Robots Move, Collaborate and Learn at FANUC America

In the old days, an industrial robot was rooted to one spot, fenced off from human contact, and simply did what it was told. Times have changed. FANUC America Corp.’s booth features robots that roll around place to place, robots that closely collaborate with human workers and robots that learn on the job.

Mobile Robots

FANUC mobile robots combine an automated guided vehicle with an articulated-arm robot. The challenge of making such a robot practical goes beyond the chore of getting the device from one place to another, notes Mike Cicco, FANUC America Corp. president and CEO. Relocating a robot used to also require the tedious reteaching of all its movement points using a pendant control. FANUC’s mobile robots instead automatically recalibrate their movements with a vision system that reads fiducial markers—reference dots—placed on the work station, saving potentially hours of programming time.

Collaborative Robots

Among the recent technology developments for collaborative robots are more responsive sensors and more computing power in networked connections, Cicco says. Demonstrations of FANUC collaborative robots show how efficiently and safely they work with human colleagues. Through the implementation of the ISO/TS 15066 safety standard (which was introduced in 2016), “We’ve been able to let the robots go faster, depending upon what the application is,” Cicco says. “When workers are nearby, the robot goes a little bit slower; if they leave, it can go faster.” Through sensors and software, there is the ability to run a risk assessment and to adjust the cobot’s parameters accordingly.

Learning Robots

One of the over-the-horizon automation systems being demonstrated combines a vision system with artificial intelligence (AI) and a deep-learning algorithm so a robot can teach itself to 3D pick an entire bin. The robot uses a camera to take a picture of what is in the bin, but it has no idea what the part looks like or where and how to pick it up. The algorithm gauges whether the pick succeeds or fails, enabling the system to learn how to make a good pick as it empties the bin.

“The cool thing about deep learning is that if two robots perform the same operation and share their successes and failures over a neural network, they will cut learning time in half because they will never repeat the same error,” says Cicco. “If hundreds connected AI robots perform the task, we could find answers to complex problems pretty quickly.”