⚡ electrical engineering

Decentralized design of leader-following consensus protocols for asymmetric matrix-weighted heterogeneous multiagent systems

This paper proposes a decentralized design approach for leader-following consensus protocols in heterogeneous multiagent systems with asymmetric matrix weights and directed spanning trees, utilizing DST-based linear transformations and matrix diagonally dominant methods to ensure stability with either minimal or full neighbor information.

Original authors: Lanhao Zhao, Yangzhou Chen

Published 2026-03-18

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Lanhao Zhao, Yangzhou Chen

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a large orchestra where every musician plays a different instrument (some are violins, some are drums, some are flutes). They are all trying to play the exact same melody in perfect sync with the conductor (the "Leader"). However, there are two big problems:

The Instruments are Different: Because they are different, the sheet music and the way they respond to the conductor's baton are unique to each musician.
The Communication is Messy: In a real orchestra, musicians don't just listen to the conductor; they listen to each other. But in this paper's scenario, the "volume" and "direction" of that listening are weird. Sometimes Musician A listens to Musician B more than B listens to A, and the "volume" changes depending on which part of the melody they are playing. This is what the paper calls an "asymmetric matrix-weighted" system.

The paper by Zhao and Chen is essentially a recipe for a conductor (or a set of rules for the musicians) that allows this chaotic, mismatched group to eventually play in perfect unison, without needing a giant, central computer to tell everyone exactly what to do.

Here is the breakdown of their solution using simple analogies:

1. The Problem: Too Much Noise, Not Enough Clarity

In the past, researchers tried to get these groups to agree by having everyone talk to everyone else. But that creates a traffic jam. Also, most previous rules assumed that if A talks to B, B talks to A with the exact same "weight" (symmetry). The authors realized that in the real world, relationships are often one-sided or complex (asymmetric).

They asked: How can we get this diverse group to agree on a plan using the least amount of talking possible, and without a central boss calculating everything?

2. The Solution Part 1: The "Tree of Trust" (Minimal Communication)

The authors propose a clever trick called the DST-based Linear Transformation.

The Analogy: Imagine the group is a family tree. The Leader is the Grandparent. Some kids listen directly to the Grandparent. Others listen to their parents, who listen to the Grandparent.
The Trick: Instead of everyone shouting their opinions to everyone else, the paper suggests that each person only needs to listen to one specific person in their "line of trust" (their parent in the tree).
Why it works: If the tree is connected (everyone is linked back to the Grandparent eventually), you don't need a mesh network. You just need the branches. This is called the "Minimal Communication Links" protocol. It saves energy and reduces confusion.

3. The Solution Part 2: The "Decentralized Tuning" (Designing the Gains)

Now, how do we tell each musician how to adjust their instrument? Usually, you need a master sheet that lists everyone's stats to calculate the right settings. That's slow and hard for big groups.

The authors invented a Decentralized Design Method.

The Analogy: Think of it like tuning a guitar. Instead of a master tuner coming to every guitar, each musician is given a simple rule: "Look at your own string tension and the tension of the person you are directly connected to. Adjust your tuning peg until you match them."
The Magic: The paper proves that if everyone follows this local rule, the whole orchestra will eventually tune itself perfectly. They don't need to know who the 50th violinist is; they only need to know their immediate neighbor.

4. The Solution Part 3: The "Safety Net" (Using All Neighbors)

What if the group wants to talk to everyone, not just their parent in the tree? Maybe they want to be extra sure.

The authors extended their method to handle this too. They used a mathematical concept called "Diagonal Dominance" (think of it as a "Safety Net").

The Analogy: Imagine a tightrope walker. If they only hold one hand (the tree method), they are stable. If they hold hands with everyone around them, they are even more stable, but it's harder to calculate the balance.
The Trick: The authors created a rule that says, "As long as your own strength (your own adjustments) is strong enough to overpower the combined noise of everyone else pulling you in different directions, you will stay on the rope." This allows them to use all neighbor information while still calculating their settings locally.

5. The Result: Harmony from Chaos

The paper runs computer simulations (like a video game) to prove this works.

They created a group of 5 robots with different engines (heterogeneous).
They gave them weird, one-way communication weights (asymmetric).
They applied their "Tree" and "Local Tuning" rules.
The Outcome: The robots started at different positions and speeds, but within seconds, they all moved in perfect lockstep with the Leader.

Summary

This paper is about teaching a diverse, messy group to work together efficiently.

Old Way: Everyone talks to everyone; a central brain calculates the plan; assumes relationships are perfectly fair (symmetric).
New Way (This Paper): Everyone only listens to their "parent" in a chain (or uses a safety net if they talk to everyone); everyone calculates their own settings based only on local info; works even if relationships are unfair or one-sided.

It's a blueprint for making self-driving cars, drone swarms, or power grids work together smoothly, even when they are all built differently and the communication lines are messy.

1. Problem Statement

The paper addresses the leader-following consensus problem for heterogeneous multi-agent systems (MASs) under two specific and challenging conditions:

Heterogeneity: The agents possess different dynamic models (different state-space matrices $A_i, B_i$ ).
Asymmetric Matrix-Weighted Topology: The communication weights between agents are represented by asymmetric matrices rather than scalar values or symmetric matrices. This allows for complex interactions where different state components of one agent influence different components of another with varying weights.

Key Challenges:

Decentralized Design: Existing methods often require global information (e.g., the full Laplacian matrix or eigenvalues of the network) to design feedback gains, which is impractical for large-scale systems.
Communication Efficiency: Standard protocols often utilize information from all neighbors, leading to high communication loads.
Stability Analysis: Asymmetric weight matrices complicate stability analysis, as traditional methods often rely on symmetric or positive/negative definite assumptions.

2. Methodology

The authors propose a decentralized design framework based on Directed Spanning Tree (DST) properties and linear transformation.

A. System Modeling

The system consists of one leader (node 0) and $N$ followers.
Dynamics are linear time-invariant but heterogeneous: $\dot{x}_i = A_i x_i + B_i u_i$ .
Communication is modeled by a directed graph with an asymmetric block weight matrix $W$ .
A DST is assumed to exist with the leader as the root, ensuring the leader's information can reach all followers.

B. Core Transformation Technique

The paper utilizes a DST-based linear transformation to convert the consensus problem into a partial variable stability problem.

Transformation: An incidence matrix $P_0$ of the DST is constructed. A linear transformation $P$ is applied to the system states, separating the leader's state from the relative errors.
Partial Stability: The consensus condition ( $\lim_{t\to\infty} \|x_i - x_0\| = 0$ ) is proven to be equivalent to the global asymptotic stability of the transformed error system's zero equilibrium point.
Auxiliary System: The stability of the transformed system is analyzed using an auxiliary system involving matrices derived from the observability of the system. The consensus is achieved if specific block matrices (derived from the system dynamics and gains) are Hurwitz stable.

C. Two Protocol Designs

The paper proposes two distinct protocols with decentralized gain design:

1. Protocol with Minimal Communication Links (DST-based)

Strategy: Each agent $i$ uses only its own state and the relative state of its parent node on the DST ( $k_i$ ).
Control Law: $u_i = -G_i x_i + K_i W_{i,k_i} (x_{k_i} - x_i)$ .
Design Logic:
- The gain $G_i$ is designed first to ensure the stability of a specific block matrix ( $L_1 \bar{D} L_3$ ) related to the heterogeneous nature of the agents.
- The gain $K_i$ is then designed locally to ensure the stability of the diagonal blocks ( $A_i^* - B_i^*$ ) of the system matrix.
- Decentralization: Since $K_i$ depends only on agent $i$ 's parameters and its parent's weight, no global network information is needed.

2. Protocol Using All Neighbor Information

Strategy: Agents utilize information from all neighbors in the original topology.
Control Law: $u_i = -G_i x_i + K_i \sum_{j \in N_i} W_{i,j} (x_j - x_i)$ .
Design Logic:
- The authors extend the decentralized approach using the Gerschgorin Circle Theorem for block matrices.
- They derive conditions where the Gerschgorin circles of the system matrix must lie in the open left half-plane.
- This allows each agent to calculate its gain $K_i$ based on local parameters and the weights of its immediate neighbors, avoiding the need to solve global matrix inequalities.

3. Key Contributions

Relaxation of Weight Assumptions: Unlike most existing literature that assumes symmetric or definite weight matrices, this work handles generic asymmetric matrix weights, significantly broadening the applicability to real-world networks.
Minimal Communication Protocol: The paper introduces a "minimal communication links" protocol where agents only communicate with their DST parent. This reduces communication load and simplifies the network structure required for consensus.
Fully Decentralized Gain Design:
- The method allows each agent to design its own unique feedback gain matrix ( $G_i, K_i$ ) without global information.
- This is a significant improvement over previous works that either required global information or forced all agents to use identical gains (which limits flexibility for heterogeneous systems).
Theoretical Framework: The combination of DST-based linear transformation and partial variable stability theory provides a rigorous mathematical foundation for analyzing asymmetric matrix-weighted heterogeneous systems.

4. Results

The theoretical findings were validated through numerical simulations involving a 4-follower heterogeneous system with asymmetric matrix weights:

Example 1 (Minimal Links): Demonstrated that using only DST parent information, the state errors converged to zero. The decentralized calculation of $K_i$ was successful.
Example 2 (Impact of $G_i$ ): Showed that the gain matrix $G_i$ (associated with self-feedback) directly influences the convergence rate of the consensus process.
Example 3 (All Neighbors): Verified the protocol using all neighbor information. The Gerschgorin circle condition was satisfied, and the system achieved consensus, proving the effectiveness of the decentralized design even with full topology usage.

5. Significance

Scalability: By enabling decentralized design, the approach is scalable to large networks where gathering global topology data is impossible or too costly.
Flexibility: Allowing asymmetric matrix weights makes the model more realistic for applications like formation control, sensor networks, and power grids where interactions are directional and multi-dimensional.
Efficiency: The "minimal communication links" protocol offers a practical solution for bandwidth-constrained environments, proving that full network connectivity is not strictly necessary for consensus if a DST exists.
Theoretical Advancement: This is the first work to provide a decentralized design method specifically for heterogeneous MASs with asymmetric matrix weights, bridging a gap between matrix-weighted network theory and heterogeneous control.