Projects and Publications

ZiMPedance: Impedance-Aware ZMP Modeling and Control for Payload Carrying with Quadruped Robots

  • Core Challenge: Using passive spring-based arms for load transportation reduces a quadruped's weight and complexity, but their spring-damper dynamics introduce oscillatory forces that severely degrade locomotion stability.
  • The Approach: I derived an extended Zero Moment Point (ZMP) formulation that mathematically relates stiffness, damping, and payload mass to the stability margin. Using this insight, I augmented a Single Rigid Body Dynamics (SRBD) model with the passive subsystem dynamics and integrated it directly into a Model Predictive Control (MPC) framework.
  • Results: The proposed controller reduced stability violations by up to 10× and increased locomotion efficiency by 15% in simulation. During real-world hardware experiments, the robot successfully maintained stable locomotion with a 2kg payload under heavy pull-release disturbances where the nominal baseline controller completely failed.
  • Tech Stack: C++, Python, Model Predictive Control (MPC), SRBD, ZMP Modeling, Hardware-in-the-Loop Testing
Read Paper

Load-bearing Assessment for Safe Locomotion of Quadruped Robots on Collapsing Terrain

  • The Goal:This work involved traversing no-mercy terrains, where one misstep could be critical. The work involved the use of motion planning, Model Predictive Control, terrain probing and load-bearing analysis..
  • My Contribution: I was involved in all aspects of the projects, including designing and building the wooden platform used for the experiments. On top of that I took care of running the real robot experiments on my own. I took care of integrating the controller on the propertary dls framework base on ROS 1. I tuned the mpc on the robot, added the mapping frameowrk base on the eth elevation map.
  • Tech Stack: ROS, C++, Python, Trajectory Optimization, MPC, Elevation Mapping, Foothold Adaptation, Real Hardware Integration
Read Paper

VINUM Project

  • Core Challenge: The main goal of the VINUM is to develop robotics solution for grapevine winter pruning automation.
  • My Contribution: I worked across multiple parts of the project, developing mobile navigation for guiding a platform between orchard pots and integrating MoveIt with ROS 1 to enable motion planning with a floating-base system. I also improved the manipulation framework to handle more challenging cutting scenarios. During the project, I worked with real hardware—including a Summit XL, the HyQReal robot, and a Kinova arm with a custom end effector—and participated in five field tests, gaining hands-on experience with real-world integration challenges.
  • Tech Stack: ROS, C++, Python, Motion Planning, Control, MoveIT, Computer Vision, Mapping and Localization, Sensor Integration, Field Testing
Read Paper

Training humanoid locomotion in load carrying tasks.

Learning stable humanoid locomotion through reinforcement learning in a high-dimensional physics simulation.

  • Overview: Developed a custom reinforcement learning pipeline for humanoid locomotion using NVIDIA Isaac Lab. The project focuses on learning stable and efficient walking behaviors by combining physics-based simulation with policy optimization. The environment includes high-dimensional observations (joint states, velocities, orientation) and continuous control over multiple degrees of freedom.
  • My Contribution: Designed and implemented the full RL environment from scratch, including observation space, action scaling, and reward functions. Built a custom reward structure combining progress, stability (upright posture), energy efficiency, and motion smoothness. Integrated PPO-based training and handled debugging of instability, NaNs, and control issues in high-DOF systems.
  • Tech Stack: Python, PyTorch, NVIDIA Isaac Lab, RL
Humanoid locomotion reinforcement learning project preview

ETH ROBOTICS CLUB HACKATON 2026

A humanoid robot that can follow you, understand your gestures, and respond to voice commands — built in just 48 hours.

  • Overview: Built an interactive human–robot system in a 48-hour hackathon using the Unitree G1 humanoid platform with the GROOT whole body controller \. The robot is capable of real-time human following, gesture interpretation, and voice-command interaction, combining perception, control, and AI into a unified behavior pipeline.
  • My Contribution: Focused on system integration and deployment, connecting perception modules to robot control via ROS2 . Worked on software setup, debugging, and real-world testing, ensuring reliable end-to-end performance under tight time constraints.
  • Tech Stack: ROS2, C++, Python, Whole-Body Control, Computer Vision, Human-Robot Interaction, Rapid Prototyping, System Integration, Testing & Debugging

Other Activities

During my PhD I also took part in different activities:

  • Machine Learning Crash Course, MALGA, University of Genova (2024)
  • Optimization for Robotics Summer School, University of Patras (2025)
  • Dissemination Activities,Exposcuola 2024, Robot Valley 2025, Mechatronics Days PoliTo 2025, demos and workshops