AI breakthroughs in robotics won't change your life any time soon (www.technologyreview.com)

Tesla's Optimus humanoid robot, championed by CEO Elon Musk, promises to transform labor by performing tasks traditionally done by humans, with a target price of around $20,000 each. Musk envisions these robots operating in factory settings and households, claiming they could soon achieve human-like dexterity. However, the hype surrounding humanoid robots has been met with skepticism from the robotics community. Experts like Yann LeCun emphasize that current AI technologies, despite their success in NLP, are ill-suited for mastering the complexities of the physical world that robotics entails. This distinction highlights the gap between developing a humanoid robot's appearance and its functional capabilities. Advancements are being made through tools like Google DeepMind's ALOHA 2, which employs vision-language-action models to improve task performance by allowing robots to learn from examples rather than relying on hard-coded instructions. While impressive, these models still face limitations, as they often fail outside their training data. Researchers argue that achieving true generalist robots will require innovative approaches, including the development of world models that predict real-world physics rather than solely relying on extensive datasets. Companies are investing heavily in this emerging field, but significant challenges remain before humanoid robots can become commonplace in everyday life.