AI-enabled robotics and automation

We have been automating things for thousands of years to speed up a litany of laborious time-consuming tasks. Early automation tools were rudimentary and mechanical, but thanks to technological advances, automation has evolved into sophisticated systems capable of incredible feats of productivity and precision. The computing power of today’s processors and the advance of artificial intelligence (AI) make it possible not only to automate many human tasks, but also to extend or “augment” human abilities. As AI continues to develop, where might automation go from here?

Romans and the ancient Greeks used gravity to power their automation devices: water wheels ground wheat into flour and water screws drew water from ships’ hulls and irrigated crops. Waterpower continued to play a critical role into the early stages of the industrial revolution, when it was superseded by steam. Factories were built around central steam-powered turbines, with machines requiring more torque placed closer to the turbine and those needing less placed further away sometimes on different floors, connected via a series of drive-belts and pullies.

Electricity was a significant step-change for automation because power could be delivered and independently controlled for each machine around the factory. Modern batteries take this a step further. Lightweight and rechargeable, they allow automation systems to be untethered from a fixed power supply and therefore mobile. Automated trollies (AGVs or AMRs) are used to deliver components to work-cells around the factory floor, airborne drones are used to perform inventory checks in logistics centers, and underwater drones to inspect and maintain subsea infrastructure, such as bridges and telecom cables.

Thanks largely to the ambitions of electric vehicle makers, battery technologies are likely to make further advances, and this will enable even more mobile automation systems.

While early automation systems used intricate mechanics to create synchronized movements, in the 18th century the concept of programming was developed to control weaving looms. The looms used strips of paper punctuated by a sequence of holes, and 200 years later early computers known as “adding and accounting machines” still used essentially the same concept: instead of paper strips, the machines read instructions from “punched cards”.

Punched cards were superseded by magnetic tape and later discs, and eventually were made largely obsolete by solid state memory (DRAM and NAND). But regardless of the media used, floppy disc or DRAM, the machines all ran on pre-defined instructions and once set in motion would continue to run until switched off, or an error occurred. A modern robot programmed to weld car doors, will continue the welding sequence whether a car door is actually in front of it or not. This makes it dangerous. What if someone walked in front of the robot or, due to a problem further up the production line, the car door is not in the right place at the right time.

Over the last ten years, increases in the speed of processors have made it possible for automation systems to adapt to changes in surroundings, simply by building a library of different scenarios. In one scenario, if the car door is not in the correct position (perhaps determined by a vision system from Keyence or Cognex) the robot might pause its operation, and in another scenario, if someone walks too close to the robot (perhaps defined by a laser-based virtual safety fence from TI or Hexagon) the robot might slow down its motion or perhaps stop altogether. This approach provides the system some autonomy, but clearly the degree of autonomy is limited by the number of pre-programmed scenarios available to it.

More recently, advances in AI technology, in particular machine learning, afford automation another significant step-change. In fact, machine learning may prove to be as significant to automation as the introduction of electricity in industry 150 years ago.

With machine learning algorithms systems can learn by example or identify patterns and anomalies by themselves or through trial and error. This process can be accelerated by simulating millions of different scenarios virtually, in software. As this field advances, automation systems are likely to become more autonomous, able to adapt and respond appropriately to changes in the environment around them. This will make them easier to use, safer to work with and more capable of performing a wider range of tasks – not just in physical tasks, but also in cognitive challenges such as problem solving.

As a result, the commercial opportunity for smarter, more autonomous automation systems is likely to be significantly larger than the niche market that has been established by their “mute and brute”¹ predecessors. We therefore believe these technological advances will create a large and multi-decade-long opportunity for the patient investor.

While smarter automation is likely to provide a great leap forward in productivity growth for the global economy, address issues of labor shortages, and allow people to avoid dirty and dangerous tasks, could the same intelligent systems be applied to solve major challenges of our time, such as mitigating climate change, finding cures for chronic disease or solutions to tackle overcrowding in cities and wealth inequality?

This may be the future, but today’s AI systems are still not advanced enough to tackle such complex, multifaceted, and interdependent problems. Some early progress has, however, been made. One of the stand-out and successful AI applications so far is AlphaFold, developed by Google DeepMind², which produced an accurate estimate of the 3-dimentional structure of 200 million proteins. Google has made the database publicly available, giving researchers a deeper understanding of protein architecture and its implications for biological function. Prior to this, only 200,000 protein structures were understood. Kudos to AlphaFold! We believe that innovation leads to further innovation and that this process is naturally accelerating. We remain hopeful that significant new breakthroughs will follow.

¹ ”Mute and brute” was a phrase coined by Andrea Thomaz, co-founder and CEO of Diligent Robotics Inc. Cited in National Geographic magazine, “The robot revolution has arrived”, August 2020
² AlphaFold (n.d.). AlphaFold Protein Structure Database; accessed on 21.11.2024