Model Predictive Control
Optimization-based feedback for constrained systems, with a focus on safety, robustness, and implementable guarantees.
Postdoctoral Researcher - EPFL
Danilo Saccani
DECODE group, EPFL - Lausanne, Switzerland
Building trustworthy autonomy with model predictive control, safe learning, and formal guarantees.
I am a postdoctoral researcher at EPFL in the DECODE group, working on control theory for reliable autonomous systems. My research is centered on Model Predictive Control and its extensions to safe learning, distributed decision-making, and neural controllers with formal guarantees.
I am particularly interested in how predictive control can become a practical backbone for trustworthy autonomy: systems that learn and adapt online, while still respecting safety, stability, and performance requirements. I enjoy connecting mathematical guarantees with algorithms that can actually run on autonomous systems, and I like explaining control ideas through simple geometric and visual intuition.
Research at a glance
Control methods for systems that learn, coordinate, and stay safe.
Distributed MPC
Predictive control methods for multi-agent systems that must coordinate under limited communication and changing networks.
Stable Neural Controllers
Neural control architectures that improve performance while preserving stability and robustness certificates.
Certificate Reservoirs
A research direction on certified flexibility: turning unused safety margin into a budget for reliable online adaptation.
Experimental motivation
From theory to experiments
My work is mainly theoretical and algorithmic, but I am strongly motivated by control methods that can eventually run on real autonomous systems. Over the years, I have been involved in experimental activities with drones, tethered aerial platforms, and small-scale autonomous vehicles, often in collaboration with students and colleagues.
These platforms are useful testbeds for studying how predictive control, safety filters, distributed coordination, and learning-based controllers behave when they meet real constraints, imperfect models, communication limits, and hardware implementation issues.
Latest highlights
Recent updates.
Published journal article
Our paper “Constrained Performance Boosting Control for Nonlinear Systems” is now published in Engineering Applications of Artificial Intelligence! We combine stability-by-design neural control with ADMM to improve performance while explicitly handling state and input constraints.
Workshops on DMPC at ECC26
The workshop I organized at ECC 2026 will present distributed MPC approaches for multi-agent systems and will take place on 7 July in Reykjavik, Iceland.
Safety-aware performance boosting
Safety-Aware Performance Boosting will be presented at ECC 2026 in Reykjavik, introducing a learning-based approach to guarantee stability and constraint satisfaction while improving closed-loop performance.
Online adaptation of neural maps
Online Adaptation of Neural Maps will be presented at ECC 2026 in Reykjavik, showing how neural controllers can be safely updated online while preserving closed-loop stability.