This Robot Requires Just One AI Model to Perfect Humanlike Actions
Although significant work remains, Tedrake believes that current evidence indicates that the methodologies applied to LLMs are also applicable to robotics. “I think it’s changing everything,” he states.
Assessing advancements in robotics has become increasingly difficult, especially with videos showcasing commercial humanoid robots performing intricate tasks, such as loading refrigerators or taking out the trash effortlessly. However, YouTube videos can be misleading; humanoid robots are often either teleoperated, pre-programmed with precision, or trained for specific tasks under strict conditions.
The recent developments with Atlas serve as a vital indicator that robots are beginning to undergo the kind of transformative progress in robotics that eventually led to the general language models like ChatGPT in the generative AI domain. This progress could eventually result in robots that can navigate various chaotic environments proficiently and quickly acquire new skills—from welding pipes to brewing espresso—without requiring extensive retraining.
“It’s definitely a step forward,” remarks Ken Goldberg, a roboticist at UC Berkeley, who receives partial funding from TRI but was not part of the Atlas project. “The coordination of legs and arms is a significant achievement.”
Goldberg cautions, however, that emergent robot behavior should be approached with caution. Just as the unexpected capabilities of large language models can often be traced back to examples in their training data, robots may reveal skills that appear more novel than they truly are. He emphasizes the importance of understanding how often a robot succeeds and the nature of its failures during trials. TRI has been transparent with its previous work on LBMs and may release further data regarding the new model.
Whether simply increasing the data used to train robot models will result in more emergent behavior remains uncertain. At a debate in May during the International Conference on Robotics and Automation in Atlanta, Goldberg and others advised that engineering approaches will also play a crucial role in future developments.
Tedrake, for his part, is confident that robotics is approaching a pivotal moment—one that will facilitate more practical applications of humanoid robots and other automated systems. “I think we need to deploy these robots in the real world and start having them perform actual work,” he asserts.
What are your thoughts on Atlas’ new capabilities? Do you believe we are on the verge of a ChatGPT-like breakthrough in robotics? Feel free to share your opinions at ailab@wired.com.
This is an edition of Will Knight’s AI Lab newsletter. Read previous newsletters here.
Rajat Sharma
https://in.linkedin.com/in/rajat-mediaHelping D2C Brands Scale with AI-Powered Marketing & Automation 🚀 | $15M+ in Client Revenue | Meta Ads Expert | D2C Performance Marketing Consultant