Matteo Cercola

Matteo
Cercola

Reasoning and control in vision–language–action models.

PhD candidate, Politecnico di MilanoVisiting researcher, Stanford ASL
Matteo Cercola
Selected workThree papers under review, 2026
01 — About

Does the reasoning drive the action?

I study how robot foundation models reason, and whether that reasoning actually controls what they do.

I am a PhD candidate in Information Technology at Politecnico di Milano, advised by Prof. Simone Formentin, and currently a Visiting Student Researcher at Stanford University's Autonomous Systems Lab with Prof. Marco Pavone.

My recent work looks at vision–language–action (VLA) models for driving and manipulation: whether their chain-of-thought is faithful to the policy, how to represent reasoning as a distribution over plans that can be sampled and steered, and why a model can perceive a hazard yet fail to act on it. I come to this from control and learning from human feedback, and I care about learning-based systems that are interpretable, steerable, and safe.

02 — News

Recent

03 — Research

Three threads

01

Reasoning in VLA models

Does a model's reasoning drive its actions? Faithful chain-of-thought, plans as steerable distributions, and the gap between perceiving a hazard and braking for it.

02

Learning from human feedback

Sample-efficient preference learning, from RLHF for language models to human-in-the-loop calibration of vehicle controllers.

03

Generative models for control

Diffusion models that design stabilizing controllers, and LLM agents with optimization-backed decisions for highway operations.

04 — Publications

Papers

VLA reasoning · New
Matthew Foutter*, Matteo Cercola*, Lena Wild, Yunshan Wang, Michelle Li, Daniele Gammelli†, Marco Pavone†
*Equal contribution · †Equal advising
Under review at IEEE Robotics and Automation Letters, 2026
▶ clip
VLA reasoning · New
Matteo Cercola, Katie Z. Luo, Milan Ganai, John Alora, Simone Formentin, Marco Pavone
Under review at ICLR 2027 · arXiv coming soon
▶ clip
VLA reasoning · New
Matteo Cercola, Katie Luo, Simone Formentin, Marco Pavone
Under review at ICRA 2027 · arXiv coming soon
Human feedback
Matteo Cercola, Valeria Capretti, Simone Formentin
IFAC Journal of Systems and Control, 2026
Human feedback
Matteo Cercola, Michele Lomuscio, Dario Piga, Simone Formentin
IFAC Journal of Systems and Control, 2026
Generative control
Matteo Cercola, Donatello Materassi, Simone Formentin
IFAC World Congress, 2026
LLM agents
Matteo Cercola, Nicola Gatti, Pedro Huertas Leyva, Benedetto Carambia, Simone Formentin
7th Learning for Dynamics & Control Conference (L4DC), Ann Arbor, 2025
Vehicle dynamics
Method for Estimating Expert Test-Driver Subjective Handling Assessment
Matteo Cercola et al.
Patent — Maserati S.p.A. and Politecnico di Milano
05 — Projects

Systems I've built

Foundation models for autonomous driving — Stanford ASL

Fine-tuning 4B-parameter vision-language-action models for driving on a 32× H100 cluster with PyTorch distributed training.

Agentic LLM decision support — MOVYON / Autostrade per l'Italia

An agentic decision-support system (LangChain, RAG) with optimization-backed guarantees, deployed for highway radio operators to accelerate incident response.

LiDAR segmentation — Indy Autonomous Challenge 2023

Real-time point-cloud segmentation (PyTorch + ROS2) deployed on a full-scale autonomous race vehicle, contributing to the team victory at the Monza Circuit.