While there is a lot of work to do, Tedrake says that all evidence indicates that the methods used in LLMS also work with robots. “I think it changes everything,” he says.
The measurement of progress in robots has become more challenging, in nature Download refrigerators or Take garbage output With ease. YouTube clips can be deceptive, however, human robots tend to be known remotely, pre -programmed, or trained to do one task in very controlled conditions.
The new ATLAS work is a great sign that robots have started experimenting with this type of equivalent progress in robots that ultimately led to the general language models that were presented to us Chatgpt In the field of artificial intelligence. In the end, this progress can give us robots that are able to work in a wide range of chaotic environments easily and have been able to quickly learn new skills – from welding tubes to espresso making – with wide -ranging reinforcement.
“It is definitely a step forward,” says Ken Goldberg, a automatic scientist at the University of California, Berkeley, who receives some funding from Tri but did not participate in the atlas’ work. “Coordination of legs and weapons is great.”
However, Goldberg says that the idea of an emerging robot behavior should be treated carefully. Just as the amazing capabilities of large language models can sometimes be followed to examples listed in their training data, he says that robots may show skills that look more new than they are already. He adds that it is useful to know details about the number of times the robot succeeds and in the ways he fails during experiments. Tri was previously transparent with the work on LBMS and more data is released about the new model.
Whether expanding the data range used to train robot models will open the emerging behavior than ever, it remains an open question. in discussion In May, I was held at the International Conference on Robots and Automation in Atlanta, Goldberg and others that engineering methods will also play an important role to move forward.
Tedrik, for example, is convinced that robots are close to the turning point of that which will enable more use in the real world of human beings and other robots. “I think we need to get these robots out of the world and start doing real work,” he says.
What do you think of the new Atlas skills? Do you think we are heading to a hacker similar to Chatgpt in robots? Allow me to know your thoughts on [email protected].
This is an edition of Will Fares Newsletter of the AI. Read the previous newsletters here.
https://media.wired.com/photos/68b8794b52f451c993bb2dd9/191:100/w_1280,c_limit/AI-Lab-New-Frontier-Robotic-AI-Business.jpg
Source link