Observer-Based Control for Multi-Agent Systems under Communication-Task Graph Mismatch

QC 20261104

Abstract

The growing deployment of robotic platforms in complex and large-scale environments has shifted control system design from centralized architectures toward distributed and cooperative frameworks. In these systems, agents coordinate their actions based on locally available information and information received from their communication neighbors. The effectiveness of such coordination depends not only on the agents' individual dynamics and control laws, but also on the availability of the state information required by the collaborative task objective. In particular, the states relevant to a task may belong to agents with which no direct 1-hop communication link exists. This discrepancy between task requirements and available communication links, referred to as a communication-task graph mismatch, poses a fundamental challenge for distributed control.

To address this challenge, this thesis develops decentralized observer-based control methods for multi-agent systems, in which each agent locally reconstructs the missing information about agents beyond its communication neighborhood using only 1-hop communication. The resulting estimates are used in the control laws in place of the unavailable states, allowing the original task to be pursued over the existing communication network. To account for the effect of estimation errors on the closed-loop system, each observer design establishes bounds on the estimation errors. These bounds are then used to derive conditions under which the collaborative tasks are achieved despite the estimation errors, for objectives ranging from consensus to spatio-temporal specifications. The first part of the thesis introduces a distributed k-hop observer for homogeneous multi-agent systems, which allows each agent to estimate, in finite time, the states and inputs of the agents in its k-hop neighborhood, i.e., located between two and k communication hops away. The observer is then reformulated within the prescribed performance framework, yielding k-hop Prescribed Performance Observers whose estimation errors evolve within funnels specified a priori. The framework is further extended to heterogeneous agents subject to bounded external disturbances, without requiring gain tuning or global network information. In each case, set Input-to-State Stability (set-ISS) of the estimation-based closed-loop system characterizes how estimation errors affect the achievement of the team objective.

The second part addresses spatio-temporal objectives expressed in Signal Temporal Logic (STL), which may involve task dependencies beyond the direct communication neighborhoods. The estimation-error bounds are incorporated into a reformulation of the spatial robustness of the specifications, accounting for the worst-case estimation errors. Based on this reformulation, a decentralized continuous-time feedback controller is designed to guarantee satisfaction of the STL specifications, including during the observer transient.

The final part extends the observer-based control framework to directed communication graphs with switching topologies, under which agents may enter or leave the k-hop neighborhoods at switching instants. To preserve prescribed performance, auxiliary funnels are constructed online at each switch, and conditions ensuring their feasibility are established together with a lower bound on the dwell time. As in the previous parts, set-ISS of the estimation-based closed-loop system characterizes the effect of estimation errors on the distance from the team objective. The effects of communication delays and packet losses, together with the experimental validation of the proposed methods, are discussed as directions for future research.

Link to DiVA