Ai+ Training

Robots are high-dimensional systems, and moving them in the real world can be harder than “playing chess” with them: our algorithms are often better at planning than at physically executing actions. Meanwhile, the human sensorimotor system has evolved over billions of years to generate efficient movement trajectories. If we can capture those trajectories at scale, we can train AI models that learn to manipulate the world more effectively.

This talk introduces Martino's work at Hugging Face’s LeRobot project to democratize teleoperation and embodied AI research through low-cost, open-source hardware. I will present the “Homunculus Glove,” a wearable exoskeleton designed to extract the maximum useful degrees of freedom from the human hand.

Attendees will learn how to leverage open hardware and software to collect datasets, design teleoperation interfaces, and contribute to a growing ecosystem of transparent, reproducible robotics research.