Robotics Technical Lead & Product Owner at Generative Bionics
Humanoid locomotion/whole-body control/learning
I turn advanced humanoid robotics research into working robotic products — from reinforcement-learning policy design and model predictive control to simulation, deployment and on-robot tuning.
Background: “Cosmic Cliffs” in the Carina Nebula · NASA, ESA, CSA, STScI
I'm Giulio Romualdi, Robotics Technical Lead and Product Owner at Generative Bionics, where I lead the development and delivery of locomotion and whole-body control technologies for humanoid robots. I own the technical roadmap, priorities, backlog and delivery criteria for the locomotion team, drive technical decision-making, and coordinate execution toward product objectives.
I mentor engineers and researchers, contribute to hiring and candidate evaluation, and work cross-functionally with other teams — while staying hands-on across the humanoid locomotion stack, from reinforcement-learning policy design and motion generation to simulation, deployment and on-robot tuning.
In 2026, I led the locomotion delivery for GENE.01, contributing to its public live demonstration at AMD Advancing AI 2026. Previously, at the Istituto Italiano di Tecnologia, I worked on locomotion, whole-body control and teleoperation for iCub, iCub3 and ergoCub.
I led the locomotion delivery for GENE.01, contributing to its public live demonstration at AMD Advancing AI 2026 — reinforcement-learning locomotion running on the full-size humanoid platform, in front of a live audience.
Product ownership and technical leadership, backed by hands-on expertise across the humanoid locomotion stack.
A few projects that show the full path from research to delivered robotic capability — what the problem was, what I contributed, and where to see the result.
Bringing reinforcement-learning locomotion to a full-size humanoid. I led the locomotion delivery — RL policy design, motion generation, simulation and on-robot tuning — culminating in a public live demonstration at AMD Advancing AI 2026.
The C++/Python backbone for planning, estimation and control of humanoid locomotion. I lead its development and maintenance; it underpins both published research and behaviors shipped on real robots.
Adversarial Motion Priors for legged-robot reinforcement learning: PPO with motion imitation that turns reference and human motion into natural, deployable locomotion policies.
A learned trajectory generator combined with model-predictive step adjustment for human-like, disturbance-robust walking on ergoCub. First author; Finalist, Best Oral Presentation Award, IEEE-RAS Humanoids 2024.
Large-system humanoid integration: I contributed the locomotion and teleoperation integration enabling fully immersive remote embodiment of a humanoid robot across a 300 km link.
Learning human-like whole-body trajectory generators for humanoid control. Co-author; the work received a Runner-up Best Poster Award at NAVER LABS Europe's AI for Robotics workshop.
Evidence that I ship usable robotics software, not only prototypes. I lead or maintain the libraries below, used across research and real-robot deployment.
A suite of C++/Python libraries for planning, estimation and control of bipedal locomotion on humanoid robots. The backbone of my research and products.
A simple, widely used Eigen C++ wrapper around the OSQP quadratic programming solver.
DCM-based bipedal locomotion software for the iCub humanoid robot platform.
Adversarial Motion Priors for robotic reinforcement learning: PPO with motion imitation for legged robots.
An interactive visualizer to inspect and replay data logged from robot experiments.
A header-only C++ library of controllers designed to operate directly on Lie groups.
Peer recognition and professional service that back up the technical track record.
Finalist — Best Oral Presentation Award for Online DNN-driven Nonlinear MPC for Stylistic Humanoid Robot Walking with Step Adjustment. Also on the organizing committee of the Designing Interactive Humanoids workshop.
Contributed to the iCub3 Avatar System as part of the ANA Avatar XPRIZE Finalist iCub team, integrating humanoid locomotion with the teleoperation system.
ADHERENT received a Runner-up Best Poster Award at the AI for Robotics workshop (team recognition, co-author).
Deep robotics research, growing technical leadership, and delivery on real humanoid platforms. Full details on the experience page.
Lead the development and delivery of humanoid locomotion and whole-body control; own the roadmap, priorities and delivery criteria; drive technical decisions; mentor engineers and researchers; contribute to hiring. Led the locomotion delivery for GENE.01 (AMD Advancing AI 2026).
Istituto Italiano di Tecnologia · Artificial and Mechanical Intelligence
Humanoid locomotion, whole-body control and locomotion architectures for iCub, iCub3 and ergoCub. Mentored PhD students and researchers, and served as IIT technical point of contact for the ergoCub project (IIT–INAIL collaboration).
University of Genoa & Istituto Italiano di Tecnologia
Thesis: Online Control of Humanoid Robot Locomotion — nonlinear and model predictive control, optimization and whole-body robotics, validated on iCub, iCub3 and ANA Avatar XPRIZE activities. Visiting PhD student at LAAS-CNRS, Toulouse (TALOS).
University of Pisa · 110/110 cum Laude
Thesis on capture-point-based controllers for bipedal locomotion, implemented on the iCub platform.
University of Pisa