Our Team

Multi-Agent Systems

A multi-agent system consists of multiple entities that collectively act as a single system. Such systems appear in various fields, including engineering (e.g., mobile robots, drones), biology (e.g., flocks of birds), and IT (e.g., computer networks).

A clear example of a technical multi-agent system is a swarm of drones flying together.

Unlike a centrally controlled system, the agents (e.g., drones) do not rely on a central computing unit. Instead, each agent makes its own decisions independently. Communication occurs via a network, where agents exchange local information, such as position and velocity, only with their immediate neighbors.

Despite this decentralized behavior, multi-agent systems can collectively act intelligently and achieve global objectives.

Example Tasks for Multi-Agent Drone Systems

• Consensus: The drones synchronize by agreeing on a common flight altitude. The final altitude is not predetermined but emerges from local interactions—each drone adjusts its altitude based on its neighbors.
• Leader-Follower: The drone swarm (followers) follows a leading drone (leader), which follows a predefined altitude and flight path. The followers align behind the leader in a formation.
• Formation Control: The drones maintain a specific formation while flying a path, even under disturbances such as wind. A real-world example of this principle is drone light shows.

Research Area

To enable multi-agent systems to perform their tasks successfully, their overall system behavior must be analyzed mathematically. The mathematical description of a multi-agent system consists of several key components. First, the agent dynamics describe how each agent follows a nonlinear dynamic model. In some cases, the dynamics of different agents may vary, such as in systems that include both robots and drones. Second, the communication network is typically represented as a graph, where agents act as nodes and communication links form the edges. The structure of this network significantly influences the system’s behavior, as agents can only exchange information with their direct neighbors. Third, the control law determines how each agent adjusts its actions based on its own state and the states of neighboring agents. Lastly, additional effects such as communication delays (latency) must also be considered, as they can impact real-time decision-making and coordination.

Mathematical Methods Used

The study of multi-agent systems relies on several mathematical methods. Nonlinear control theory is essential for modeling the complex behavior of individual agents, while systems theory provides a framework for analyzing the collective dynamics of the entire system. Graph theory plays a crucial role in representing and studying the communication structure, as the connectivity of the network affects the coordination among agents. Optimization techniques are used to improve performance by finding the most efficient strategies for agent coordination. Machine learning methods allow agents to adapt and optimize their actions based on experience and environmental feedback. Additionally, model predictive control (MPC) enables agents to anticipate future states and make decisions accordingly. These methods, among others, provide a strong foundation for designing and analyzing intelligent multi-agent systems.