Humanoid Data (www.technologyreview.com)

A new trend in the robotics industry is emerging as companies are turning to real-world data on human movement to train humanoid robots. Inspired by the success of large language models like ChatGPT, which utilize vast amounts of training data, robotics developers believe that capturing and analyzing human actions can enhance the dexterity and capabilities of humanoid robots, making them more suitable for tasks currently performed by humans. This shift comes as significant investments flood into robotics, with projections of $6.1 billion for humanoid development by 2025. To gather movement data, various methods are being employed, from individuals wearing cameras while completing household tasks to advanced setups involving exoskeletons and virtual reality to ensure accurate and repetitive motion data collection. While these initiatives mark a departure from reliance on imperfect simulations, challenges remain regarding the feasibility of scaling this process to produce meaningful results. The significance lies in the potential for humanoids to transform labor markets, but questions linger about the effectiveness and profitability of training robots through the countless daily actions of workers in diverse global locations. Ultimately, this could reshape how we view labor and data collection in the age of robotics.