Cutting-edge AI techniques will help the industrial robots of today enable the manufacturing of tomorrow on a global scale.
Robotics has emerged as one of the most exciting use cases for AI, with labs specializing in the combination of robotics and AI research springing up at corporations and universities all over the world.
To most casual observers, the only time they see the topic of industrial robotics in the media, the message is similar to when they see discussions of artificial intelligence — the accelerating loss of jobs through automation.
But if we look closer at the robotics industry, a different picture emerges.
Robots can improve the speed, accuracy and safety of production. They enable new classes of products to be created and allow for manufacturing to take place in a wider range of geographies than previously. If we follow the deployment of industrial robots around the world, we see that higher robot deployment is directly related to higher industrial employment. In short, robots enable modern manufacturing. If a country wants to have a robust manufacturing industry, then deployment of industrial robots goes hand in hand with that.
For example, according to IFR World Robotics, Asia accounted for 69.1% of industrial robot shipments in 2018. Europe accounted for 17.2%, and the United States accounted for 13.7%. So, Asia has the most manufacturing jobs as well as the highest number of industrial robot shipments. Industrial robots enable manufacturing, as well as the jobs that go along with it.
That is not to say that these robots are perfect. In fact, there is massive room for improvement in terms of ease of programming, safety in working in proximity to human workers, and flexibility of robots’ capabilities. Most robots are strong and can move with precision but are incapable of doing anything they haven’t been specifically and painstakingly programmed for.
Industrial robots have been in use since the 1960s. At that time they were enormous, heavy, expensive, used only for a single task and extremely dangerous for use around human workers. Initially they were used primarily in heavy industry such as auto manufacturing and steel production Today, however, traditional industrial robots see major use in the electronics and high tech manufacturing industries as well as food processing and logistics. To give some examples of what we are talking about when we say ‘industrial robots, here are some examples. Some of the most common manufacturing robots in use today are welding robots, assembly robots and picking and packing robots.
Arc-welding robots are used in automobile and metal production, and improve precision and uniformity of results as well as worker safety by reducing exposure to dangerous fumes and burn risks.
Assembly robots are used on automated assembly lines including automotive, consumer electronics and home appliances. and support lean manufacturing processes — as well as improve speed and precision. These systems can employ computer vision and sensors to identify, grasp and manipulate parts that may not lie in specific locations on the line and to assemble parts using feedback to gauge the level of force applied and how well parts are being assembled. Robots can also manage assembly of parts that are too small for human hands to manipulate accurately and to dramatically speed up production times.
Picking and packing robots are involved in taking the products off a production line and packaging them. In the future, we may see more robots taking a mixed range of products from inventory and packing them for shipment to fulfil custom orders. Picking and Packing robots require flexibility and a range of motions. Amazon primarily uses people for picking and packing
Other manufacturing robots include cutting, drilling, milling and painting robots.
By Industry, automotive accounted for 36% of robots sold this year, electrical/electronics accounted for 32% and is currently the fastest growing segment, Metal industries were 10% and the remainder were for food, logistics and other unspecified industries. The annual market size for industrial robots is $15bn.
But there is reason to believe that use cases can grow dramatically and robotics can play a much bigger role in modern business and manufacturing.
Some major concerns with the current generation of robots is safety in working in close proximity to humans and the ability to reconfigure robots quickly and simply. Also researchers are seeking to improve the ability of robots to grasp delicate items of varying shapes, such as food, without harming them, which would dramatically expand the use cases for robots.
These are all areas where artificial intelligence research is emerging as an extremely important addition to robotics technology.
AI-powered Industrial Robotics
AI-powered computer vision systems are one of the most important avenues of research in robotics. In short, AI technology can give robots the power to see, and to identify and interact what it is they are seeing.
Amazon sponsors an ongoing “robot picking” contest, where robots compete to try to pick up various different types of products. Every team in the contest is currently using AI to train their robots and performance is improving dramatically.
Robots are also being outfitted with force sensors, which provide information about how hard they grasp, manipulate or nudge parts in a manufacturing environment or to measure the impact of a collision with an object or worker. These sensors along with machine learning techniques are being used to develop collaborative robots that can work safely alongside humans.
A major obstacle in the development of AI-powered industrial robots is the availability of training data.
Elon Musk recently tweeted about a project from Open AI in which robots were trained entirely on synthetic data, in a simulated environment, and then when tested on tasks in the real world were able to function properly, demonstrating that it is possible to use synthetic data to train AI powered robots for real world applications.
This is similar to the way a human pilot can be trained to fly a plane in a flight simulator. While the representation of the world in the flight simulator may not look exactly like the real world, it looks and behaves enough like the real world that this simulated experience is almost as valuable as ‘real’ experience. And, given flight simulator time is far cheaper than real flying time, it is a fast and cost effective way to train pilots.
To complete the metaphor, training a neural network machine learning model is similar to a human ‘learning from experience’, with data standing in for experience in the AI context and synthetic data standing in for simulator time, or simulated experience.
As synthetic data is a Neuromation core specialty, we are currently working on several projects in the area of industrial robotics, including quality control computer vision systems for Linfinity Foundation in China, allowing us to identify faulty products on a production line, identifying faulty equipment performance and speeding production through more accurate and flexible robot performance.
We look forward to maintaining leadership in the industrial robotics space through key industry partnerships and close attention to leading research in the space.
We invite you to connect with Neuromation!
by Angus Roven
Senior Analyst, Neuromation
