All posts by Eric Schweikardt

I love Cubelets. I feel like my identity is pretty closely tied to those little robotic cubes. You might have discovered Cubelets just now, or maybe you’ve had a kit for a year, but I’ve been working on Cubelets since 2006. That’s a decent chunk of my adult life! It feels like it’s been a really long time since I’ve been able to show off something brand new, so it’s particularly exciting for me to introduce you to MOSS. MOSS is a brand-new construction kit for building dynamic little robots.  We’ve been working on it in secret for almost three years and we’re launching on Kickstarter. The video does a far better job of explaining how it works than I could here, so please take a look!

We’re announcing something awesome on November 7. I promised the team that I wouldn’t announce it early, so consider this a meta-announcement or a second-order announcement. No further announcing follows. Well — we’ve been so focused on details and finishing touches recently that the origins of our new thing have faded a little bit. The story of how it came to be is worth telling. Back in 2009-2010, some of our startup funding came via SBIR grants from the National Science Foundation. These grants are free money, but they come with some unanticipated benefits too. About three years ago, I got email from the American Society of Engineering Education announcing their new post-doc scholar program. As a NSF SBIR recipient, the ASEE would pay for almost all of a postdoctoral scholar’s salary and health insurance if we hired someone who had recently finished (or was about to finish) their PhD in a technical field. I immediately thought of Jon Hiller. When I was a post-doc at the Cornell Computational Synthesis lab (now called the Cornell Creative Machines Lab), Jon was a PhD student. Even in a world-class lab filled with the smartest people you’ve met, Jon stood out. He was working on discrete 3D printing, figuring out how to deposit tiny beads of material in dense grids to create forms with variable material properties. He built apparatus, wrote code, ran 3D simulations, and finished some super cool research. I wondered if he was getting ready to defend. I called Jon. He was about to defend his dissertation and was ready to accept a job at Lawrence Livermore National Lab. It’s hard to compete with a cush government job but modbot had a trump card: Jon’s a rock climber and we’re in Boulder, CO. Proposals were written and he moved out here a month later to start. Under normal circumstances, there’d be no way that a little, underfunded startup like Modular Robotics in 2010 could embark on the design of a second, totally unique robot project. But when Jon came out to join us, much of the work on commercializing Cubelets was complete. Since Jon was basically “free” to Modular Robotics, we decided that we should make the most of this bonus: we’d put him to work on a brand new something. Here we are, a couple of years later.

Thanks everyone for your calls and emails.  We’ve got some amazing flooding going on right now in Boulder but we’re all OK.  A few modbot people are out of their homes and have flooded basements but modbot HQ is dry with very minimal damage.  We expect to be shipping Cubelets out the door again on Monday. Everyone knew that this was coming sooner or later.  Our last 100-year flood was in 1894, and Boulder is basically built in the flood plain.  To be clear, 100-year floods don’t occur every hundred years, it’s just a funny way of labeling a flood that has a 1% chance of occurring each year.  Regardless, Boulder’s in a low spot at the mouth of many canyons that drain the Continental Divide, it’s a pretty likely spot for serious flooding.  The weird part of this, though, is that flooding doesn’t normally happen in the Fall, it happens in the Spring, when all of the snow melts and fattens all of the rivers and creeks. It’s still raining.  I’ve been in and around Boulder for almost 20 years and I’ve never seen it rain continuously like this for days and days.  It generally sprinkles in the afternoon for 15 to 20 minutes during the Summer, but a grey day of rain is a once-a-year novelty.  People here look forward to it, we get all Seattle and drink coffee and watch movies during the day.  But it’s been raining here for a week. I catch a little flack around Modular Robotics sometimes because I can get get pretty excited about charts.  Some people can look at huge spreadsheets of numbers and make sense of them, but I generally cannot.  I usually think of data as an input, and a chart as the output.  Well, anyway, take a look at the rainfall chart above! I’m stuck in Nederland, the little mountain town West of Boulder where I live; all of the roads in and out are closed.  We’re at 8,236 feet above sea level, so all’s fine here, all of our rain just washed down to Boulder.  It feels a little odd not being able to leave, but we have ample stores of coffee and chocolate and also a big vegetable garden, so we’ll be just fine.  It just started raining again. If you’re interested in reading about the meteorology behind the current flooding, Bob Henson wrote a great piece on the UCAR/NCAR site.

I’ve been trying not to travel too much this Summer. There’s always a lot to do at Modbot, and in general I’ve tried to spend a little more time riding bikes in the forest and sharing dinners with friends instead of waiting in airports and hotels. I jumped at one interesting invitation, though, and went to Foo Camp in August. It’s been a few weeks and I think I’ve finally calmed down enough to reflect on it. O’Reilly Media has been organizing and hosting Foo Camp each year since 2002 at its lovely little office/farm complex in Sebastopol, California. It’s a three day unconference: there’s no agenda and the attendees figure out what to do as they go along. O’Reilly invites about two hundred people and provides space, food, and drinks at the Foo Bar. The general idea is that putting a bunch of smart people together will probably result in interesting discussions and collaborations and O’Reilly probably hopes that some of these will result in new books or conferences.

There was a funny pervasive insecurity among new attendees; people wondered why they were there. During the second day, I started to figure things out. This year, attendees seemed to be working predominantly in three areas: synthetic biology, government hacking, and hardware. I fit pretty clearly into hardware and felt a little more comfortable, but the people making glow-in-the-dark plants kept me guessing. Foo Camp was intense. Wake up in an orchard, expand your mind for 18 hours, pass out and repeat. It turns out that I can only handle about 12 intense, stimulating conversations in a row before I start to shut down so I made liberal use of the little forested paths in the area to spend a few minutes alone clearing my head. I could ramble here for quite some time about the events of the weekend, but I’ll try to give this a focus and just mention the Local Manufacturing session. Sessions are set up on the first evening. We arrived, ate, and then milled around a series of huge whiteboards to try and decide what to do with the weekend. A grid of seven locations and 20 time slots provided for a total of 140 sessions, and the group filled them up in just a few minutes. I put up a session called “Local Manufacturing: Tech, Tools and Strategies for Making Stuff” for Sunday morning and hoped someone would show up.

Someone showed up! There’s Z Holly, founder of TEDx. And Mike from Otherfab, who makes the rad little PCB mill spun out of Otherlab. And Adam, one of the Makerbot founders. Nick, who makes hardware design tools, Angie who’s working with girls and code, Heather, who’s thinking about hardware at O’Reilly, Alex from the Media Lab whose little cardboard robots made 200 people laugh at the same time. Modbot hero Chris Anderson showed up! I can’t really say that we accomplished anything aside from getting ourselves excited about USA manufacturing. We all told our manufacturing stories, I showed some pieces of FARKUS, our robot kit for factory automation, we talked about how valuable it is for designing (engineering) and making (manufacturing) to be colocated, and BAM!, time was up and we needed to head to a session in the Lemur tent for Luddites Working in Technology. Thanks Tim and Sara and everyone.  I had a very nice time.

Molex is a big, 75-year-old company that makes electronics connectors and other little fiddly bits that you’ll find in iPhones, network gear, and consumer electronics. Today, Molex announced that they were going to be acquired by Koch Industries. If you don’t know or care who the creepy Koch brothers are, then super, we’re done here. But if you do, I thought I’d mention that Modular Robotics uses no Molex connectors in any of our products, and after today’s news we’ve written that rule into our design guidelines for all future products.

I just returned to Boulder after a lovely week totally unplugged in Panama.  First off-grid week since 2011!  We snorkeled and surfed and ate fish and fruit.  I read books and relaxed, and barely thought about work at all except for one anagnorisis that happened in the middle of the night. I woke up and looked out the open sides of our little thatch-roofed cabin to see the moonlight flickering on the ocean swell as it rolled lazily onto the beach.  The thought occurred to me that Modular Robotics was happily functioning while I was gone.  We’ve put together such a great team that I’m not even really necessary for day-to-day functioning of the company.  I can leave, turn off my phone for a week, and we keep pumping out Cubelets. If I think about our company as a little train, we’ve built the train cars, loaded up on fuel, figured out how to mix drinks in the bar car, and left the station.  It will happily chug along for a while without the need for me to walk up and down its length chatting with passengers and crew and patrolling for problems.  I suppose that I could even just get off the train more often, and I probably will.  The realization here was more profound, though.  With the train carrying on smoothly, I’m free to head to the caboose, pull out some maps, do some reading, and determine where I think the train should go. I’m the CEO.  It probably seems obvious to you that my role should focus primarily on strategy and direction.  While that seems fairly obvious to me in the abstract, it took a little time away and a different setting to make it seem real in the present.  I think the reason for my slow uptake has everything to do with our growth.  When there were two of us, I designed circuits and sourced magnets and wrote code.  When there were ten, I answered questions and did a lot of hiring and wrangled our bookkeeping.  Strategy and direction happened at interstitial times: at night, over a meal, or during a flight.  Over the last couple of years we’ve made hires sequentially, and each new team member has taken a role that I muddled through, and filled it fully and expertly.  Now that the train is rolling, our next hire is someone to focus not on operations, but on strategy, new products, team building, and design.  That hire is me!

Thanks for the photo, Sawyer!

We just made a fairly monumental decision that almost everyone in the toy industry will tell you is asinine. We decided to build a big factory in Boulder, Colorado, and manufacture all of our products ourselves. Electronic stuff is mostly made in China. It’s been that way for a while. Modular Robotics has a tiny factory here in Boulder, but it seemed obvious to almost everyone that as we scaled up to making millions of tiny robots each year, we’d move manufacturing to a contract manufacturer (CM) in China who would make our stuff and send us pallets of shrink-wrapped robot kits. We have a lot of parts made for us in China. The stamped metal Cubelet connector pieces are made in Wuxi, and the plastic Cubelet shells are injection molded in Zhuhai. These parts get sent to us via UPS and we build the electronics and assemble and test everything in Boulder. It’s not trivial to get things made all the way across the world, so I’ve been visiting China once or twice a year for the last 3 years to oversee production, solve little problems on the factory floor, and audit our factories. During my most recent trip in December 2012, I visited three different CMs to begin figuring out how we might have them manufacture Cubelets (and our next product) for us. It was a crazy few days. One CM’s mould making shop had a dirt floor. Another had ten thousand people working there and we drove through the campus in a golf cart. One had lighting so bad I could barely navigate, another had a museum of products they make that included almost every toy you might think of. But one difference between the factories struck me: at the low-end factories (they call them Tier 3), there were people everywhere. 3 workers ran each injection moulding machine, placing inserts, removing parts, trimming flash, and touching up funny white streaks with a hair dryer. The mid-level factory had fewer people, they seemed to simply have labor that was better organized to do several things at once. But at the high-end, tier 1 factories, there was nobody around. The moulding machines whizzed and klunked along on their own, aided by robotized jigs that removed parts and filled bins that ran on tracks. One worker supervised four SMT lines, simply maintaining the conveyorized, automatic machines as they did their work. China is known for its cheap labor force. Why so much automation to reduce the number of workers, I asked? The answer was simple: labor rates have increased tremendously in the last ten years. Wait a minute. If Chinese labor costs have gotten so expensive that we need to build a robotized, automated assembly line, why would we build it in China, exactly halfway around the world, instead of in our back yard? On the long flight home, I convinced myself that we could build our own factory, right here in Boulder, to make our tiny robots. I convinced myself that on a certain level, it’s pretty much insane to build products all of the way around the world just because the people there are poorer. I convinced myself that it would be fun, interesting, and a generally good thing to do for the world. I convinced myself to make a really unlikely decision. There’s an alternative, by the way, that lots of companies are taking: chasing cheap labor. Huge factories are popping up in Vietnam and Thailand, Mexico, even Burma. But honestly, this seems short sighted. I don’t think it’s ethical or sustainable. I just don’t think it’s cool. I called a couple of people that weekend to talk through the idea of manufacturing robots at scale here in the old USA. That’s where the word, “asinine” came from. A friend who works for a clothing manufacturer suggested that I must have eaten some really bad Chinese food on my trip. I was undeterred. I’m in the unique and interesting position of being able to make big decisions for our company without giving a shit what anybody else thinks. I sort of like having that card up my sleeve but I felt like this was the wrong decision on which to play it. If we ended up building a factory in the USA just because I said so, it could quickly turn into Eric’s Folly. It seemed dumb to make a decision like this without full consensus; otherwise, when things invariably started to go wrong, it’d be because of my stupid idea. So we formed a Manufacturing Task Force to explore the ramifications of building our robots here and not in China. There are four of us on the task force: me, Tascha (Director of Finance), Scott (Director of Supply Chain), and Matthew (Head of Production). We bought blue terrycloth headbands that we wear both to remind ourselves of the importance of our task force and to ensure that we look like huge dorks. We’re a good mix of preconceived notions. I started out staunchly in favor of USA manufacturing, and Tascha thought it would be impossible. Scott has ten years of experience with Chinese manufacturing, and Matt, who oversees our mini-factory, was eager to expand it but admittedly a bit glassy-eyed when looking at the number of tiny robots we expect to manufacture over the next year or two. We decided to take no more than eight weeks to build out a financial model of the alternatives and see what the bottom line looked like. The finished model is pretty complicated. I’ll break it down a bit in a future article. We took what we know about making all of the Cubelets we’ve made to date, and compared it with quotes from contract manufacturers, and projected these options out over time. That part was pretty easy; the interesting part was assigning costs to some of the “soft” metrics involved. Is it worth anything to be able to say, “Made in the USA” on our box? How much more innovative can we be with production and engineering co-located? What’s that worth? How much is the annoyance of flying for 14 hours worth over just walking downstairs? There is no straight conversion from annoyance to yuan, but we tried to put numbers on everything we could think of. I’ll digress for a moment to let you know that I feel like data isn’t always the best answer to a question. This kills our engineers. But seriously; fundamentally we’re all ruled by predictable physics, but we can’t plan for the future at an atomic level. The “data” that we use for something like a business decision is so high-level and abstract that it can be misleading and certainly incomplete. I’m not saying that there’s magic involved, just science and causality that we don’t understand yet. Thanks Dave, for the pointer to David Brooks’ recent article that does a much better job of explaining this than I am doing. Even a huge amount of data can’t predict the future. But we can talk about trends and I can say confidently that I think the difference in cost between manufacturing in the USA or China is decreasing and will continue to do so. Ten years ago, it was a no-brainer; manufacturing was outsourced. But recently we’ve seen big companies like GE and Apple bring some manufacturing to the USA. I first went to China in 2008 and witnessing the change in cost and economic climate since then has made me confident that making stuff in the USA, while still not as cheap as in China, is getting much more attractive. The last two paragraphs might sound a little bit like justification. They are. At the end of the eight weeks, we didn’t end up with a clear answer that making tiny robots in the USA would be cheaper. Nor did we end up with the opposite answer. We ended up in between. All of our soft costs were estimated as a range. If we calculated based on one side of the range, we’d end up with “do it ourselves.” If we calculated toward the other limit, we’d end up with “have a factory in China do it.” We did end up with the conclusion that we could, in fact, do it one way or the other and have a high likelihood of success. So we decided to build a big factory and make our robots here in Boulder. Woo! The task force is into it, I’m into it, the Board of Directors is into it, and our whole team is into it. When the task force announced its decision to everyone at scrum, a couple of assembly elves even piped up spontaneously to talk about how amped they were to work for a company that made stuff they were proud of. Thanks Kristen and Joe! We’ll keep our injection molding and metal stamping in China for now, but examine bringing them over here after we’ve figured out the other parts of our manufacturing-at-scale process. We’ve already been working on getting a stencil printer, board washer, and AOI for our SMT electronics line. We’ve hired four new employees in the last couple of weeks. And we’re looking around at big (15k square foot) manufacturing spaces in town. It’s going to be an interesting couple of years. I’ll try to write often about how it’s going robotizing and automating our little factory. Know anyone who wants a job building tiny robots?

As we’ve grown from 2 people to 20 and from hundreds of Cubelets to hundreds of thousands of Cubelets, we’ve hit a few stumbling blocks.  There was that time that we received a huge shipment of plastic parts that were all the wrong size, for example.  In the general scheme of things, we’ve been pretty lucky, and we’ve learned quite a bit from (and enjoyed) the Sparkfun blog, where they document their travails trying to make stuff.  There was that time that they got all of the counterfeit Atmel micros.  Or the cease and desist letter.  Or the subpoena.  Now we’ve got a new problem, though, and it’ll be interesting to see how it plays out.  US Customs seized our stuff. We’ve been waiting on a shipment of voltage regulators for a while.  They’re tiny little parts; miniscule electronic chips that stabilize the varying battery voltage in Cubelets before power gets to each microcontroller inside.  Every single Cubelet contains one, and it’s the FAN2500S25X, made by Fairchild Semiconductor.  For the last couple of years, we’ve been buying them from Zhengke Electronics in China because they offer better prices and much shorter lead times than any of the electronics distributors in the USA.  Since the parts are so tiny, the shipping doesn’t get too expensive.  Whenever we work with a small, offshore vendor, we scale up gradually.  We place small orders, make sure they’re good, and then increase quantity as we trust each other more.  There’s always some risk of wiring a whole pile of money to China and getting a box of rocks in return.  We’ve been working with Zhengke for three years now, though, so we’re pretty comfortable placing large orders with them. When this shipment of voltage regulators seemed to be taking a long time to arrive, we tracked the shipment on UPS.com and were fairly well surprised to find a status reading: GOVERNMENT AGENCY HOLD.  Seized.  Seize them!  It sounds like a Scooby Doo episode. We called UPS and they told us to call US Customs and Border Patrol, where we reached Officer Ayala.  He explained that one of his agents inspected our package and suspected that it might contain counterfeit parts, so they seized it, took it to an undisclosed location, and were going to investigate.  He explained that Fairchild has been seeing a lot of counterfeit parts recently, and that Customs was on the lookout to prevent such things.  He said that Customs had taken photos of the packaging and of the individual parts, sent them to Fairchild, and would wait for their response.  He suggested that we should call again in a week or so. I’m really curious to see how this plays out.  Since we’ve had such a great relationship with Zhengke, my intuition is that the parts are genuine, not counterfeit.  But we’re buying them through an “unauthorized channel”: a little Chinese vendor who buys leftover parts from big contract manufacturers and then resells them.  This doesn’t bother me in the slightest, and it’s perfectly legal, but Fairchild probably doesn’t like it; I imagine that they want to control distribution through a set of specific vendors.  That way they can control the price and who gets what.  In my head I’ve already made up this story of international intrigue in which the parts are genuine, but Fairchild has Customs working as their little personal goon squad to stop any shipments from an unauthorized reseller, but I have a vivid imagination.  Stay tuned!