Dynamic Topology Inference via External Observation for Multi-Robot Formation Control

Abstract

Network communication topology through which robots achieve intelligent collaboration in multi-robot formation control systems is of fundamental importance. Existing works focusing on security issues of multi-robot systems usually assume that the topology is priori knowledge. However, the topology graph is not accessible for inner security policies or ID identification from the outside of the system. An intriguing question is how to construct the communication topology via observation from the outside point of view. This work studies the problem of constructing topology graph that represents magnitude of robots interaction via observing trajectories. The main novelties of this work include: i) It is the first time to consider the topology inference problem in multi-robot formation control systems. ii) We transform the inference issue into a linear regression problem. The optimal estimation of Perron matrix that contains the interaction profile is derived using l2-norm least square algorithm (l2-LS). iii) Considering the link failure and creation, we propose a novel dynamic window least square algorithm (DWLS) to identify dynamic changing topology. Finally, simulation results demonstrate that l2-LS has 95% inference accuracy averagely when noise parameter μ = 0.05, and DWLS is robust and stable in identifying time slices, moreover, accuracy approaches 90%.