Distributed MPC for
Multi-Agent Systems

European Control Conference 2026

Reykjavík, University of Iceland

July 7, 2026

Multi-agent systems (MAS) are rapidly evolving from theoretical constructs to large-scale applications in robotics, logistics, and cyber-physical infrastructures. Examples such as warehouse automation, drone swarms, and smart grids show the need for control architectures that address safety constraints, network limitations, and real-time computation.

While Model Predictive Control (MPC) provides a principled framework for constrained optimization, centralized approaches become computationally intractable and fragile as the number of agents grows. Distributed MPC (DMPC) addresses this by decomposing the global problem into local subproblems coordinated via communication; however, bridging the gap between theoretical stability guarantees and practical, real-time implementation remains a significant challenge.

This workshop offers a comprehensive view of the state-of-the-art in DMPC, structuring the landscape into three complementary modules:

Foundations of Distributed Optimization and Estimation, moving beyond classical consensus schemes to explore sensitivity-based decomposition and the often-overlooked challenge of Distributed Moving Horizon Estimation (MHE).
Learning-Accelerated DMPC, examining how data-driven warm-starting and learning-based policies can drastically reduce computational overhead while maintaining verifiable safety certificates.
Contracts and Coordination Primitives, showing how global constraints, such as collision avoidance and connectivity, can be enforced locally via set-based contract abstractions or offline contingency rules when communication is unreliable.

By combining methodological rigor with implementation-oriented insights, the workshop aims to define the roadmap for the next generation of scalable, safe, and efficient multi-agent control systems.

Registration: https://ecc26.euca-ecc.org/conference-registration/

Time and venue: July 7, 2026 09:00 AM (Local Time) | Reykjavík, University of Iceland.

Invited Speakers

Andrea Carron

Danilo Saccani

Giuseppe Notarstefano

Matthias A. Müller

Pierer von Esch Maximilian

Timm Faulwasser

Azita Dabiri

Georg Schildbach

Max Studt

Schedule

Module 1 - Foundations of Distributed Optimization and Estimation

Real-world MAS face a dual challenge: handling complex (often unknown) couplings between agents and solving the resulting optimization problems efficiently. This module revisits the algorithmic core of DMPC to address these hurdles. We explore hierarchical and data-driven frameworks to handle uncertain or aggregative interactions, and introduce sensitivity-based decomposition methods that improve scalability beyond standard duality-based schemes. Crucially, we also bridge the gap between control and sensing by integrating Distributed Moving Horizon Estimation (MHE), ensuring that robust state information is available to drive the control loop.

Module 2 - Learning-Accelerated DMPC with Certificates

This module comprises two sessions, the details of which are outlined below. It is structured to provide a comprehensive framework for integrating machine learning into the control loop, specifically to reduce computational complexity and enable real-time operation without sacrificing theoretical stability or feasibility guarantees.

Module 3 - Contracts and Coordination Primitives

This module comprises two sessions on distributed MPC for multi-agent systems, moving from iterative distributed optimization methods to contract-based and communication-free coordination approaches. The focus is on how tasks such as coverage, collision avoidance, and connectivity maintenance can be addressed under nonlinear dynamics, limited local information, and possibly time-varying or absent communication.

Organizers:

Andrea Carron (carrona@ethz.ch)
Danilo Saccani (danilo.saccani@epfl.ch)

The workshop is sponsored by the NCCR Automation.

Distributed MPC for Multi-Agent Systems