Research

Research Overview

My work lies at the intersection of generative modeling, reinforcement learning, and robotics.

The central question behind my research is:

How can generative optimization enable robotic agents to move beyond the imitation of skills learned from labeled demonstrations?

My research investigates how generative priors can be combined with reinforcement learning to guide exploration and optimize behavior beyond the training data. I initially explored the out-of-distribution behavior of generative policies, studying how they behave when deployed beyond their training distribution. More recently, I have focused on integrating generative priors—such as diffusion and flow-based policies and vision-language-action models—with reinforcement learning, allowing robotic agents to leverage pretrained behaviors while continuing to improve through interaction with the environment.

Main Research Directions

Generative Imitation Learning Priors

I study generative policies as behavioral priors for robot learning, including diffusion and flow-based policies and vision-language-action models. I am particularly interested in their generalization and out-of-distribution behavior beyond the demonstrations used for training.

Online Exploration and Optimization

I investigate how generative priors can be combined with reinforcement learning to guide online exploration. The goal is to move beyond pure imitation learning, enabling agents to optimize and acquire behaviors beyond their training data through interaction with the environment.

Real-World Robotics

I validate these ideas on real robotic systems, with a focus on manipulation and embodied decision-making. My work connects generative modeling and reinforcement learning with real-world experimentation, multimodal perception, and sim-to-real evaluation.

Research Experience

Ph.D. Researcher — IDSIA USI–SUPSI

2024–present · Lugano, Switzerland

Research on generative optimization and exploration for robotic learning, with emphasis on diffusion/flow policies, VLA models, reinforcement learning, generative in-context learning, and multimodal foundation models.

Visiting M.Sc. Researcher — IRIDIA, Université Libre de Bruxelles

2023 · Brussels, Belgium

Developed Byzantine-fault-tolerant Swarm-SLAM methods using blockchain-based smart contracts and geometric consensus.

Funded Projects & Industrial Collaborations

ACHEAS — Automatic Robotic Platform for the Aerospace Industry

2024–2026

Research contributor to the EUREKA Eurostars project, Grant Agreement No. 3616.

Worked on learning-based methods for industrial robotic manipulation and automation.

Klepsydra Technologies AG

2024

Industrial research collaboration on high-performance AI and robotics.