PhD researcher in robotics working on language-conditioned, long-horizon manipulation and multi-agent robotic systems.
My work focuses on building real systems that can understand and interact with the world in a human-like manner.
I also share some projects on YouTube to make robotics and Artificial Intelligence more accessible.
Turning ambiguous failures in long-horizon robot tasks into validated recovery supervision through counterfactual re-execution.
MAGMA-GEN started from a practical problem I kept encountering in long-horizon robotics: when an agent fails, the final error rarely tells us which earlier decision actually caused it. Yet failed rollouts often contain useful information.
This project was initiated during a robotics hackathon organized by REVEL and LycheeAI, with the objective of building a useful, manipulation-oriented robotic application.
This project was developed during a research internship at Ostfalia University of Applied Sciences (Germany). The objective was to design a complete robotic bartender system using a Sawyer manipulator, integrated with a Pepper robot acting as a waiter.