Identifying Anomalous and Antagonistic Behavior in Networks of Multibody Systems

EasyChair Preprint 13342

Abstract

Networks of systems are of increasing relevance, in particular in robotics, where a cooperating and communicating network of robots can achieve things unachievable by a single robot. However, while networks of systems with underlying distributed decision-making algorithms bring increased flexibility, they can also be vulnerable. In particular, in uncontrolled environments, it may happen that malevolent agents enter the network, derogating the performance of the whole network. To deal with such scenarios, this contribution proposes two methods to identify anomalous behavior in robotic networks, one of which is based on a model-driven, inverse optimal-control approach, whereas the other uses machine learning in a stochastic framework. To analyze the proposed methods, two well-known problems from distributed robotics, namely formation control and the coverage problem, are considered. The contribution shows that both methods can yield very high detection rates of anomalous behavior, which ranges from merely erroneous to actively antagonistic behavior.

Keyphrases: Resiliency, Robotics, inverse optimal control, machine learning, networks

@booklet{EasyChair:13342,
  author    = {Henrik Ebel and Ingeborg Wenger and Peter Eberhard},
  title     = {Identifying Anomalous and Antagonistic Behavior in Networks of Multibody Systems},
  howpublished = {EasyChair Preprint 13342},
  year      = {EasyChair, 2024}}