Donate to Science & Enterprise

S&E on Mastodon

S&E on LinkedIn

S&E on Flipboard

Please share Science & Enterprise

Human-Friendly Robot Designed for A.I.

Blue robot and developers

Blue, in foreground, with its developers, L-R, Pieter Abbeel, David Gealy, and Stephen McKinley. (Philip Downey, Univ of California, Berkeley)

9 Apr. 2019. A robotics team at University of California in Berkeley designed a device to learn about humans’ needs with artificial intelligence and respond gently to their presence. Researchers from Berkeley’s Robot Learning Lab plan to describe Blue — short for Berkeley robot for Learning in Unstructured Environments — next month at the IEEE International Conference on Robotics and Automation in Montreal, and started a company to market the systems.

The Blue project is led by Robot Learning Lab director Pieter Abbeel, a professor of electrical engineering and computer science, with postdoctoral research fellow Stephen McKinley and doctoral student David Gealy. The initiative reverses the usual process of adding artificial intelligence to an already-designed robotic device, and instead designed a robot for human interaction with artificial intelligence, or A.I., algorithms that learn about the humans in its environment.

“A.I. has done a lot for existing robots, but we wanted to design a robot that is right for A.I.,” says Abbeel in a university statement. “Existing robots are too expensive, not safe around humans, and similarly not safe around themselves. If they learn through trial and error, they will easily break themselves. We wanted to create a new robot that is right for the A.I. age rather than for the high-precision, sub-millimeter, factory automation age.”

The Blue device consists of 2 computer arms with grippers and joint-hinges resembling shoulders and elbows of a body-builder made from durable, but everyday plastic on a central vertical metal pillar. It’s algorithms use deep reinforcement learning, which are rewarded or reinforced when making the right decisions and penalized when they go wrong. An algorithm known as AlphaGo, published in October 2017, uses reinforcement learning to master the Chinese strategy game of Go and even beat world champion players, starting from absolutely no knowledge of the game.

Previous generations of robots, as Abbeel observes, were built to automate factories, requiring consistent strength and precision to carry out highly structured and repetitive tasks. These kinds of devices would likely wreak havoc in many American homes, with children and pets, and where structure and tidiness are more the exception than the rule. But machine learning, where algorithms change as new data are encountered, will lead to trials and errors, thus a robot interacting with humans through machine learning in this environment needs to be more sensitive to its surroundings.

“One of the things that’s really cool about the design of this robot,” notes Gealy, “is that we can make it force-sensitive, nice and reactive, or we can choose to have it be very strong and very rigid. Researchers can adjust how stiff the robot is, and what kind of stiffness. Do you want it to feel like molasses? Do you want it to feel like a spring? A combination of those? If we want robots to move toward the home and perform in these increasingly unstructured environments, they are going to need that capability.”

Another human-like quality designed into Blue is limits on strength and endurance. For example, Blue’s arms can hold 2 kilograms, but not forever. Instead, the Blue team designed in agility with a wide range of motion for the arms, sacrificing consistent strength and precision. “What we realized was that you don’t need a robot that exerts a specific force for all time, or a specific accuracy for all time,” says McKinley. “With a little intelligence, you can relax those requirements and allow the robot to behave more like a human being to achieve those tasks.”

Abbeel, Gealy, and McKinley founded the start-up company Berkeley Open Arms to market Blue robots. The developers aim to have beta versions of Blue available in May 2019 and production models on the market by 2020. The following video demonstrates some of Blue’s capabilities.

More from Science & Enterprise:

*     *     *

Comments are closed.