Still, some experts are doubtful about the allure of Blue. They notice that it is not that different from Baxter, another bot with pincers and arms which was supposed to work alongside humans.
This makes easier to work around but also suitable for research using a kind of AI training method that’s becoming popular in robotics, reinforcement learning. Learning operates by asking an agent to complete a job when it does, and rewarding it. It is basically trial and error, with the broker slowly teaching itself over time and then starting out without the knowledge of how to complete its goal.

International robot sales have skyrocketed over the previous five decades, but the bots nowadays, deployed in warehouses and factories are pretty much the same as the ones we had decades. They are strong and precise but expensive to get and dangerous for humans.

Picture: UC Berkeley
Blue the robot is”backdrivable,” meaning that you can push it around even when it’s turned away.

Photo: UC Berkely
From left to right: Pieter Abbeel, David Gealy, and Stephen McKinley.

Abbeel and his staff, research fellow Stephen McKinley and student David Gealy, expect Blue will function in precisely the exact same manner. It’s a central vision module with a depth-sensing camera, and motors with rubber bands which give it flexibility control its arms. It is like pushing against a brick wall, if you push an industrial robot arm. However, Blue is similar to a person in a crowded subway car: jostle it, and it’ll move aside.
Ankur Handa, a robotics researcher at Nvidia, stated Blue’s pincers limit the range and its lack of precision would be a issue with AI controls. “Overall, I do not think they are offering something fresh,” Handa informs The Verge.

However, Abbeel is bullish about Blue’s future. The robot has been constructed in tiny batches but Abbeel expects to scale upward, eventually moving to outsourced manufacturing to produce numbers that are bigger. The target customers would be research labs and universities where robots are shared among teams, much like computers were in the 1960s. Offering a robot can make them widely accessible, boosting the output of robot study.
This is another area where a difference might be made by Blue. PR2, a research robot built by Willow Garage that also has a set of pincers and arms, place researchers back. The bill of materials for Blue is only $3,000. Abbeel states a final price hasn’t been decided on by the team, but they are hoping to target the range.
Blue, a brand new robot out of UC Berkeley, intends to break that mold with the assistance of AI.

The roboticist directing the project, pieter Abbeel, needs to change this, and he says Blue has been built from the ground up to take advantage of developments in AI. “The simple fact that AI is getting more competent gave us an chance to rethink the way to design a robot,” Abbeel informs The Verge.

Abbeel explains that many robots in use today are built to be accurate and powerful. Their movements are predefined, and they just repeat the same action over and over again, whether that’s lifting pallets welding cars, of goods, or fastening screws.
More importantly, Abbeel hopes that Blue will offer a blueprint for the home robot of the future might look like: something that performs nicely with humans, and is low cost, flexible. “The residence is absolutely what we have in mind with this sort of layout,” he says. “There is still a great deal of challenges ahead, and it’s not like we think this specific robot is moving in a house. [However ] this is a design paradigm which takes us in a new leadership .”
Employing conventional robots with reinforcement learning could be costly. Their lack of versatility makes them fragile and prone to damage. Plus, reinforcement learning takes time to create results, and because the robots are costly, the prices quickly accumulate.
Blue looks a bit it’s made from bulky, 3D-printed parts, and it’s a set of robot arms using pincers for handson. It can be controlled with VR handsets, which allow operators wave their arms about and then Blue will wave its arms in tandem. In addition, it can be trained to control objects using artificial intelligence, a management system that’s still surprisingly rare in robots.
“That’s possible once you’re prepared to forego sub-millimeter precision because you understand you do not need it with AI-based control,” states Abbeel.

The robots of the future, in contrast, will be lively and reactive. They will be able to work safely alongside humans without crushing themand rather than getting their activities planned in advance, they’ll browse the world in real time using cameras and detectors.

“If you look at traditional robots, then they are designed around the principle of very high precision and repeated moves,” says Abbeel. “But you do not necessarily need sub-millimeter repeatability.” (That’s the ability to execute the same task over and above with gaps in movement of less than a millimeter.) “Humans don’t have sub-millimeter repeatability. Insteadwe use our eyes and feel of touch to get things done through feedback.”

Picture: UC Berkeley
Blue is capable of complicated tasks such as folding a towel.

Plenty of startups and research labs are targeting this paradigm, trusting to teach robots how to work using artificial intelligence. Abbeel is the president of a few of these, a startup called Embodied Intelligence. Kindred AI, a company that builds robots that select items in warehouses, is just another . The Elon Musk-founded research laboratory OpenAI has performed similar work utilizing robot palms , and Google is also researching AI-training for robots.