Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Although there is a lot of work to do, Tedrek says that so far all the evidence shows that the procedure used in LLMs also works for robots. “I think it’s changing everything,” he said.
Gazing progress in robotics has become more challenging late, of course videoclips show commercial humanoids perform complex tasks, as Loading the refrigerator Or Garbage It seems to be comfortable. YouTube clips can be fraudulent, and humanoid robots are either telegrained, carefully programming in advance, or trained to do a single job in very controlled situations.
The work of the new Atlas is a big symptom that robots have begun to experience the kind of equivalent progress in robotics that lead to the last of the common language models given to us Chatzipt In the case of the generator AI. Eventually, such progress can give us robots that are easily able to handle in various messy environments and are able to quickly learn new skills – from welding pipe to creating aspresso – without a wide re -training.
“This is definitely a step ahead,” a roboticist from UC Berkeley said Goldberg, who received some funds from TRI but was not involved in Atlas’ work. “The combination of legs and arms is a big subject” “
Goldberg says, however, the idea of ​​emerging robot behavior should be carefully treated. The amazing skills of greater language models such as sometimes can be identified in examples included in their training data, as he says that robots can demonstrate skills that seem more fancy than them. He added that a robot is helpful to know how many times a robot is successful and how it fails during the test. TRI was transparent about the work done on LBMS before and could reveal more data in the new model.
Whether simple scaling of data used for training robot models is an open question whether it will unlock even more emerging behavior. This A debate Goldberg and others warned at the international conference on Robotics and Automation in Atlanta, warning that engineering methods would also play an important role in moving forward.
Tedrek, for one, is convinced that robotics are moving closer to a reflection point — it will enable more real-world use of humanoid and other robots. “I think we need to put these robots out of the world and start the real work,” he said.
What do you think about Atlas’s new skills? And do you think we’re going to the breakthrough of a Chatzipt-style in Robotics? Let me know your thoughts on ilb@wired.com.
This is a version Will Nights AI Lab NewsletterThe Read the previous newsletters Here
