About me

I am an AI Engineer at Hexagon Robotics in Zurich, where I build locomotion and manipulation capabilities for the AEON humanoid – the first humanoid with wheels that can roll like a car and step like a human. My work spans reinforcement learning, vision-language-action (VLA) models, sim-to-real transfer, and scalable training infrastructure for real-world deployment in industrial environments.

Prior to Hexagon, I completed my PhD at EPFL where I pioneered the use of tensor networks for robot learning, and interned at Disney Research Studios working on locomotion for the BDX droid.

Current Work: AI for the AEON Humanoid

AEON -- the autonomous humanoid by Hexagon Robotics

At Hexagon Robotics, I am delivering end-to-end AI capabilities for the AEON humanoid platform:

Locomotion (RL)

• Developed RL-based models enabling robust locomotion across factory floors, stairs, and uneven terrain
• Deployed on physical humanoid systems for real-world industrial tasks

Manipulation (VLAs + RL)

• Built diffusion transformer-based VLA models from scratch for general-purpose manipulation
• Outperforming baselines (e.g., Hugging Face LeRobot) on benchmark and real-world industrial tasks
• Curated evaluation benchmarks and built frameworks to systematically assess VLA performance

Infrastructure

• Optimized training and inference pipelines using GPU acceleration and model optimization tools from PyTorch and NVIDIA
• Collaborated with Microsoft Azure and NVIDIA on large-scale training infrastructure

BMW Deployment

BMW became the first company to deploy humanoid robots in production in Germany, partnering with Hexagon Robotics.

BMW x AEON -- Humanoid deployment at BMW Leipzig plant

Disney Research Internship

Disney's BDX droids at Star Wars: Galaxy's Edge

During my internship at Disney Research Studios (Summer 2023), I developed novel deep reinforcement learning algorithms for concurrently learning a locomotion policy and state estimator, enabling robust sim-to-real transfer for Disney’s famous BDX droid.

Selected Impact

• ICLR 2024 Spotlight (top 5%) – Generalized Policy Iteration using Tensor Approximation for Hybrid Control
• Best Paper Award – IEEE RAS TC on Model-based Optimization for Robotics (2023)
• Idiap Paper of the Year (2021) – Ergodic Exploration using Tensor Train

Research Background: Tensor Networks

During my PhD at EPFL (2019–2024), I pioneered Tensor Networks as a computational paradigm for robot learning – developing fast and memory-efficient algorithms for exploration, learning from demonstration, reinforcement learning, and motion planning. See the Tensor Networks page for details and demos.

Technical Skills

• ML: Reinforcement Learning, Imitation Learning, Vision-Language-Action Models (VLAs), Diffusion Models
• Programming: Python, C++
• Frameworks: PyTorch, TensorFlow, JAX, Hugging Face
• Robotics & Simulation: ROS2, NVIDIA Isaac Sim, MuJoCo, Gazebo
• Infrastructure: Linux/Unix, Git, Nix, Docker, Microsoft Azure