.2019.2929015
• ISSN: 1552-3098

The ability to reliably detect faults is essential in many real-world tasks that robot swarms have the potential to perform. Most studies on fault detection in swarm robotics have been conducted exclusively in simulation, and they have focused on a single type of fault or a specific task. In a series of previous studies, we have developed a robust fault-detection approach in which robots in a swarm learn to distinguish between normal and faulty behaviors online. In this paper, we assess the performance of our fault-detection approach on a swarm of seven physical mobile robots. We experiment with three classic swarm robotics tasks and consider several types of faults in both sensors and actuators. Experimental results show that the robots are able to reliably detect the presence of hardware faults in one another even when the swarm behavior is changed during operation. This paper is thus an important step toward making robot swarms sufficiently reliable and dependable for real-world applications.