Hi, I’m Antonio, founder of r4robot.
I’ve been thinking a lot about intelligent robots. Especially about how they might transform manufacturing. Let's explore some ideas here.
Very broadly, manufacturing assembly lines today follow a model where a complex multi-step process is split into different stations. At each of these stations, just a few simple steps are performed. Goods are built up as they progress through the long chain of stations and at the end of the process, we end up with a big pile of stuff.
This is the best setup we’ve come up with that balances quality, speed, and cost. It allows us to make quality goods quickly at affordable prices. The affordability comes from exploiting economies of scale so that costs are spread over a large number of goods. To ensure that factories are profitable, they impose minimum order quantities.
Forecasting and Overproduction
One issue with this model is that producers must accurately predict how much product they can sell. If they make too much, products are stuck in warehouses, eating up storage costs. If they make too little, they’re not capturing as much profit as they could. Innovations like just-in-time manufacturing address this issue, but overproduction and waste remain big challenges.
Opaque Supply Chain
Stepping back from our single factory and considering its place globally, we can see that each of our factory’s inputs is likely produced by a similar process in a different region of the world. Multiplying this out, this results in an increasingly complex and opaque supply chain. For a manufacturer at one end of the chain, it can be impossible to be sure of a sustainable and responsible process just a few steps back.
In our globalized world, the economics of manufacturing have resulted in categories of production concentrating in particular regions of the globe. As we’ve seen during the pandemic, these bottlenecks can wreak havoc for local economies, with scarcity of essential products, price surges, and supply uncertainty.
So — there are a few problems with the way we manufacture goods today across the world. Globalized supply chains are an astonishing human achievement, but they have their problems, and we have some interesting opportunities to address them.
Assembly line manufacturing today is still largely driven by human labor. Mostly humans, some robots. While we may be tempted to think that modern factories are run by rows and rows of robots as in some automotive factories, this isn’t the case across all products and industries.
It may not be entirely clear why this is the case, aside from cost considerations, but I’d like to propose a theory. Consider two broad categories of products. On one side are products whose design doesn’t change or changes very little year after year. These products can take advantage of current automation technologies, since the automation tools don’t need to be dynamic. The design of a can of soda isn’t likely to change by the season, so it makes sense to invest in automation.
On the other side are products that undergo frequent design changes like consumer electronics or apparel. Because today’s automation technologies are best suited for static repetitive tasks, investing in automation here doesn’t make sense. The robots can’t easily adjust to changing designs.
So this is the state of much manufacturing today. Mostly humans, some robots.
As we continue to introduce robots into assembly lines, one risk is that we maintain our current manufacturing models and simply automate them with current automation technologies. In doing so, we would be missing out on opportunities opened up by upgrading our automation tools and introducing intelligent robots into assembly lines — robots that can quickly and reliably adapt to changing designs and processes.
Critically, the assumption that robots are a one-to-one replacement for human labor misses the important point that introducing robotics to a process is not only a change in labor, but a change in process. Robotics introduces its own inefficiencies and surfaces new issues that might come from the redesign of any existing process.
Still, intelligent robots offer massive opportunities to transform our manufacturing processes for the better. For instance, instead of executing a few simple steps at each station, we could combine stations and perform multiple steps at each new station. We could start to collapse the assembly line into fewer stages, shrinking the factory, reducing real estate and production costs.
Additionally, intelligent robots could perform different tasks at each station depending on the input. This would allow for different but similar goods to be produced on the same line, simultaneously, at scale. In this way, we could start to drive manufacturing from operating on economies of scale to economies of scope.
So, manufacturing today operates on this model where big factories pump out large volumes of identical goods. With intelligent robots, we could start to move to a model where smaller factories produce a wide range of goods.
Further, what one robot learns, all robots learn. So we could imagine having a network of small factories across the globe that learn from each other in real time. Each factory could produce a wide range of goods instead of mountains of a single good. They could do this close to the final customer in a sustainable and transparent manner.
Now, intelligent robots bring massive opportunities for reshaping global manufacturing, but, this is all happening within a social context of growing inequality and a political context of rising tensions between global powers.
There are multiple dimensions to consider here, and businesses need to operate across all of them. Each one of these dimensions - cultural, social, political, and technical - is a discussion that merits its own dedicated discussion.
For now, I want to focus on the technical side and dive into considerations around how we might design these machines.
It’s worth mentioning that we’re very far away from this vision of automation that I’ve laid out. We’re at the very beginning of making robots that can learn and adapt to their environments.
This is an area that mostly exists in academia and research labs and is just starting to seep into industry. Often, those robots we do see out in the world have been trained behind a curtain and emerge with a given set of abilities.
At r4robot, we’ve been exploring ideas around how we might design interactions with robots that learn from us. If we want to develop robots that are easy to train, we need to design the interactions and systems that allow us to train them.
After all, making advanced robotics accessible, intuitive, and widespread is what we're all about.