Berkeley’s Blue AI-Powered Robotic Arm Is Human-Friendly and Cost-Effective

Researchers from the University of California have developed a robotic arm platform, named Blue, which uses AI and deep reinforcement learning to perform intricate human tasks — such as folding towels, arraigning flowers, pouring a cup of coffee and cleaning up the mess afterward. Its arms are about the size of a bodybuilder, made of durable plastic parts, and are sensitive to outside forces — like a hand pushing it away, meaning it’s safe to work around.

AI has been tapped for robotic use in many industries, including Boston Dynamics, who recently highlighted their Handle reimagined robot for logistics, and Pick platform, which uses 3D vision to sort boxes on pallets.

“AI has done a lot for existing robots, but we wanted to design a robot that is right for AI. 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 AI age rather than for the high-precision, sub-millimeter, factory automation age.” — Pieter Abbee, lBerkeley professor of electrical engineering and computer sciences

Under the hood, Blue is designed around a QDD (Quasi Direct Drive) and uses a series of links (or joints) outfitted with a 2-DOF geared differential module that combines two planar QDD timing belt transmissions into output pitch and roll motions, giving it a high degree of dexterity. They also make the robot highly backdrivable with no tension in the limbs when being pushed or pulled. The arms are quite stout too and are capable of lifting a continuous payload of 4.4 pounds.

Blue even features a central vision module with a depth-sensing camera that it uses for deep reinforcement learning, which is done through observing a task, and being rewarded when accurately recreated. The robot costs less than $5,000 and is currently in Beta form with 50 or so units up for production. The researchers state that a final consumer version is expected to hit the market sometime at the end of this year or early 2020.