Blue the robot could be the workhorse of the future at the IA

Global robot sales have doubled in the last five years, but robots deployed in factories and warehouses are pretty much the same as we had decades ago. They are powerful and precise, but expensive to buy and dangerous for humans.

Blue, a new UC Berkeley robot, aims to break this mold with the help of AI.

The blue is a little like the drawing of a robot designed by a child: it is composed of large pieces printed in 3D and a pair of humanoid robot arms with a clip for the hands. It can be controlled using VR handsets, which allow operators to wave, then Blue to signal his arm in tandem. He can also be trained to manipulate objects using artificial intelligence, a control method that remains surprisingly rare in robots.

Pieter Abbeel, the robotist who runs the project, wants to change that. Blue was built from scratch to take advantage of recent improvements to AI. "The fact that AI is becoming more efficient has given us the opportunity to rethink the design of a robot," Abbeel said. The edge.

Abbeel explains that most of the robots used today are designed to be powerful and accurate. Their movements are predefined and they simply repeat the same action, whether lifting pallets of goods, welding cars or fixing screws on a smartphone.

The robots of the future, by comparison, will be reactive and dynamic. They will be able to work safely alongside humans without crushing them. Instead of planning their actions in advance, they navigate the world in real time with the help of cameras and sensors.

"If you look at traditional robots, they are designed around the principle of very high precision and repeated movements," says Abbeel. "But you do not necessarily need less than a millimeter repeatability." (It's the ability to continually perform the same task with movement differences of less than one millimeter.) Instead, we use our eyes and sense of touch to make things happen through comments. "

Abbeel and his team, researcher Stephen McKinley and graduate student David Gealy, hope that Blue will work the same way. It has a central vision module with a depth detection camera and its arms are controlled by motors with elastics that give it great flexibility. If you press against an industrial robot arm, it's like leaning against a brick wall. But Blue looks more like a human in a crowded metro car: jostle her and she will move away.

This makes Blue safer, but also suitable for research using reinforcement learning, a type of AI training method that is gaining popularity in robotics. Reinforcement learning involves asking an agent to complete a task and rewarding it when it does. Essentially, it is a question of trial and error, the agent starting without knowing how to reach his goal and then teaching himself little by little over time.

Photo: UC Berkely

The use of traditional robots with reinforcement learning can be expensive. Their lack of flexibility makes them fragile and subject to damage. In addition, reinforcement learning takes time to produce results and, because robots are expensive, costs add up quickly.

This is another area where Blue could make the difference. PR2, a popular research robot built by Willow Garage, featuring two arms and a forceps, reports researchers about $ 400,000. By way of comparison, Blue's nomenclature is only $ 3,000. Abbeel said the team had not decided on the final price, but she hoped to target the $ 5,000 range.

"It becomes possible when you are ready to give up less than millimeter accuracy, because you realize you do not need it with AI-based control," says Abbeel.

Many other research labs and startups are also interested in this new paradigm, hoping to teach robots how to work with artificial intelligence. Abbeel is the president of one of them, a startup named Embodied Intelligence. Kindred AI, a company that makes robots that pick up items from warehouses, is another. The OpenAI research laboratory, founded by Elon Musk, did a similar job using robots; Google is also exploring the possibility of training AI robots.

Some experts remain skeptical about Blue's call. They note that this is not so different from Baxter, another robot with arms and forceps designed to work alongside humans. The company that manufactured Baxter, Rethink Robotics, shut down last year.

Ankur Handa, a robotics researcher at Nvidia, said Blue's forceps limited the range of tasks she could perform, and that her lack of precision would be a problem even with AI commands. "Overall, I do not think they are proposing anything new," said Handa. The edge.

But Abbeel is optimistic about Blue's future. The robot is being built in small quantities, but Abbeel hopes to move up a gear, eventually moving to outsourced manufacturing to produce larger quantities. The first customers targeted will be the research laboratories and universities where the robots are currently shared between teams, much like the computers of the 1960s. By offering a cheaper robot, they will be more widely available, which will increase the results of the robot research.

More importantly, Abbeel hopes that Blue will provide a detailed blueprint of what the home robot of the future might look like: something that is inexpensive, flexible, and works well with humans. "The house is absolutely what we think with this type of design," he says. "There are still many challenges ahead and it's not like we're thinking this specific robot goes into a house. [But] It's a design paradigm that takes us in a new direction. "