About me

I am currently a first-year PhD student in Robotics and Intelligent Machines at the Italian Institute of Technology (IIT) in Genoa, within the Humanoids and Human-Centered Mechatronics Lab, under the supervision of Nikolaos Tsagarakis.

My research focuses on embodied robot autonomy, aiming to equip legged and hybrid wheeled–legged systems with the ability to reason about their environment while acting in it. I investigate how foundation models can support semantic understanding, task grounding, and action generation, and how these components can be combined into robust navigation and manipulation systems. A central aspect of my work is the integration of VLMs and VLA models with failure-aware control strategies for reliable real-world operation.

During my Master’s studies in Artificial Intelligence and Robotics at Sapienza University of Rome, which I completed with honors, I carried out my thesis in collaboration with Leonardo S.p.A., under the academic supervision of Prof. Giuseppe Oriolo and the industrial supervision of Dr. Navvab Kashiri. My thesis, “Learning Distance Functions for Robot Obstacle Avoidance”, investigated neural distance functions for obstacle avoidance, starting from geometrically constructed datasets and studying their use as differentiable representations for motion generation in mobile manipulators.

My research interests include foundation models for robotics, vision–language–action learning, and failure-aware navigation and manipulation.

Education

Ph.D. in Robotics and Intelligent Machines

Italian Institute of Technology (IIT), Genoa

Humanoids & Human-Centered Mechatronics (HHCM) Lab

Nov 2025 – Present
MSc in Artificial Intelligence and Robotics (with honors)

Sapienza University of Rome, Rome

Sep 2023 – Oct 2025 · Graduation grade: 110/110 cum laude
BSc in Applied Computer Science and Artificial Intelligence

Sapienza University of Rome, Rome

Sep 2020 – Oct 2023 · Graduation grade: 110/110 cum laude
Scientific High School Diploma

I.I.S. “Galilei Vetrone”, Benevento

Sep 2015 – Jul 2020 · Graduation grade: 100/100 cum laude

Experience

PhD Researcher

Humanoids and Human-Centered Mechatronics Lab, IIT

Nov 2025 – Present
Robotics Research Internship

Leonardo S.p.A., Genoa

Apr 2025 – Oct 2025
Computer Vision Research Internship

VisionLab, Sapienza University of Rome, Rome

Jun 2023 – Oct 2023

Thesis
Work

Master’s Thesis — Learning Distance Functions for Robot Obstacle Avoidance

Sapienza University of Rome × Robotics Lab Leonardo S.p.A.

This thesis investigates data-driven distance representations for robot obstacle avoidance in mobile manipulation. A geometric dataset is first constructed by sampling collision-free robot configurations and corresponding workspace obstacle points, from which exact distance values are computed. These distances are then approximated through neural models, enabling smooth and differentiable distance functions. The learned representations are designed to support gradient-based control, with the long-term goal of integrating distance constraints into quadratic programming formulations for real-time motion generation of mobile base manipulators.

Slides

Robot Learning Obstacle Avoidance Neural Distance Fields Mobile Manipulation Gradient-Based Control

Bachelor’s Thesis — Unmasking Deception: A Deep Learning Approach with Attention Mechanisms

VisionLab, Sapienza University of Rome, Rome

This thesis investigates deception detection from facial micro-expressions using a deep learning approach. A unimodal architecture combining an autoencoder for temporal sequence reconstruction and an attention-based classifier is introduced to identify salient facial dynamics associated with deceptive behavior. The method is evaluated on a real-life trial dataset of high-stakes interview videos, featuring both truthful and deceptive subjects, with facial features extracted through OpenCV-, Dlib-, and MediaPipe-based pipelines.

Slides

Facial Micro-Expressions Attention Mechanisms Autoencoder-Based Models Computer Vision Deception Detection

Selected
Projects

TIAGoCare: Emotion-Aware Interaction for Assistive Robotics

ROS1-based assistive robotics system for emotion-aware human–robot interaction. The project integrates real-time multimodal emotion recognition from face, audio, and body pose with semantic scene understanding and knowledge-graph-based symbolic reasoning, enabling adaptive and socially appropriate robot behavior in healthcare and assistive scenarios.

Code Report

ROS1GazeboPyTorchTensorflow Knowledge GraphsHRIAssistive RoboticsMultimodal Perception

Trajectory Optimization for Humanoids via Centroidal Dynamics

Trajectory optimization framework for humanoid robots based on a stiffness-based formulation of centroidal dynamics, modeling CoM motion and angular momentum in multi-contact scenarios. The problem is formulated as an optimal control problem with contact and friction constraints and solved using CasADi and IPOPT, with validation in simulation on the HRP-4 humanoid robot.

Code Slides

CasADiIPOPTDART HumanoidsCentroidal DynamicsOptimal ControlMulti-Contact Planning

Model-Based Offline RL with Trajectory Pruning

Model-based offline reinforcement learning approach using an ensemble of autoregressive dynamics models for trajectory rollouts. Uncertainty-aware trajectory pruning combined with Q-function evaluation is employed to mitigate out-of-distribution actions, with performance evaluated on MuJoCo benchmark environments.

Code Slides

MuJoCoPyTorchOffline RLModel-Based RLTrajectory Pruning

ClipFusion: Cross Modal Transformer for Video Question Answering

Cross-modal transformer architecture for multiple-choice Video Question Answering, integrating BERT-based textual representations with CLIP-extracted video embeddings. Multimodal alignment is achieved through attention mechanisms and a dual-stage decoding process that sequentially models question–video and answer interactions.