LumenLabs is dead. They died over a year ago, and their machines are not for sale. Their last tweet was in December of 2011, so they’ve probably been dead since January of 2012. Their website is now owned by a domain parker.
I’ve been watching LumenLab for a long time. The robot they had designed was ingenious. It’s jut plain beautiful, and a nice simple, elegant design. I would love to own one. I nearly bought one, just for the beauty of it a few years ago. Why didn’t I? Because I read their forums, and saw all the people complaining about delivery dates. Now that the company is dead, I feel that it’s important to do a postmortem to better understand why they failed ( so I don’t make those mistakes ).
Here’s my personal opinion — why I believe their company either failed ( or at least, had public relations problems. I’m not 100% sure what’s up with them — they’re just silent right now ). There seem to be two problems that killed them as a company — the first a design problem in their robot, and the second — their build philosophy.
Now, I lust after their robot — so what’s the design problem? The problem is machining time. Each of those monolithic blocks must be milled. They have geometries and are made of material that doesn’t injection mold very well, can’t be laser cut, can’t be water-jet cut, and require plastics that can’t be printed. As a result, the only way to make a large number of them is with a bot mill. Bot mills don’t scale well — they’re linear. Everyone who has gone down the bot mill path has had nightmares. Bot mills are labor intensive operations, and usually, the number of people involved in a project like the micRo is too small to effectively run a bot mill. They also didn’t go down the bot-mill path. They tried to scale by building a large bot that made the items in batches. This bot was named, RoGR. See the robot analysis section for more info on how they machined each micRo.
So, from the get-go, they had a serious problem. They had a design whose production process can’t scale. Now, this type of problem can be managed. Boeing is a great example of a company who custom builds each plane, but can scale their design. How? Cost! Boeing sells planes for millions!
Which brings me to the second problem — they ran the business like a couple of guys and a machine shop. I have seen this model so often in the 3D printer and CNC mill community, that it’s not even funny anymore. It’s almost always 2 guys, 1 woman, and a machine shop. It’s a recipe for early success — and for slow, horrid failure. Every company I’ve seen that starts off this way ( sometimes it’s a single guy and a lathe — but the idea is the same. Small number of people, and 1-2 pieces of really nice equipment. Almost always, one of the men is an engineer or a machinist. ) They design their robots around what they have — some integral component can be manufactured easily by their machines. Let’s say they have a really nice CNC mill ( which LumenLab did have — at least 2, and a large-form CNC lathe. I watched their videos on youtube, blip.tv, etc… — even the early ones back from 2005/6. Those videos are now all private, and you can’t see them anymore. I’ve seen their shop on video. It would put my hackerspace to shame. ) — they’ll make an awesome design around parts that can be made in their machine shop. What happens when any machine reaches capacity? Backorders. Lots and lots of backorders. Those backorders become a major business problem. Those backorders become one of three things:
- Pissed off customers — people have paid for the thing, and they don’t get the thing in a manner they consider timely. Without good communication, this becomes a killer.
- Costs. Your wonderful profit margins are gone — you need to invest in more machines and the labor to run them, or you need to outsource at a higher cost. Suddenly, your 200-300 profit per machine is now a big loss. You never really had profits to begin with — you co-mingled labor with profit, and forgot to account for costs like depreciation. You were “eating seed corn” as it were, and didn’t realize it until those backorders showed up, and you had to pay the full costs of your parts. This threw your marginal costs out the window, and you could never afford to invest enough to get this problem under control.( Profit margins on custom-machined parts is very low. When you pay $70 for a screw at a machine shop, perhaps 15% of that is actually profit to the machine shop. The rest of that is costs. )
- Risks go up. Any mistake you make now — anything that takes your time — is going go kill the business. And the business is now really just you. One of you gets sick, and the entire business ends. You no longer have the structure in place to handle all the tasks needed to dig out of the backorder hole. The machine that is now at capacity breaks — guess what — you’re done.
Backorders of this type have 4 possible solutions:
- Raise prices — have your supply meet the capacity of your shop, and only satisfy the demand you can satisfy comfortably at your current scale.
- Outsource — have other companies help you make the part. This may involve redesign, to accommodate what capital the other companies have.
- Scale — buy more machines and hire people. This is hard — harder than all the other solutions.
- Fail — you’re in over your head — tell people and give up.
Almost everyone in the “two guys, a woman, and a machine shop” scenario goes for solution 3 — it’s the only choice that works when the problem shows up. It’s also wrong. You never had priced right in the beginning, and now you have to buy new equipment and take on the expense of labor. Likely, your margins aren’t high enough to do this. For some unknown reason — no one ever picks 1 or 2 — even though those solutions are highly likely to succeed. Other types of businesses will pick these. The “this is my side business” guy will pick any or all of the other 3. Prntrbot did this. ShapeOko did this. The both ran into back-order problems, and realized the only solution was to raise price, outsource, and even exit( I don’t believe that the creators of those projects are directly involved anymore. At most, they’ve become figure-heads. )
The micRo ( especially the V3 variant ) was horribly underpriced. The V3 was literally bigger and better than any other mill in its class. To date, no one has even come close. It should not have cost $1,000 as a kit. It should have cost, $3,000 as a kit, and $5,000 as a built machine. ShopBot level prices. Why? Because that’s at least the minimum point the business could have succeeded at. At lower prices, the business doesn’t have the ability to hire people or scale as needed. It doesn’t have the ability to outsource parts. Let’s take a quick look at ShopBot — the ShopBot Buddy Desktop is 24″x24″x5″ — exactly the same size as the micRo v3. The ShopBot costs $5,800 with a porter-cable router at Education pricing! LumenLabs had an ER11 class spindle — they sold it for $100, while ShopBot sells theirs for $1,000! LumenLab was selling a 24″ desktop class mill for the same price as Taig sold their 3″ mill! The micRo v1 ( the smallest mill they had ) sold as a kit for slightly more than a Taig micro-lathe kit ( a much simpler tool ). LumenLab was probably adding $100-400 to the BOM of their product, and calling it good. The V3 probably felt like a rip off to them ( I’ve spent time reverse engineering their product from the Internet. I believe I’ve figured out their costs.
The micRo v3 had a BOM between 300 and 700 dollars in raw materials, excluding machining costs. The v1 was almost the same as the V3, but about $100 less in raw material. So, $200 to $400, just in parts. From what I can see, they were using Vista metals ATP-5 tool plate as their base, 1.5″ thick, 2″ wide HDPE blocks as their liners, which they bought by the foot and cut with a CNC saw, then milled. ATP-5 is very expensive — around $200 for a 24″x24″x1/4″ plate. HDPE blocks are cheap — $9/foot. Their ways are probably $25/way. So, $300 just in plate and ways. Another $20 for all the HDPE raw bars. Let’s say 3 hours of machining time per HDPE block — which they didn’t count, since they owned the machine( typically, machine time is around $70-$100 an hour. ) So, $70/hour x 3 hours for each block x 8.5 blocks — $1,785 in machine time they didn’t charge for. Another 1 hour of machine time for the plate. Don’t forget lathe time for the leadscrews. so let’s say 2K in machine time, not charged. Another $150 in misc parts ( screws, motors, washers, thrust bearings, couplers, collars, etc… ) and $100 in electronics. The V3, without the machine time, costs them about $650( With good sourcing, they may have been able to cut that price by 200 at most ). They sold it for 999 — 350-550 profit( so they think ). But once at capacity — they couldn’t pay for the machine time. That’s almost 2K/robot they hadn’t accounted for, since they viewed machine time as free, and actually had negative real profits for every machine sold! When they went to get those HDPE blocks milled elsewhere — it would cost them 2k, but they sold the machine for 1K. They would take a loss of 1K on each machine to meet their ship dates. So, they tried to get out of the backorder problem a few ways. They thought about building a bot-farm to manufacture the blocks, but didn’t because it would cost too much and they couldn’t run it. They tried outsourcing the part production — but couldn’t because it cost too much per block. They eventually found a somewhat cheap supplier — with poor quality control and a high defect rate. Those blocks just can’t be machined by the lowest bidder.
Watching how LumenLab failed, ( and to a lessor extent, other projects surviving by the skin of their teeth ), has shown me some risks in small-volume robot sales that I had been ignoring. I realized that at very low volume, robots have to be very expensive, or the company fails. You end up dead one of two ways: (1) Dead due to supply costs, or (2) dead due to inability to sell. In case (2), you die because you keep slipping your delivery date, doing R&D to try and hit your BOM in volume, and by the time you ship, someone else eats your lunch.
It’s a big problem in the maker community. I’ve seen so many wonderful robots at my hackerspace, whose designs and companies will go nowhere. The problem is price. Once your design is open source, you can’t charge enough to scale. If you are popular enough a product, some yahoo will come in and rip off your design, undercutting your price. You’ll always be stuck at the marginal point.
As an aside — I reverse engineered a single part of theirs, just to see how much it would cost to make. It was one of the simplest parts — a motor stanchion. I got it quoted — the lowest cost was $400. They had 9 plastic parts that all needed to be machined. Each block has 4 setups, all of which must be perfect. You’d have to design jigs and hold-downs to make one. So, machining costs for them, if they priced it correctly, would have been $3,600 per robot — just in machining time for the plastic blocks! No-one will ever copy and sell LumenLab’s design — it would be a stupid move to make. You’re spending at least 4K in machining time per robot and you’re selling the robot for $1,000. You can see here exactly what happened to their business. They could have designed their parts to be made in a single setup — at huge amounts of waste per part. If they designed their machine to waste 33% of the plastic, have no counter-sinks, tabbed hold-down, etc — they could have dropped the machining costs to $50/block at volume. I’ve seen versions of their designs that were easier to machine — and those changes were ugly. This machine’s beauty comes from the block design — a design that makes those blocks cost an arm and a leg…
As another aside — It is such a beautiful robot, that I’d love to have one. If any of you readers out there have one for sale, please let me know. I’d be interested in buying it. Or, if you have the design files for it, I’d love to have them. You can contact me at ipeerbhai [ symbol for at here ] aol [ symbol for dot here ] com.
I wanted to talk a bit about the micRo and how they made it. First of all, Grayson and Robyn ( I think those are their names ) had a machine shop. They started with some sort of mill and a lathe. Both their mill and their lathe were huge beasts, each man-sized. I’ve been watching their stuff from blip.tv, vimeo, youtube, and reading their forums for a long time, so I have a good sense of what they had, how long it took to make the first micRo, etc… Before they could build the micRo, they had to first build a CNC mill that could make the micRo. Now, they already had a large mill. But they felt the tolerances of it weren’t good enough. They spent 2 years designing a mill that can make perfect holes in HDPE blocks. They had to settle on HDPE, since any other plastic couldn’t forgive the error they still had in the part. The called the mill they designed to make the micRo RoGR. RoGR itself is an interesting robot — it was made from powder coated metal attached together by bolts, with what looks like a 1″ linear bearing system and 1/4″ thick steel plates. RoGR must have weighed close to 1/2 a ton or maybe even a ton, and getting RoGR working was their first real challenge.
RoGR is a large, and pretty amazing robot. It easily had a work area of 3′ by 3′ by 8″, with tolerances around .001″ ( or better! ). I don’t even know if a ShopBot can achieve that over the distances involved. So, it took 2 years of building and tuning a large CNC mill before they had the required machine needed to make their HDPE monolithic blocks or mill the base plate of the micRo. I’m not 100% sure it was needed — they might have been able to make do with a much smaller machine, but I think their idea was to make blocks in large batches on RoGR. Those blocks are only 2″ X 5″ x 2″ in size, roughly.
Going further back in time, to when the micRo was conceived — they had a lot of design prototype for the micRo. I don’t know what their original prototype was — the first one I saw on youtube was using Vista Metals ATP-5 with SBR-12 rails and bearings. The SBR rail system is supported, and they had milled HDPE blocks to screw into the roller bearing. This was an existing CNC bearing system, very cheap and easy to get in China, even back then — and they didn’t go with it. I think I know why, too. SCS and SBR type rails/bearings clog easily. They have wonderful linear characteristics — but any dirt, and they fail. They also have poor radial flex characteristics — they’re designed to compensate for shaft misalignment, but in the process, sacrifice linear precision in a single bearing design. Failed bearings means a tear-down and rebuild — not too hard to do, but annoying as sin. I think they wanted to avoid all tear-down/rebuild operations. I see from early videos that they tried Delrin and a few other plastics besides HDPE, and didn’t pick them, either. Grayson explains in a video that the problem is how well it forms a bearing surface when machined. HDPE holds up the best to machining while remaining bearing grade. They also are running the HDPE blocks for 2″ in the longest direction. This means that the holes drilled into the HDPE block had to be perfectly round, and bearing smooth, through the entire 2″ throw of the bearing. I haven’t yet found a machine that can do this without a lot of calibration time.
Then, after they had built RoGR, they got into production. The micRo itself was made of very few parts — 7 monolithic blocks, 2 plates, a motor spacer, assorted screws, nuts, rods, and pins. The rest of the story I’ve already told. Those monolithic blocks had to be machined, and the few machine shops with machines precise enough to do the work would charge more than the cost of the final robot. So, back-orders happened( LumenLab was always back-ordered to many months, almost a year ), then Grayson got sick. That was the end of it — without enough money to hire a crew, and with angry customers asking for their money back — well, it didn’t go well from there.
On a closing note — I always thought the micRo was beautiful — if the company had a better reputation for delivery, I would have bought one! Even today, I keep thinking of building one — just for the sheer beauty of it.