Don’t Fear the Robots
Every time one of these Boston Dynamics videos hit, they are met with a mix of amazement and predictions of doom at the hands of our new robot overlords.
It is impressive.
It is also very well choreographed and stage.
Impressive as the tech here is, it is important to remember that the Boston Dynamics videos are trailers for a movie that isn’t yet made. Look hard enough, you can find the blooper reels of the robots that are so impressive running, jumping, and doing other things failing as they work out the kinks.
Because the videos are not just videos, they are advertisements. For funding and investments. An actual warehouse environment is much different than the staged demo shown here. The AI will get there, someday, but it is only as good as it’s programing, and anyone who has spent time in the logistics and transportation industry will tell you it only takes one misread SKU code, or one mis-bayed item, or one wrong label, to make a whole lot of mess. The problems in becoming more efficient in logistics isn’t just the streamlining of processes, but the ability to problem solve and adapt on the fly in an very dynamic environment. That decision making ability is make-or-break in most situations in an environment that is long stretches of mind-numbingly repetitive actions punctuated by moments of intense and desperate revisions to the process. Automation is fantastic at the first, but the latter is a great challenge, and also the difference in success or failure. Robots that seek to fully replace humans have to do that, not just the manual labor aspect, and the most advanced robot is still only as good as it’s programing. As of yet, whatever “thinking” they do still has to be programmed in, and industries like logistics which are organized chaos on the best of days have far more variables than can be planned in advanced for.
While the day may come when robots fully replace human box kickers, the more realistic — not to mention profitable — use of automation right now is not in fully replacing the humans, but in enhancing and improving what they can do with robotic assistance. Variations on LEAN concepts have migrated from manufacturing and taking up space in everything from transportation to healthcare, and have fully taken hold in logistics. Smarter, better, and more efficient are the buzzwords for improvement. Such philosophy is naturally suited for introduction of automation to make processes more efficient. Compare the Boston Dynamics vid to this one of Amazon Robotics, which unlike the Handle Robot’s strictly controlled demo, is currently deployed and functional world wide right now.
More relevant than the hyperbole over the pending rise of Skynet is the issue of how much disruption disruptive AI tech is going to cause the job market. The spectrum of opinions runs the gambit from mildly disruptive to concerns whole industries will go human free. The truth is technology always interrupts jobs, and jobs constantly change. The refrigerator killed the ice delivery business, cars didn’t do the liveries any favors, and when was the last time you bought a wall phone?
It is also true whole industries can change suddenly. Manufacturing and mining have seen steep drops in the required manpower with inclusion of technology and automation. But examples abound on both sides of the evolving tech and economy. For every Youngstown or rust belt region gutted by these changes there are Pittsburgh and wide swaths of the sun belt that are thriving with new growth. Until they learn to do it themselves, the robots still have to be designed, manufactured, programmed, and maintained by humans, which creates it’s own new and disruptive industry.
AI will disrupt, and robotics will change the job market. But they also have to be commercially viable to do so. The robotics that best enhance and enable increases in human productivity will be the ones that take hold and change the world, like the Amazon Robotics currently are with little fanfare. When the Boston Dynamics machines make life better for many consumers the way Amazon has, then they will be truly revolutionary, not just potentially so.
But until then, the videos are really cool, trailers to a movie we really want to see. If they ever get around to making that movie as good as the tease hints it could be.
The thing about Boston Dynamics is that they are solving problems in robot movement that have been constant bugbears since we started toying around with robots. Using living things learn how to do amazing movements that robots struggle with. And since we learn such things often unconsciously, figuring how to to sense the things we automatically sense, and code for it, is a huge challenge.Report
And they somehow ended up with veloceraptors?Report
Veloceraptor is an extremely robust design.Report
…living things learn how to do amazing movements that robots struggle with. And since we learn such things often unconsciously…
Two of the things I learned back in my golfing days were: (1) the modern golf swing is one of the most unnatural human movements anywhere in sports, and (2) it took well over a hundred years for people who were thinking about it professionally to arrive at that motion.Report
Pitching a baseball overhand is extremely unnatural, which is why it took so long to catch on and causes so many injuries.Report
“An actual warehouse environment is much different than the staged demo shown here. ”
I dunno. If a human is just executing a standard procedure on standardized products, aren’t they a “robot” in all senses including the original? And all it takes after that is a bit more work on standardization and mechanical robots can handle it.Report
@densityduck
“If a human is just executing a standard procedure on standardized products…”
You left your comment while I was typing mine, but I address that below. A warehouse has no real standardized product. The average building size for my company is about 800,000 square feet and there might be 100,000 different unique SKUs within that facility. For example, I have a new customer coming online in May. their footprint is about 30,000 sq ft. They have told us that they will have nearly 1,000 unique SKUs to start out. And that’s actually a fairly small number.Report
” A warehouse has no real standardized product.”
Doesn’t is not can’t.
What if your customer with thousands of unique SKUs put them all in the same size box? A size which is cartoonishly huge for most items, but you can guarantee that it’s always the same size box to handle and always marked the same way?
“When I can pay a recent immigrant $12.50/hr plus good benefits…”
Congratulations, you’ve identified why robots will eat the future.Report
Corrugation (boxes) are really expensive. And a lot of my customers have very specific storage requirements. I deal primarily with aerospace and high-tech customers and there are all kinds of weird requirements there. Plus, knowing they pay for every square foot of space, they aren’t going to want to pay for a lot of extra space just to make it easier to automate.
The biggest perk for automation is error reduction. As someone said, barcodes are one of the best features.Report
The attitude taken was that Of Course there’d never be warehouse-picker robots because Of Course the whole situation was irreducibly complex because Of Course the inefficiency of standardization outweighed the gains of automation.
And, the thing is, this is kind of the inverse of Slippery-Slope reasoning (the Inverse Pyramid argument?) Because if that last thing stops being true then the whole argument falls apart. Like, I’m sure there were a lot of very good arguments for why you definitely couldn’t ship (thing) in a standardized container, but it’s hard these days to find a (thing) for which that’s true.Report
Surprise, surprise, I have a few opinions about this…
My company is one of the top logistics providers in the U.S. but with regards to technology we are somewhere in the middle of the industry. I have been to facilities run by other companies that are proud of themselves because they manage their inventory with an Access database instead of spreadsheets. I have also seen videos of facilities that are piloting AR to improve their efficiency and safety. Google Glass has essentially become an enterprise project and it’s the drum I probably beat the loudest when we talk new tech at work.
It’s absolutely possible to heavily automate a warehouse environment but it’s cost prohibitive in most cases.It’s not like a manufacturing facility where a robot is cranking out the exact same thing all day. In a warehouse environment an employee might pick dozens of different SKUs in one shift, all with different sizes, shapes, etc. Humans are still the best resource for that because they can make hundreds of micro decisions per day to adjust for changing factors. When I can pay a recent immigrant $12.50/hr plus good benefits and they can do this work with a high degree of accuracy, it’s still a much better financial decision to stick with people.
The primary thing we focus on regarding technology is constantly improving our Warehouse Management Systems with better functionality for our operators and also looking at ways to improve flow within the facility. Transportation, Motion and Over-Processing are the three primary types of wastes that we are constantly trying to eliminate through Lean methodologies.
Another word on KIVA…If you notice in the video Andrew posted, there is a lot of wasted space above those KIVA modules. My company charges our customers for every square foot of storage, so we want to maximize both horizontal and vertical storage. We only have a couple of automated facilities but the most recent one uses a multi-shuttle system like this:
https://www.youtube.com/watch?v=DQq_S4RoSKMReport
One of the best things about this place is the access it provides to people who know things because of what they do and where they live. Thanks.
Now, I have a question. When you look at the video Andrew posted about the robots at Amazon, do you think, “Meh, that’s an awful lot of money for a small improvement” or do you think, “Man, I wish we had those”.
Because for me, just watching the demo video, it’s hard to evaluate efficiency gains.Report
I think it’s a lot of money for a small improvement. I do like automation, but usually it’s stuff we can use in a more general sense. For example, we have a lot of automated shrink wrappers where you can put a whole pallet of goods on a pedestal and walk away while the machine wraps it. Saves a lot of time when we send out hundreds of pallets per day. Also, the automation like the video I posted can work with a variety of parts. Those totes can hold pretty much anything that will fit in them.Report
The veloceraptor robot has a lot of mass in it’s counterweight and spends considerable time adjusting it.
Force, mass and acceleration still apply in automation, which the energy involved will eventually be pulled from the electrical grid. The shuttle system appears to have the least amount of mass acceleration per unit of product, so it would likely be more efficient for the boxed styled handling we are seeing here.
Typically a robotic arm fastened to the concrete floor is utilized to pick and place the conveyor items on a automatic palletizer/wrapper, the floor is the default counter balance, and it is not moved in the process.
Moving counter balance mass ends up in more wear on parts, and eventually to more electricity usage.Report
You make a good point, and as others have pointed out the largest problem with the Boston Dynamics robots so far is lack of any energy source capable of sustaining the functions you see them performing for any length of time without a teether of some kind. Especially in a warehousing environment which is 27/7 tempo in most cases, anything under multiple hours of uninterrupted use is more problem than it is worth.Report
Sometimes it’s a small gain, but on an operation performed a billion times a year it adds up. IIRC, every delivery truck that leaves a UPS facility now has an optimal (or near-optimal) route programmed in for the driver to follow. Each run takes into account the addresses to which all the packages are going, where the driver will stop to pick up outgoing parcels, local road conditions, which stops have TOD requirements,etc. Basically the software solves a traveling salesman problem with additional constraints for each truck. We’ve only recently gotten to where there’s enough cheap computing power available to do that. They don’t save a lot of fuel or time on each trip, but over the course of a year it saves on the order of 10 million gallons of fuel and reduces other costs.Report
I know a little bit about the route optimization that UPS does 😉 You’re right that small efficiencies add up. As I said in another comment, the problem in a logistics warehouse is really about finding technology that can handle a multitude of different sized products.
We routinely run into issues where customers do not accurately convey the size and weight of their product to us before we implement their account or when they add new SKUs. This creates a ton of capacity issues related to storage. I can’t imagine what that would mean for automation.Report
I have used it, it works pretty well once you have all of the specific constraints dialed in. But, it does come up against the human factor really hard. Drivers hate being told when and where to be. The have someone whom they want to flirt with, a favorite place for lunch, want to take a side trip to put something in a storage locker, they want overtime, and a million other little things. But, it does give a lot of cause to bring to the union.
When I was with Iron Mountain, we based everything on a basic bankers box (we called a cube) and if a customer mistakenly referred a 3.6 as a 1.2 it would through the whole thing off. Either you picked it up an forgot about everything else that day jeopardizing all of those customers, or you jeoped the customer you needed to refuse.Report
Package size hurts my side of the business mostly in the storage sense. When the product won’t fit in the storage location, our Putaway folks get really creative (because they have a productivity goal to meet). Then we get parts commingling in the racking, which causes mispicks, etc. It’s a waterfall effect that often starts with the customer giving us bad info. We have tried to push back at times and insist we measure stuff as it comes in, but our operators are so good at keeping their heads down and pushing forward that they just bring it in and try to deal with it. They are often their own worst enemies.Report
At Neiman Marcus Direct we used a scanner that also sized the packages (LxWxH) as they entered the sorter, and the information was used so the RS/6000’s database software could better calculate the cube for the trucks.
One of the other fun things about that project was that all our test boxes were for famous Hollywood stars like George Clooney, with their real shipping addresses.
One time when I was there, the programmers I was working with had just figured out why all the packages toward the end of the prior month had ended up shipping via FedEx Next Day Air.
Management was blaming their software, but they are really good IT guys who love a mystery. It turned out that the vice president (of whatever) was *that* close to hitting the monthly target where his massive bonuses kicked in, but the only way he could put it over the top was by upgrading those last shipments at Neiman’s expense. Then he tried to cover his tracks in the computer. He was immediately and unceremoniously fired.
Report
We occasionally use a cubiscan device to capture accurate dims for parts for new SKUs but customers rarely want to pay for the extra cost and operations doesn’t want to lose the productivity time it takes to perform the task.Report
I’ve programmed a whole lot of shipping and distribution centers. On occasion it turns out that some parts of the automatic systems are actually less efficient than doing things by hand. That’s usually showed up on automated receiving systems where most of the incoming pallets should’ve been kept as a pallet instead of being broken up in individual cartons which are thrown onto the conveyor system that ends up sorting the cartons into pretty much the same pallet they started from.
There are other cases where a carton just needs to go from A to B really quickly, but the automated route would have it spend an hour or so riding around the building through multiple sortation systems, so they’ll instead use humans to handle it as a hot pick.
But these situations also provide information on where automation works well and where it doesn’t.
One of the problems with AGV’s (automatic guided vehicles) is that they don’t mix well with people. They either have to move slowly and beep a whole lot (which is irritating), or you have to use a lot of safety barriers to make sure nobody gets run over. That means floor space has to get dedicated to them.
Report
Heh. Over in the autonomous-vehicles threads it’s just taken as a given that dedicating space to autonomous vehicles is a Nightmare Capitalist New Gilded Age Bullshit Anti-Human Car-Fetish Garbage American Society That Sucks thing.Report
@georgeturner
We do the hot picks for new product launches (we have one big cellphone customer). When the new iPhones launch we will just park pallets of them on the floor because we know we’re just going to have thousands of single quantity orders.Report
The interesting thing is, looking at those Amazon robots, is that it would be pretty simple to build a lower level for them to route _under_ where people walk.
Obviously, if they’re carrying something, they can’t do that, but it would let them move around empty very easily.
And if you were really clever, you could make the human walking paths be drawbridges that flip up and physically stop people from walking where a robot is going to be moving something through. Or, alternately, someone is on top of them and they can’t flip up, so the robot stops.
I say this like Amazon hasn’t already thought of this and isn’t already doing it.
Having eight-inch tall robots is actually a pretty useful design.Report
The cool thing about the KIVA system is that the shelves have long legs and the robots just roll under them, so it creates this internal ‘highway’ they can travel through. A couple of ideas we had for using them was to create overpasses between our buildings so we could flex the robots back and forth between buildings depending on which accounts had the highest volume for the day. We also discussed building a mezzanine with an elevator so the robots could travel between floors. Alas, once Amazon bought the tech we stopped white boarding those ideas and went to a multi-shuttle system.Report
Robots are, I think, more of a distraction. The real threat to human labor is software and machine learning that replaces the human skills of computing, logic analysis and pattern recognition.
Those skills are the core of our highest professions like law, engineering and medicine. Its not like one day you will walk in and see a robot doctor. The threat is deskilling, where the machine does the difficult, highly valued tasks, and the human just monitors and tends to it.Report
A couple of things. My first logistics job was in a DC for a big box store. There were areas in that DC (half a mile long) that had lots of open space and other areas that were extremely close quarters. Not much room to get a robot in like the Boston D video above. Also, bar codes are what drives this whole world. If for some reason it can’t be read, the system shuts down to a certain degree. The more this happens, the greater the shutdown.
Robots work wonders in a standardized environment with constant throughputs. When those change, they don’t work wonders anymore. This is neither good nor bad, just a fact. Humans, on the other hand, are infinitely flexible. If I needed an extra 50 people, I just call a staffing agency and they are there within the week, at $12.50/hr. 50 robots? I would need a year of lead time and about a million dollars each. For something that can only do one job.Report
I didn’t understand what the robots in the Amazon clip were actually doing, but got the impression that for practical purposes, the pickers/packers could stay in one place and the “shelf” with the appropriate product came to them. Is that actually what I was seeing?Report
@michaelcain
You are correct. I should also note that the Kiva system does optimize horizontal storage (no aisles necessary for pickers) but when we looked at the system we found it was still a net loss in billable storage space. Plus the shelving units are relatively small and would not accomodate the product for many of our customers.Report
How many jobs exist today that did not exist when you were born?
How many of those jobs didn’t exist when your granddad was born?
Even taking into account the funniest of the blooper videos and even when I *KNOW* that the perfect take for the robot working perfectly took 35 takes, 20 years ago, we were still playing the N64 and the PS1 and the GameBoy Color.
20 years later, the PS4 is slowing down and we’ve started talking about the PS5.
Those robots above are the Battle Arena Toshindens.
We’re going to have Final Fantasy VII, soon enough.
(We’re not going to get to the “changes everything” until management is deemed equally replaceable, though. When it’s not just the simple repetitive tasks that can be robot-sourced and/or code-sourced but the ones requiring nuance and judgment. When Capitalism no longer requires people to perpetuate the creation of wealth, *THEN* we need to start considering freaking out.)Report
Meh, socialism will continue resetting capitalism to the Iron Age, so we’re good for awhile.Report