Academic Project Page

The emergence of autonomous skills in humanoid robots requires a deep integration of multimodal sensory streams and motor control. We present \textbf{LeRobot-iCub}, an open-source framework designed for the study of \textbf{sensorimotor learning and active perception} through scalable VR teleoperation. By bridging the iCub humanoid—a platform rooted in cognitive developmental robotics—with the LeRobot ecosystem, our system enables the generation of contact-rich datasets in MuJoCo or Gazebo, and real-world environments. Our framework currently captures high-fidelity proprioception and tactile feedback from the robot's hands, providing a foundation for training Vision-Language-Action (VLA) models in complex bimanual tasks. The simulation environment and the generated datasets facilitate research in sensorimotor learning, active perception and cognitive developmental robotics by providing a low-cost pipeline for human-guided data collection.