German startup trades free meals for robot training data
german startup shift sends camera-wearing chefs into homes to record cooking tasks for training humanoid robots, highlighting the rising demand for high-quality egocentric data.
shift, a division of microagi, is recording human chefs preparing meals to create training datasets for future humanoid robots, specifically targeting complex dexterity tasks like knife work. by trading free labor for data, the company is building a marketplace for 'egocentric' data collection, where workers record their physical actions to train ai models. this approach signals a shift in robotics development, where real-world human demonstration data is becoming as critical as language data for large language models. the company believes this data is essential to bridge the gap between current robotic limitations and the goal of capable household helpers.