Episode 8

WORLD'S LATEST DEEP LEARNING

Boosting productivity in manufacturing


PLAY MOVIE

Innovation Japan

WORLD'S LATEST DEEP LEARNING

Boosting productivity in manufacturing

Manufacturing sites worldwide use industrial robots to streamline and automate operations. Japan has led the world in industrial robot technology, and now is about to spur an even greater evolution by combining this technology with an open source deep learning framework developed in Japan. Deep learning enables industrial robots to make judgments in complex operational situations by learning from past examples. Moreover, the learning can be shared between robots to increase efficiency. Watch this video to learn how this innovative approach will deliver previously unavailable levels of advanced automation in manufacturing.

00:17

Industrial robots are used in manufacturing plants around the world to streamline and automate operations.

Japanese robots have led the drive toward factory automation, and today they’re evolving to offer even greater gains in efficiency through deep learning.

00:40

Toru Nishikawa
President & CEO
Preferred Networks, Inc.

Deep learning technology, which is our specialty, is driving the evolution of industrial robots.

This technology makes it possible for robots to provide more flexible operations and capabilities.

We developed Chainer, an open source deep learning framework that allows extremely flexible implementation of deep learning algorithms that is widely used around the world by researchers and developers.

01:14

Eiichi Matsumoto
Researcher
Preferred Networks, Inc.

This is an industrial robot’s picking process that utilizes image recognition technology.

By utilizing image recognition with three-dimensional images and deep learning, the robot gains the ability to select the best positions to pick-up blocks in a disorganized pile.

At first, the robot is incapable of picking up the blocks correctly, but after about eight hours of training, it gains the ability to perform the picking operation with greater than 90% precision.

01:56

Although human beings can’t share the implicit knowledge resulting from training, robots can share skills learned instantaneously.

So if 10 robots are connected together on a network, they will learn 10 times as quickly.

02:14

Deep learning can also be used to enable robots to cooperate with other robots, and to detect and prevent imminent malfunctions to avoid manufacturing line stoppages.

02:28

Toru Nishikawa
President & CEO
Preferred Networks, Inc.

By combining our deep learning technology and software with Fanuc’s world-class hardware, we can lead the world in this field.

02:39

Deep learning is driving a variety of changes in manufacturing by enabling higher throughput and shorter turnaround in production.

02:51

Toru Nishikawa
President & CEO
Preferred Networks, Inc.

I look forward to tapping the full potential of our deep learning technology outside the field of industrial robots, for example to dramatically advance transportation systems through technologies such as self-driving and to solve the complex, difficult-to-elucidate phenomenon of cancer in the life sciences.

This video was created in collaboration with Preferred Networks, Inc.