Dexterous manipulation
Multi-fingered robotic hands performing manipulation tasks that require coordinating many contacts at once, rather than a single grasp point.
robotics · real-time control · dexterous manipulation
Control that stays stable when robots touch the real world.
PhD researcher at IDSIA USI-SUPSI, in the LEON Group. I work on dexterous manipulation with multi-fingered robotic hands, designing real-time control and learning-based strategies for compliant multi-contact interaction.
Research
Contact is the hard part.
Manipulation and interaction break down exactly where the robot meets an uncertain environment. My work sits on that boundary: controllers that stay real-time and stable through contact, and learning that makes them general.
Compliant multi-contact interaction
Compliance control that lets a robot share space and forces with people and objects safely, under contact conditions that are neither known in advance nor static.
Real-time whole-body control
Modular C++ controllers that meet hard real-time constraints on redundant and floating-base platforms — mobile manipulators, quadrupeds carrying arms, redundant manipulators.
Teleoperation under uncertainty
Keeping teleoperated systems stable and usable when the channel is not: high latency and jitter of the kind lunar and space missions impose, and operator interfaces in mixed reality.
About
I am a PhD researcher in robotics at IDSIA USI-SUPSI, in the LEON Group, in Switzerland. My research is on dexterous manipulation with multi-fingered robotic hands: .
Before the PhD I spent about a year and a half as a robotics engineer on real hardware. At the Dynamic Legged Systems Lab of the Istituto Italiano di Tecnologia I built a modular whole-body controller for a quadruped carrying a redundant manipulator, and a real-time compliance controller for mixed-reality teleoperation. Before that, at the Automatic Control Group of the University of Salerno, I designed a ROS2 control framework for teleoperation under the latency and jitter conditions relevant to lunar and space missions, validated on Franka Emika Panda, UR10, and Husky A300 platforms.
A recurring part of that work was integration: defining the software interfaces between control, perception, and mixed-reality modules across several teams, then designing the experiments that show the whole system actually holds together on hardware. I care about the seam where clean control theory meets a robot that has to work in front of stakeholders.
Education
- MSc Computer Engineering, cum laude University of Salerno · Oct 2024 · GPA 29.58/30
- BSc Computer Engineering, cum laude University of Salerno · Jul 2022 · GPA 29.51/30
Experience
Where I have built things.
Apr 2026 — present
PhD Researcher
IDSIA USI-SUPSI · LEON Group
- Research on robotic dexterous manipulation with multi-fingered robotic hands
Nov 2025 — Mar 2026
Robotics Engineer
Istituto Italiano di Tecnologia · Dynamic Legged Systems Lab
- Designed and implemented a modular C++ whole-body controller meeting real-time constraints for a quadruped (Unitree B2) equipped with a redundant manipulator (Kinova Gen3)
- Developed a real-time compliance controller for teleoperation in mixed reality, enabling stable and safe robot-environment interaction under uncertain and dynamic contact conditions
- Led end-to-end system integration and tests across 3 teams, defining software interfaces across control, perception, and mixed reality modules; designed experiments proving robust performance to stakeholders
Nov 2024 — Nov 2025
Robotics Engineer
University of Salerno · Automatic Control Group
- Designed a modular C++ control framework in ROS2 meeting real-time constraints for teleoperation of robots under high-latency/jitter conditions relevant to lunar and space missions
- Developed a ROS2-based compliance controller for human-robot interaction
- Validated the framework on real hardware, including Franka Emika Panda, UR10, and Husky A300
- Designed and conducted experiments integrating simulation and physical testbeds for manipulators and rovers
- Facilitated Scrum ceremonies for a team of 4 engineers, improving project delivery and collaboration
Projects
Selected projects.
Locomotion, navigation, and perception work with code on GitHub.
2024
repo ↗
Humanoid Locomotion via Primitive Composition
A data-driven control algorithm that combines model-based RL primitives, trained in PyTorch, into robust bipedal locomotion in MuJoCo while avoiding obstacles. Roughly 90% success on held-out simulation trials, with sustained locomotion beyond 20 s in most successful runs — a baseline controller could not complete the same tasks.
2024
repo ↗
Autonomous Mobile Robot Navigation
A ROS2 navigation stack for TurtleBot 4 using QR-code signposts and Nav2 planners, with an FSM-based global planner, kidnapping recovery, and dynamic cost-map updates. The QR detection pipeline stayed reliable up to 3 m in motion under varied lighting; validated in Gazebo and on hardware, ranking 1st in race time among student teams.
2023
repo ↗
ONFIRE Contest 2023
Fine-tuned existing models and designed neural networks for fire and smoke detection on embedded cameras, reaching 84.4% mean validation accuracy — 5.4 percentage points above the FireNet baseline.
Publications
Publications.
Simplifying ROS2 Controllers with a Modular Architecture for Robot-Agnostic Reference Generation
IEEE Robotics and Automation Practice · 2026 IEEE Xplore ↗ Code ↗
Keeping the Franka Emika Panda alive: a ROS 2 stack with a reliable position interface
arXiv:2608.19740 · 2026 arXiv ↗ Code & videos ↗
Contact
Open to collaborations, chats, and good problems.
Working on dexterous manipulation? I would be glad to talk.