Robot Learning Researcher

Develop generative robot policies for robust and generalizable manipulation. Built DynaMimicGen, a Dynamic Movement Primitive-based data-generation framework that synthesizes manipulation datasets from one or two human demonstrations and improves downstream diffusion-policy performance in simulation and on real Franka Emika Panda hardware.

Developed DROM, a language-guided diffusion framework for learning and composing 17 manipulation skills into long-horizon instructions. Validated the approach on FANUC CRX-25iA and Franka Research 3 robots.

Conduct research on vision-based activity recognition for industrial human–robot collaboration, including a multimodal RGB-D dataset, transformer-based few-shot recognition, and integration with a Hierarchical Task Network planner. Contributed to the Horizon Europe Fluently project by developing and integrating ROS 2 collaborative robotic-cell capabilities, and supervised one MSc student.

Prepared and delivered MSc-level lectures, exercise sessions, and Python laboratories on diffusion models for approximately 100 students.

Covered forward and reverse diffusion processes, ELBO, DDPM, score-based models, conditioning, and applications of diffusion policies to robot learning. Created and maintained an open-source educational repository with theory notes and runnable code examples.

Taught BSc-level lectures and laboratory sessions on ROS 2, robot kinematics, and computer vision for 25 students.

Supervised hands-on robotics projects and graded coursework and assignments.

Taught first-year BSc physics lectures and laboratory sessions covering classical mechanics, fluids, and thermodynamics.

Prepared problem sets and quizzes, led written-exam preparation sessions, and graded assignments and examinations.

Thesis on diffusion-based generative models for robotic manipulation and human–robot interaction. Advisors: Luca Maria Gambardella, Anna Valente, and Stefano Baraldo.

Final grade: 106/110. Thesis: Implementation of Active Multi-Preference Learning for a Collaborative Robot with Ergonomics Assessment.