Ever since “Unimate,” the world’s first robot arm, hit a GM factory floor in 1961, robotics companies have focused their time and innovation solving factory problems.

It’s not hard to understand why. After all, the Unimate cost $350,000 in today’s dollars to build, and the best trick it could muster for a live “Tonight Show” audience was pouring a beer into a glass. Very few companies could justify spending that kind of money on a machine capable of a handful of crude, pre-programmed tasks.

Over the past decade, though, that’s all changed. The new economics of production we described in Part 1 of this series means that startups are producing far more sophisticated machines — for far cheaper — than ever before. This allowed founders to follow in the footsteps of software developers and adopt a subscription model, robots as a service (RaaS), that significantly reduces the cost and burden for customers.

As a result, founders greatly expanded the universe of potential customers — and use cases — they could pursue, solving problems not just in factories but in hospitals, office towers, parking lots, retail stores and dozens of other settings.

Scaling faster

Today RaaS is doing for robot makers what SaaS did for software developers: expanding the market and speeding innovation.

The market for robots as a service (RaaS), which hardly existed a decade ago, is projected to grow to a $35 billion industry by 2022. Markets ripe for disruption are big ones: agriculture (click to read our views), industrial cleaning, healthcare (click to read our views), building security, warehouse management, and retail.

Reflecting the enthusiasm for this industry, Amazon, Google, Honda and Deloitte all announced their own RaaS platforms in the past several years.

Just as Microsoft no longer sells software out of a box with installation instructions, robotics companies no longer sell their machines to customers and wish them luck. Rather, they offer a service for a modest monthly fee and then deploy their robots to provide the service, whether it’s cleaning, delivering, sorting, packing, picking, or a growing number of other tasks. 

Smaller companies can now make a business case for adopting automated technology that they couldn’t make just a few years ago. And industrial clients — long the backbone of the robotics industry — now have more flexible options to choose from, such as per unit pick and per hour pricing.

In 2001 — five years before Google CEO Eric Schmidt introduced “cloud computing” into the lexicon — Paul Graham wrote a blog post outlining the many advantages of web-based software: “You should get new releases without paying extra, or doing any work, or possibly even knowing about it,” he wrote. “Over time applications will quietly grow more powerful. This will take some effort on the part of the developers. They will have to design software so that it can be updated without confusing the users.”

This aptly describes the experience of most customers using robots today — a notable departure from just 10 years ago, when buying a robot looked a lot like buying a car from a dealership.

Operations experience required

At the same time, RaaS places new burdens on founders.

Building the robot is just one part of the solution. Now companies must develop a robot and turn it into a service. This requires them to solve a whole host of new problems:

• How do I manage this service? 
• How do I deploy robots into the field? 
• How do I install them?
• How do I monitor and repair them? 
• How do I update them?

This transition reminds us of the one we experienced at Yahoo! When we first launched My Yahoo!, the entire company ran on two servers. As we scaled and added millions of users, we began operating entire data centers, which required a team of engineers to run. We built developer operations (dev-ops) teams to run the data centers, so that we could run the data centers remotely. 

Companies that are successful scaling with RaaS will confront similar operations challenges as they expand their businesses. They will need to figure out how to deploy, monitor and service an ever-growing fleet of robots while providing a seamless experience for customers.

All of those things that go into making a service run reliably over an extended period of time require them to focus on operations more than ever before.

Real-world learning

Another aspect that is different in robotics is the advent of machine learning. Robots that can learn and adapt, that can operate autonomously can exponentially increase their value proposition in many more use cases.

The Watson supercomputer that won Jeopardy in 2011 is a much better representation of today’s automated machines than GM’s beer-pouring Unimate. Though Watson performs no physical tasks, its ability to learn and process information represents the real value of these machines. 

A robot that can pick up the same size and shape boxes and move them to pre-programmed locations is far less useful than a robot that sees and adjusts to its surroundings. Using sensors and, in many cases, computer vision, a robot collecting massive amounts of data can leverage machine learning in ways not imagined 15 years ago. 

This speaks to another key advantage of RaaS. Because robot makers manage their fleets in the field, they collect data from machines deployed in the real world. This allows them to iterate faster, fixing bugs and adding new features. The key to making that AI successful is collecting real-world data—they go hand-in-glove.

Why we love Everest Labs

This is one of the reasons we got so excited about Everest Labs. The founders had already convinced several large recycling plants to let them deploy their picking robots. By deploying directly in recycling centers and using computer vision and other sensors, they are directly in the stream of a constant flow of information: endless conveyor belts of refuse and recycling. 

By being exposed to and analyzing a constantly moving and changing flow of items in every possible configuration, they benefit from their exposure to edge cases that continuously improve their machine learning models and recognition/decision algorithms. And as their robots pick out what can be recycled from what cannot, the feedback loop adds to their knowledgebase and further improves the system. 

Today, they are already handling jobs that many workers consider too risky to take on (e.g., the risk of needles and the related extensive downtime required if they are pierced by them is real). So in many ways, Everest Labs represents a great example of a “win-win” for all stakeholders involved.