The Role of Robotics and Automation Systems in Shaping the Future of the Manufacturing Industry
The advances in computing and consumer electronics over the last twenty years along with the emergence of the internet have catalysed a new wave in technological innovation and enterprise¬¬¬. The commoditisation of electronics and computing hardware has engendered a rapid shift in what was once considered high-end, industrial grade technology to the public domain. It has allowed new companies to spring up and innovate at an unprecedented rate in what could perhaps be called the 21st century start-up model. The combination of these factors has resulted in the emergence of the Internet of Things (IoT) which is the principal underlier for what is being termed as Industry 4.0, or the 4th Industrial revolution . Industry 4.0 is the view of the modern-day factory as network of smart, interconnected nodes, wherein the nodes take the form of any equipment that adds value to the final product and that can be made accessible through the network (the IoT).
The overarching goal at a factory is to extract as much value as possible from as few resources as possible. To this end, production lines are primarily optimized to minimize two variables: takt-time and down-time . While there is a whole range of operational and workplace management techniques (including 5S and six sigma) aimed at addressing the former, the latter is much more prone to vicissitudes of the supply chain and therefore much harder to optimize. This issue is finally becoming addressable through Industry 4.0; real-time tracking information on incoming shipments and their estimated arrival times, allows the production line to schedule operation accordingly. A similar principle can be applied to part movement within a factory – network connected supply carts communicate with the production line to relay information about required parts and their estimated delivery time to the associated work-stations line-side. The tracking of such supply-carts also allows for real-time optimization of the flow of traffic within a factory . While this system is only at its nascent stages of implementation, as the infrastructure proliferates, the paradigm for part movement and warehousing itself will undergo an upheaval - replacing human driven carts and fork-lifts in favour of their autonomous counterparts.
This trend was first evident in the technology developed by Kiva Systems for automated control of warehousing, and its utility made equally clear when Amazon Inc. bought the company for $$775 million, in March 2012 . While Amazon’s decision to dedicate the Kiva technology for use solely in Amazon warehouses could have been a step back for the field, the 21st century start-up model jumped in to bridge the gap; companies like ClearPath Robotics, whose early core technology was based on consumer grade electronics, are working on systems that automate part movement across factory floors and warehouses, with the aim of making Amazon’s proprietary technology an industry standard . The economic benefit of automating these well-defined tasks in controlled environments (warehouses and factory floors) makes it easy to extrapolate the adoption of this technology over the next ten years to a stage where it would be unfeasible for a mass-producing manufacturer not utilise it. What is much harder to address and predict is the automation of tasks that are equally mundane which nevertheless require much greater dexterity and skilled labour to accomplish. Installing parts onto the chassis of a car for instance requires a level of cognition that is currently hard to attain in robots. These tasks access a specific facet of human learning termed as reinforcement learning.
Reinforcement learning (RL) is so innate to humans that we are scarcely aware of it. We subconsciously learn skills such as walking or catching a ball which are incredibly complex to encode in a robot. Empowered by the advances in computing capability over the recent years, however, we are beginning to apply RL to machines and producing results which surprise even their creators. Google Deep Mind’s AlphaGo computer (based on RL) for instance, beat the South Korean champion Lee Sedol at the Chinese game of GO which is known to be unsolvable using traditional brute force methods (based on current day computing capabilities) . This approach might prove pivotal in the next generation of automation technologies in manufacturing.
Research on Deep Reinforcement Learning applied to robot manipulation utilises similar techniques as Google’s AlphaGo but applies them to allow robots to understand the dynamics of the physical world and to interact with it in a dextrous manner . Such techniques would allow robots to accomplish our earlier example of installing parts on a car, which might have been deemed undoable.
Over the next ten years, it is improbable that robots will be capable to human level dexterity and understanding of the physical world. What is more likely is the emergence of collaborative human-robot environments where machines in a sense act as a human’s assistants or counterpart in performing repetitive tasks. Baxter by Rethink Robotics is one such robot which is at the leading edge of this trend . Although Baxter is currently based on learning through human demonstration and does not as such utilise RL, with its focus on human-friendly collaborative interaction and two high degree of freedom dextrous arms, it is ideally poised to be ubiquitous on factory floors over the coming decade and capitalise on the advancements in RL .
Lastly, an aspect of robotics that is often overlooked, since it takes a step away from the physical world, yet, is equally important in manufacturing, is machine learning applied to predictive analysis and optimisation. Machine learning algorithms for instance could make time consuming diagnostics on faulty components unnecessary by suggesting a “most likely” failure mode based on some learned parameters. In the remanufacturing department of a car company, this could mean not having to physically take apart faulty components to identify the underlying problem, thereby saving material and labour costs.
Machine learning could also be applied to optimise the performance of the factory and reorganise workflow to save resources. This application was demonstrated by Google when they used their DeepMind AI to reduce the power consumption of their data-centres by 40% . Such demonstrable economic benefit is a sure-shot indicator of the wide-spread proliferation of this technology as the new generation of engineers and scientists move into industry over the next ten years.
Any discussion of automation is incomplete and irresponsible unless it addresses the potential impact on the human work-force. Since the first industrial revolution, advancements in technology that questioned the position of a human work-force have engendered controversy and in most cases, been dismissed as the work of luddites, and to some degree, rightly so. With Industry 4.0 and the advancements in AI, the situation this time is different. While the short-term (10-year timescale) role of robotics and automation is expected to be hugely economically beneficial, in the long term, it could lead to a range of issues, from inequality to unemployment and an eventual economic downturn . In creating a ‘Partnership on AI’ leading tech-corporations such as Facebook, Google and Amazon, realise this eventuality and are conscious about ameliorating it. Given that the pace of legislation is rarely able to keep up with industrial advancement (especially when vast economic benefits are to be reaped), partnerships such are these are our best bet to prevent a reversion to the social divide and stagnant economy of the feudal ages.
Written on July 1st , 2019 by Noorvir Aulakh