As a result of the progress in sensors, computing, and networking, global manufacturing is facing industrial transformation. Industrie 4.0 in Germany, the so-called industrial Internet of Things in the U.S., and the Industrial Value Chain Initiative in Japan are entirely new concepts. Collaborative robots such as those from Rethink Robotics Inc. or Universal Robots S/A are changing processes in factories, and manufacturers are starting to apply deep learning technologies.
In my last article, we looked at how Preferred Networks Inc. (PFN) and FANUC are working on “edge computing” and applying deep learning to Japanese manufacturing. I also had a chance to discuss the future of connecting manufacturing and artificial intelligence with Keigo Kawaai, who is in charge of this engineering project in PFN.
- Deep learning can be applied to make manufacturing more efficient, according to Japanese researchers.
- Preferred Networks Inc. is among the Japanese organizations working to apply robotics and AI to industrial transformation.
- The U.S. may be the leader in robotics research, and China is the world’s factory, but Japan expects its talent and companies to keep up.
Deep learning possibilities and bottlenecks
No production lines have actually installed a deep learning application yet. However, that situation will likely change drastically in a few years. Because of international attention and top-down governmental initiatives, every manufacturer has the motivation to join industrial transformation.
If someone builds a good app, many businesses will use it, predicted Kawaai.
A significant shift already happening: Because of e-commerce, production lines need to adapt to high-variety, low-volume batches and cellular manufacturing. One of the biggest concerns for manufacturers is the labor-intensive process of manual teaching. PFN believes that deep learning can solve this issue.
Industrial automation is today most widespread in the electronics and automotive industries, with some usage in medicine and food processing. The cost of installation is a barrier to adoption for robots in more areas of these sectors and in other markets.
Systems integrators are usually needed to launch new production lines. For instance, a customer may need an integrator to help tune its production line by choosing an end effector for a robotic arm. If a deep learning system could decrease the labor for tuning unstandardized production lines, then industrial transformation could step up to the next level.
Leading the way to the factory of the future
Both object recognition and motion planning are key technologies for building the factories of the future.
Looking ahead to more fully automated facilities, Kawaai said PFN recognized that it needed to work on not just object recognition, but also motion-planning algorithms.
Therefore, the company has started a demonstration project that uses reinforcement learning to guide the movement of robotic arms.
Kawaai cited Mujin Inc., another Japanese startup that has specialized in fast and intelligent controllers for industrial robots. Major Japanese manufacturers such as Canon Inc., Honda Motor Co., and Denso Corp. already use Mujin’s controllers.
Since China and Asia in general are the world’s factory, increasing automation is an important topic to them. PFN doesn’t focus on these markets by itself, said Kawaai, but it would like to capture them through its partnership with FANUC. Unlike active U.S. investors, PFN has not yet been contacted by Chinese or Indian venture capitalists.
More on Manufacturing Automation and Industrial Transformation:
- European, Asian Robotics Look to Safer Cobot Implementation
- Robotics Companies Should Develop a GeoRobotics Strategy
- Stepper Motors Get New MCST 3601 Motion Controller From Faulhaber Trinamic
- What Does New Robotics Roadmap Mean for U.S. Manufacturing, Logistics?
- British Robotics Must Catch Up to Global Competitors
- Gentle Touch Capabilities Promise to Give Robots More Value
- Machine Vision Investments Eye Safety, New Apps
- Massachusetts Companies Start Chinese Robot Visit With High Hopes
- Japan Begins Executing on Its Robot Plans
Japan and AI
Is Japan ahead, behind, or equal to other countries pursuing industrial transformation?
In September, Google, Amazon, Facebook, IBM, and Microsoft formed an alliance around AI. Some observers expressed concern that the rapid technological evolution is led by U.S. companies and research organizations.
However, looking back Japan’s technological history, a basic concept of a neural network called “Neocognitron” was invented by a Japanese researcher, Kunihiko Fukushima, in the 1980s.
From 1982 to 1992, there was a $600 million national project called “Fifth Generation Computer Systems.” The project was not successful in the context of business, but it educated many researchers who are now running AI labs in universities or companies.
Although Japan got a late start toward the current deep learning boom, big IT companies including Rakuten, Yahoo, Hitachi, Recruit, and Dwango have invested a lot in this field. In addition, there are some promising young companies such as Metaps, Fronteo, PKSHA, and ABEJA that aim to apply AI to niche fields.
In robotics, SoftBank Robotics Corp.’s Pepper is designed to recognize human emotions. The former head of Pepper’s development, Kaname Hayashi, has founded GROOVE X, which is developing another “empathic” robot (using, of course, AI techniques).
As one of the top AI companies in Japan and the world, PFN isn’t worried about the situation. Kawaai said that it was wrong to assume that there are fewer researchers in Japan who are capable of competing with those in the U.S. or elsewhere.
“In the history of Web technology, the only difference was that Google or Facebook could build a business model,” he said. “The turning point for deep learning will be the same. In the intersection field of AI and manufacturing, there are no companies that know how to build a big business based on deep learning. We’d like to challenge that.”