Average Consensus in Multi-Agent Systems with Non-uniform Time-Varying Delays and Random Packet Losses

A Decomposition Approach to Multiagent Systems With Bernoulli Packet Loss

“…A key ingredient to the approach proposed in [9] is to analytically calculate the expectation of the Laplacian L σ k and the product L ⊤ σ k L σ k . Similar calculations were performed before in [5] without exposing the inherent structure of the result, which is required for decoupling. However, neither result is applicable to stochastic Laplacians as resulting from Assumption 1 because both require homogeneous loss probabilities and independence in time, i.e., p ij = q ij = p for all e ij ∈ E 0 .…”

“…There is a rich literature on analysing MJLSs, both in terms of stability and system performance. A direct application of the general analysis conditions does, however, lead to a combinatorial explosion for MASs with more than a few tens of agents and even applying simplifications as proposed in [5] does not allow to consider truly large-scale MAS due to the remaining quadratic scaling of the computational complexity in terms of the number of agents. For that reason, existing approaches on MASs with stochastic packet loss require additional assumptions on the packet loss model and are C. Hespe and H. Werner are with the Hamburg University of Technology, Institute of Control Systems, Eißendorfer Straße 40, 21073 Hamburg, Germany, {christian.hespe, h.werner}@tuhh.de mostly concerned with stability analysis rather than system performance: Many results rely on having identical packet loss, that is, all links fail at the same time [6], which is a very restrictive assumption and violated in almost all real world scenarios.…”

“…For that reason, existing approaches on MASs with stochastic packet loss require additional assumptions on the packet loss model and are C. Hespe and H. Werner are with the Hamburg University of Technology, Institute of Control Systems, Eißendorfer Straße 40, 21073 Hamburg, Germany, {christian.hespe, h.werner}@tuhh.de mostly concerned with stability analysis rather than system performance: Many results rely on having identical packet loss, that is, all links fail at the same time [6], which is a very restrictive assumption and violated in almost all real world scenarios. Approaches that consider not identical loss but homogeneous loss probability are closer to reality, amongst which are [2], [5], [7]- [9], all assuming Bernoulli distributed packet loss, i.e., no correlation of the packet loss across time steps. Heterogeneous loss probabilities are considered in only few previous works, e.g.…”