Angelo Moroncelli

Angelo Moroncelli working with a robot in the IDSIA robotics lab

Angelo Moroncelli

Ph.D. Researcher — Generative AI for Robotics

IDSIA USI–SUPSI · Lugano, Switzerland

I work at the intersection of generative modeling, reinforcement learning, and robotics. My research investigates how generative optimization can enable robotic agents to move beyond the imitation of skills learned from labeled demonstrations. I initially explored the out-of-distribution behavior of generative policies, investigating how they behave beyond their training distribution. More recently, I have been studying how generative priors, such as diffusion and flow-based policies and vision-language-action models, can be integrated with reinforcement learning to guide exploration and optimize behavior beyond the training data.

Google Scholar GitHub Email Download CV

Selected Research

Overview of VLA jump-starting for reinforcement learning robotic agents

Vision-Language-Action Jump-Starting for Reinforcement Learning Robotic Agents

IEEE ICRA 2026 Workshop on RL4IL

Using VLA priors to jump-start reinforcement-learning agents for robotic control.

arXiv

Comparison of RoboMorph Transformer and diffusion sequence models

Diffusion Sequence Models for Generative In-Context Meta-Learning of Robot Dynamics

IEEE CDC 2026

Generative sequence modeling for in-context adaptation and learning of robot dynamics.

arXiv

SynthLA framework for synthetic language-action policies and closed-loop robot control

SynthLA

IEEE/RSJ IROS 2026

Synthetic language-action policies for zero-shot real-world manipulation through structured perception.

Publications

Overview of the duality between generative AI and reinforcement learning in robotics

The Duality of Generative AI and Reinforcement Learning in Robotics

Information Fusion, 2025

A review of how generative AI and reinforcement learning complement each other in robotic systems.

Publications

View all publications →

Research Interests

Generative Policies

Diffusion models, flow-based models, and generative priors for robot learning.

Vision-Language-Action Models

Foundation models for grounding high-level semantics into robotic control.

Reinforcement Learning

Exploration, jump-starting, robustness, and data-efficient skill acquisition.

Robot Manipulation

Real-world experimentation, sim-to-real learning, and embodied decision-making.

Current Position

I am a Ph.D. researcher at IDSIA USI–SUPSI in Lugano, Switzerland, where I work on Generative AI for Robotic Systems.

I am supervised by Prof. Loris Roveda, with Prof. Alessandro Giusti and Prof. Luca Gambardella as advisors.