Working on robots is fun.  And building a robot that touches people’s lives and eases their burdens is satisfying in great measure.  Such incentives lead many to consider a career in robotics.  Because of my long residency in the field, aspiring roboticists often ask me questions like:  “How can I get into robotics?”  “What should I study?”  “What degree do I need?”  

There are no cut and dried answers.  Many paths lead to a rewarding career in robotics.  But there are some subtleties and cautions you should appreciate as you make your plans.

Passion versus Interest

As a child I was captivated by all things scientific and technical.  Given my fascination, it was inevitable that I’d find some way to work in those fields.  But I stumbled for years trying to figure out exactly how.  (See The Myth of the Planned Career.)  It wasn’t until seven years after graduating college that I discovered robots.  I’ve been obsessed with them ever since.  So, if you haven’t located your passion, don’t despair, you may find it yet.  But there’s hope even if you don’t.

Once a student asked me, “What if I’m interested, but don’t have a particular passion?”  The talk I’d just given, heavy on my consuming passion for robots, had probably caused my listener concern because she had no such obsession.  Not thinking well on my feet, I failed to answer her question reassuringly in the moment.  But I thought about it considerably afterward and realized that there’s an upside to interest as opposed to passion.  Someone who’s passionate may suffer if circumstances deflect them from their calling.  While someone who’s broadly interested can be happy with any of a variety of careers.  

Improve versus Invent

Engineers make things better.  Typically, engineers focus on improving existing products or processes, and in that role they can be brilliant.  Each engineering iteration of a product can boost efficiency, hone functionality, and reduce cost.  Everyone appreciates a well-engineered product.  The flip side of incremental improvement, inventing a first of its kind product, is a risky and unpredictable proposition.  But every new robotic application requires such a roll of the dice. 

Robotics needs practitioners proficient in incremental improvement and others adept at invention.  If you choose the former, more customary course, you’ll find yourself in good company on firm ground.  Data from earlier versions of a product can guide your design choices for an improved version.  Applicable theory will be found in textbooks, marketing surveys can inform your decisions, and veterans of earlier versions can answer your questions.  Your diligent efforts will nearly always be rewarded.

With invention, you’re on your own.  No textbook or manual describes how to build something that’s never been built before.  Guarantees that it’s even possible are unavailable.  In attempting to birth such a thing, you will make the mistakes that those who come afterwards will learn from.  If customers have never seen, let along purchased, your invention there is no assurance that they will.  Even if your product works exactly as you intend and everyone who hears of it proclaims their desire to buy one, it can still flop.  Worst of all, it’s possible to spend an entire career inventing but never score an unequivocal win.  So, if invention is your passion, it’s important to value the journey over the destination.  

Preparation

Incremental development favors depth, invention wants breadth.  

The more deeply you know your discipline the greater will be your ability to devise sophisticated solutions within that discipline.  Once the basic design of a robot is settled, its pieces can be optimized—the mechanics refined, control systems improved, the user interface made more intuitive.  Perhaps more than most products, autonomous robots require the confluence of many different disciplines.  Whether electrical, mechanical, software, systems, or other, which field of study you should master depends only on your interests.  No one discipline is more essential than another.

Invention comes before the robot’s design is settled.  To determine from which pieces and in what form the robot should be constructed, a broad, general understanding of the possibilities is necessary.   Deep knowledge of any one discipline is not needed, but a good appreciation of what is possible in each area is critical.  Practical experience as a tinkerer and dabbler in many fields can be helpful preparation for the robot inventor.  Anecdotally, some of the most inventive roboticists I know cast about for a time, trying different things before settling on robotics.

The best way I know of to become proficient in robotics—to understand how sensors and mechanisms interact with the real world—is to build sensors and mechanisms and have them interact with the real world.  If you would invent definitely study, but build as much as you study.

Inventive Process

A robot begins as a concept that survives the great filter.  The robot inventor must ask, of all the tasks that people want done, which ones are most appropriate for robots?  The answer is complex because robots are not competent at the same things as people.  To date, it hasn’t been possible to simply pick a task (perhaps for business reasons or because of great need) and then build a robot to perform that task.  Most tasks are filtered out, snared by abilities robots don’t have.  The plant-moving and home garden-weeding applications we chose for Harvest Automation and Tertill Corporation respectively, were both surprises that made it through the filter.

With a suitable task chosen the exercise becomes one of satisfying constraints.  The process is fraught because task constraints are often contradictory.  Which of several potential methods can be made to work—which will satisfy all the constraints at an acceptable price—is rarely obvious at the outset.  Development commonly requires guessing which method might be easiest to implement before you know for sure.  The existence of an effective, well-known manual method is of little help—and may even distract—if it’s inappropriate for a robot.

Two examples:  When Kiva System set out to build a warehouse robot, the obvious solution would have been to replace the human worker who walked the aisles picking ordered-items from bins on the shelves with a robot that did the same thing.  Research into robotic bin-picking had been in progress for years.  But bin-picking remained a research topic; putting it on the critical path to a business solution would have doomed the project.  It turned out to be much easier to figure out how a robot could bring a pod of products to a human picker than it would have been to solve the research problem.  Transporting heavy pods around the warehouse would have been folly had it been done by people, but it was exactly the right approach for robots.

Roomba’s carpet sweeping mechanism is another example.  Earlier attempts to build robot floor cleaners started with a powerful vacuum.  But the energy demands of that system violated the reasonable constraints that the robot should crawl under furniture, clean effectively, and run for a long time.  One of Roomba’s innovations was to reject the obvious vacuum solution and instead use an energy efficient carpet sweeper mechanism to accomplish the bulk of the cleaning.

Price constraints are one of the issues that make robots, especially consumer robots, so challenging to design.  Marketing can inform the tradeoff between functionality and price.  But success requires a robust give and take collaboration between the marketing and technical disciplines.  Absent that you’ll end up with either a high-tech design no one can afford or an appealing, comfortably priced robot that can’t be built.  So, another path into robotics is to become that rare person who can bridge the gap—someone fluent in the languages of both marketing and technology who understands the constraints each speciality faces.

Over preparation?

I found when I worked at iRobot that highly prepared teammates, those with PhDs, didn’t necessarily move an inventive project forward more reliably than other teammates.  It seemed to me (I could be wrong!) that all the hoops one has to jump through in the intense academic environment of doctoral preparation produced a certain mindset.  Specifically, the PhD-holding folks I knew were most reluctant to solve a problem in any way that the reviewer of a scientific paper they might later write would flag as poor form.  

Reviewers demand solutions that can be measured and evaluated.  As they should!  It’s very much the legitimate purpose of research to determine the characteristics and suitability of a proposed solution.  But all a robotic product has to do is work.  Mixing solutions in ways that makes them hard to evaluate separately is appropriate if they make the robot work.  Keeping reviewers of potential scientific papers happy is an unnecessary constraint in this case.  (After the robot was invented, I’m sure our PhDs would have been great at making it work better.)

The highly trained person is highly competent to work within a narrow niche.  If the best solution to a problem falls withing that niche, there’s no better candidate to solve it.  But if the optimal solution lies outside that niche, or straddles niches, then deep, highly specific knowledge becomes less of an advantage.   

What do roboticists do?

I met a theoretical physicist once who told me that a lot of his job involved drawing on white boards while arguing with his friends.  My day-to-day tasks are much more varied than that.  Especially when working in a startup, a roboticist must do a little of everything.  Here are a few of the (mostly) fun things I’ve done as a roboticist.  

I looked for tasks robots might do.  I constantly mined news stories and videos I ran across for possibilities.  Once I spent a week or so going through every job listed by the Bureau of Labor Statistics.   In the job descriptions I systematically looked for inspiration—but I found no revelations there.  As described in Harvest Automation, Part 1, we happened upon a nursery and greenhouse industry application called “spacing” that perfectly dovetailed with robots’ abilities. 

I always tried to learn as much as possible about the domain of the robot I was trying to design.  When we began working on Tertill I wanted to have the same experience with the robot as a customer would have.  So, I put in my first garden ever not to provide the family with fresh vegetables but to facilitate building a robot.  Now I find I like having a garden—robots can expand one’s horizons!

I investigated ways to solve problems.  Once we decided to build a robot that weeds home gardens, we needed to a way to remove weeds from the ground.  I experimented with a pair of rollers that would pinch weeds and pull them from the ground and with a reciprocating mechanism using a jigsaw blade.  But my colleague came up with a simpler more satisfying mechanism.  (See Hatching Tertill.)

I developed sensors and methods to accomplish things the robot needed.  Our Tertill robot needed a way to know when it collided with something its capacitive sensors couldn’t detect.  To figure out if physics could help, I programmed our prototype to drive forward while making a high-resolution recording of the signal from the robot’s accelerometer.  Then I ran the robot into walls and softer things that decelerated it slowly.  I build a spreadsheet and analyzed the collected signals in many different ways.  Eventually I found a reliable solution that we programmed into the robot.  (See Tertill: Under the Shell.)

After something is invented—or often while you’re still in the process—it’s advisable to file a patent.  Creating the patent usually involves writing out your ideas in a clear and concise way and then delivering them to a patent attorney.  The attorney then translates what you’ve written into patent-speak—typically making the ideas almost indecipherable to non-attorneys.

When developing a new robot it’s almost inevitable that issues and problems—unforeseen at the start—will arise.  One such problem with outdoor-living Tertill was dew.  Sensitive electronics and moisture do not go together.  To avoid reliability issues I set about to understand condensation, how it might affect our robot, and what we could do to minimize any problems.   Over the course of a few weeks, through study and experimentation, I came to understand how dew works—something I hadn’t fully grasped during my undergraduate days in physics.  That comprehension let us develop mitigation strategies.

A few other things I’ve done as a roboticist include building a sonar corner reflector for a grape cart project we did at Harvest Automation, speaking to students of various ages about robots, and doing a guest spot on a Home Shopping Network show promoting Tertill. 

The activities listed above barely scratch the surface of the many things roboticists do.  But I think they give you the flavor of the profession.  And one more item.  Like theoretical physicists, we roboticists sometimes stand in front of white boards and argue with our friends.

Whether you choose to dance with the robot or another partner, enjoy the music and the moves.  And good luck.