Giulia Benintendi

Ciao! I am a second-year master student, pursuing a double degree between the University of Brescia and Sorbonne University. I research multi-agent reinforcement learning, focusing on how social signals shape multi-agent coordination with the end goal of improving AI agents generalisation. Currently, research assistant in the ALPI Lab at the University of Zurich, advised by Prof. Giorgia Ramponi. Just at the beginning of my journey.

News

Sep 2026. Joined the ALPI Lab at University of Zurich as a research assistant, supervised by Giorgia Ramponi.
Apr 2026. Joined EuroTech Federation, a pan-European community of emerging deeptech talents and builders.
Mar 2026. Joined the Collaborative AI lab at University of Stuttgart, supervised by Andreas Bulling and Constantin Ruhdorfer, to work on multi agent reinforcement learning methods for zero shot coordination.
Sep 2025. Started Double Degree in Robotic Systems at Sorbonne University, Paris.
May 2025. 1st Place at the ISSA PULIRE AI-for-Cleaning Hackathon 2025 with AI-Orchestrator.
Mar 2025. Joined the Mechanical and Thermal Measurements group at University of Brescia for research on gaze trajectory analysis in preterm infants (VIPPSTAR project).
Sep 2024. Graduated B.Sc. in Industrial Automation Engineering, 110/110 cum laude, University of Brescia.

Selected Projects

Self-play, other-play, and MATE agents in the card game: MATE agents converge on the same card through mutual attention

Mutual Attention for zero-shot TEaming

Collaborative AI Lab @ University of Stuttgart

Mar 2026 – Aug 2026

Investigating how joint attention can support zero-shot coordination in cooperative multi-agent RL.

Preprint

Eye-contact timeline: manual annotation compared with GECO face probability and the Chong et al. score

Gaze Estimation for eye COntact detection

MMTLab @ UniBS · VIPPSTAR Project

Mar 2025 – Aug 2026

Developed an eye contact detection approach during naturalistic caregiver–infant interactions and assessed its accuracy.

Hopper, half cheetah and humanoid MuJoCo agents

Human-Robot Interaction Playground

Sorbonne University

Dec 2025

Benchmarked imitation learning algorithms (BC, DAgger, DAgger with replay buffer) on MuJoCo locomotion tasks.

GitHub

BIS kept in the 40-60 prescribed region from induction through surgery until the patient awakens

ML for Anesthesia Control

University of Brescia · ACTIVA Project

2024

Developed DL models to optimize anesthesia control systems by predicting patient's level of consciousness.