Just as simple belt- and gear-driven machines propelled quantity and quality of manufacturing and production in the 18th and 19th centuries, robotics have ushered in similar advances in the late 20th to the present. All technologies across the spectrum are improving at an exponential rate, and there is no doubt that production robotics will be riding this wave of innovation. At Motivo, we predict that the next huge shift in robotics and manufacturing will center around machine vision and learning.
Apart from the various autonomous mobility endeavors, we have seen several players across diverse, traditionally slow-moving industries approach us with their manufacturing aches and pains. Today, we are able to help them solve their problems with conventional robotics and traditional vision systems with highly accurate, repeatable actions with super-precise measurements. But the most progressive companies among them are demanding more. They want machines that adapt to and learn from changing products and environments. Motivo is here to help them navigate the technological options and implement the subsystems to make their ideals real, working products.
Machine vision combined with deep learning via artificial neural networks (what a mouthful!) allows for a level of pattern recognition as good as or better than a human’s. And a computer doesn’t get sleepy, bored, or angry. Many industries will benefit from these advances.
Orienting highly variable product for post-processing
Predicting bad product
Deciding what is good versus bad product
Genetic engineering without gene editing
Highly precise actions that would normally cause pain in human technicians
Finishing off hybrid handmade/machine-made products
Finding the most efficient build processes
Predictive worldwide product demand
High tech material imperfection non-destructive testing
Discovery of complex correlations that lead to product issues
Variable actions requiring constant human attention
Optimizing box packing
Optimizing truck loading
Optimizing shipping routes
Robotics has already helped industries where human labor is subjected to boring, precise, uncomfortable, or hazardous work. However, some jobs remain that need the keen eye of a trained employee, despite the hazards to life and limb. This is costing companies a ton of money in insurance, training, and salary. This is not what the culture of high-technology has promised us, and machine learning will be the beginning of actually fulfilling that promise.
In robotics, there are 2 primary implementations: off-the-shelf robotics (from companies like Kuka, Fanuc, Universal Robots, Denso, Motoman, etc.) and custom electromechanical actuators and control systems. Combining this with machine learning, the food and agriculture markets are finally taking steps toward using robotics to do mundane or dangerous work. The nature of a “God-made” product is such, though, that an intelligent detector is needed to make decisions about that work, lest the end-product quality suffers.
Training these neural networks takes time and investment, above and beyond what is seen in training an individual employee; however, this is not an apples-to-apples comparison. Neural networks are creating the foundation for multiple iterations and a solid foundation for multiple applications of the knowledge. In short, the development time required for creating a neural network is more akin to training multiple employees and multiple shifts, and in teaching them how to learn, we create a pathway for future, less time intensive iterations and a more reliable, accurate, and sustainable robotic workforce.