Boston Dynamics unveils Stretch, a robot designed to move boxes in warehouses
theverge.com
theverge.com
I've done some work in Mech Engg. designing grippers for smaller objects, and gripping objects is often the hardest part. From an operations research perspective, a human will figure out the best way to carry a box that is almost falling apart almost intuitively.
I can see this being extremely useful in, say, a large warehouse where it is isolated to finding and moving boxes to loading bays, particularly through an API.
Amazon is already doing this, and this video is highly relevant here — https://www.youtube.com/watch?v=4sEVX4mPuto
https://en.wikipedia.org/wiki/Automated_storage_and_retrieva...
For warehouses willing to make capital investments automating the process of getting bins to pickers is a pretty well understood problem. Getting boxes out of trucks like Boston Dynamics is doing here and getting items out of bins and into boxes like we[1] are doing are the frontiers of warehouse automation right now.
Flipkart (India's Amazon, acquired by Walmart since then) deployed similar things in their warehouse: https://inc42.com/buzz/flipkart-deploys-100-automated-robots...
Yet another warehouse robotics startup from India: https://www.greyorange.com/
Stay strong factory workers!
But at this point they aren't the only company making that style of inventory/robot system now. There are a few clones out there in use.
Does anyone know the cost?
Besides, in my experience (as a professional in this exact field of warehouse / supply chain / fulfillment operations), the biggest gains are usually process-related, and adding robots doesn't really help. The average warehouse tends to have ridiculously inefficient processes, driven by clunky and cumbersome software. Things like aisle zoning / pick path planning (i.e. minimizing the distances pickers have to travel for average pick orders, and minimizing contention between pickers and replen), data entry automation (e.g. barcode scanners and computerized scales instead of manual data entry), precise and measured use of conveyors, etc. help maximize the return on every labor dollar spent. These things can all be done incrementally to good effect.
That is, to your point: throwing robots at an inefficient warehouse just gets you an inefficient warehouse full of robots. And yes, designing the warehouse from the ground up to support automation as efficiently as possible is a good idea, but focusing on human efficiency will get you 90% of the way there - and then you can spend the other 90% of that effort on whiz-bang software and droids ;)
I did what I could with Excel macros to automate as much as I could, but I'm not a trained coder and I definitely didn't have security access to implement the tools that could have saved us huge headaches. I wondered quite a bit who the hell up top got paid to set up this process.
Which goes to show that the warehouses themselves are only part of the battle. Just as critical is the whole org holistically taking downstream impacts into account. It's why I make it a point to be on the warehouse floor, doing the ground floor work firsthand, talking to warehouse workers, understanding and observing for myself where the bottlenecks are. That's the only way to really know you're actually helping.
I got annoyed with importing data for taxes in one instance recently and actually was able to recover all the data from a printable PDF form to then turn into a standard format.
There are so many parts of Excel that seem designed by different groups at different times, with varying levels of stability and sense-making, and so I think it takes a lot of experience just to know what exists and when to access it.
My approach these days is to do anything I can in Power Query and anything I can't in VBA. And those can connect to virtually anything. I avoid other stuff such as Power BI, the data model, DAX, formulas on worksheets, etc. due to painful experience.
This is similar to a situation we often deal in electronics repair business. Some people are very adamant about being careful when reworking around water damaged components because those solder joints/balls/legs are already weakened, when in fact the proper procedure is to rip out and replace everything compromised. Whats the point of turning device back to working electrical order while leaving it mechanically deteriorated - it will come back broken the first time used bumps into something, or when the eaten away solder ball finally gives up after few more heat cycles.
Any time someone would say "be careful or it falls apart" its time to remove/replace it, not pass along hoping to make it someone elses problem.
I'd imagine the goal ("purpose") is to meet a sweet spot between money spent on packaging materials and cost of returns from broken deliveries caused by the packaging.
Sometimes less money spent on packaging also aligns with corporate environmental goals. If you're into loyal customers you probably want to spend a little more than you "have to" one packaging.
UPS sometimes breaks boxes they deliver. They just re-tape it and finish delivery. Sometimes everything is there, sometimes something is lost and you need to file a claim.
This robot at least needs to recognize broken box and hail human in such case.
"Package destroyed, contents discarded." (it was a piece of furniture with a glass component).
and
"Package delayed due to train derailment."
Can you imagine triaging a train derailment and not setting it to the highest priority?
Making that the general approach would probably solve the automation problem pretty cleanly. You could even have broken-package-packaging robots at each cargo-unloading step!
The problem here is that the perforations will easily tear under the weight of a full case so it's extremely common for it to give out if you try to pick it up by the top.
but who cares if the case makes it to the store in one piece because we only care about the packs inside.
Sometimes you're unloading or on the receiving end of disintegrating boxes, and you need to unload the container the freight came in. This is to receive in the damaged product to your warehouse before you can claim it.
Two examples I from bring on the receiving end: - ocean freight were water gets in through holes / damaged parts of the container that cause all the packaging material to breakdown - beverages being punctured or otherwise leaking causing all packaging material to be damaged
In an ideal state, you don't have damaged packaging because you can use clamps, slips, or other specially designed conveyance equipment.
In 2019 BD released their updated Handle robots which used swinging counterbalance tail to lift warehouse boxes. Looked very dangerous to be around them; this 'Stretch' version seems to have been tamed much.
APR2019 Handle: https://spectrum.ieee.org/automaton/robotics/industrial-robo...
This one-side suction system will obviously fail if the box itself cannot hold its contents without ripping or deforming.
I would rather see the robot get better than we start using bigger boxes with lighter loads just to make it easier for the robots. I already receive too-big boxes and too much paper packing stuffing.
These robots appear to have a gentler touch than what I'm guessing humans do, based on the health of the cardboard I receive from humans. They might make it feasible to create re-usable plastic boxes that have some kind of an anchor point.
I've got a friend that works in the field and have seen their machines in operation and how they work on their contracts. The use cases right now seem to be entirely human defined; i.e. it handles items that weren't modified to make it easier on the robot, and it handles items in a random orientation. Could be turned, could be upside down, could be balanced on a corner against another item.
If robots take over more of the pipeline, you can start guaranteeing that boxes are oriented correctly, which gives you a lot more practical options than "cover it in hooks, and we'll worry about how to stack it later". Space in containers is also a big concern, so anything that increases the size of a good is generally bad (last-mile delivery excepted).
And perhaps with more automation and logistics improvements, reusable last-mile containers can become common. We already receive our one of our grocery deliveries in foldable crates which get re-used (one set always sitting empty at home, and traded to the delivery driver for crates with new groceries).
There's a lot of room for improvement here.
To realize what people are going to do when they don't have to work, look at those in retirement: gardening, choir singing, lotto may be?
But a lot of people if guaranteed a minimal quality of living would just stop working and do things that they like (that don't necessarily make money) and that's the whole point of UBI.
Which means less labor supply, which means higher labor prices, which means greater incentive for people to work (since wages would no longer be suppressed by the "either I take this predatory offer or I starve to death" effect).
> Then the people still working will have higher taxes
Not necessarily. That depends on who and what you tax. A land value tax + a tax on income or wealth after a certain point (say, $1m/year) would readily pay for UBI without putting any tax burden on the working class. The other benefits of LVT (like incentivizing denser urbanization and penalizing land speculation) are nice cherries on top.
...which means higher prices of good and services, which means that the basic income you had previously settled on is no longer sufficient to survive on. How sure are we that there's a stable equilibrium?
Only if automation never happens. Clearly that ain't the case, per this very comment section and the topic thereof.
Indeed, automation's going to continue anyway, because as automation costs fall they'll eventually undercut labor costs - even those artifically suppressed through coercive means, as is currently the case in any (capitalist or otherwise) society lacking a socioeconomic safety net like UBI. So sooner or later something like UBI is necessary either way (that, or making people do pointless busy work, which is both paternalistic and grossly inefficient).
There are many things that could be automated right now (for example, in the garments industry) but are not because it's currently cheaper to do them in a third world country by poorly paid workers - but the moment those workers would have to be properly paid, the tasks would go away as then it would be cheaper to automate.
Modern American safety net is far from perfect, but it does a pretty good job making sure people don't die from hunger not freeze to death.
Chefs do “robotic jobs” repeatedly creating the same end product. Would you be comfortable if the person who prepared your Michelin course dinner was forced to urinate in a bottle while they did it? Why are we ok with treating human beings like this, simply because we do not romanticize the task of sorting and moving boxes.
https://www.youtube.com/watch?v=yYUuWWnfRsk
As far as I can see, boxes are picked up from one side, or the top. Some boxes would collapse due to this uneven distribution of force - a human worker would normally pick up a box from both sides, or the bottom.
OTOH 73kg (160lb) full beer kegs still get a fair bit of one-man manhandling, and to be a proper professional scaffolder you have to be able to lift a full-length scaffolding pole (just under 30kg, apparently?) from flat off the ground to straight up above your head, partly one-handed in one smooth movement, several times a day routinely.
Realistically, any company investing in one of these robots is going to make sure it's right for their workflow and that they're using boxes capable of being lifted this way.
Most boxes are designed to withstand inward pressures, not necessarily outward. And those specs are based on a lot more thorough of testing than "well we tried it a couple times and it didn't break, so I guess we're good"; I'd need to see one of these handling a lot of boxes without issue before I'd throw it into a warehouse.
A huge part of the gains come from completely eliminating humans, and getting rid of all the niceties that come with human labor. I strongly suspect that it will come down to the opex to maintain the robots than the capex to buy them. If opex is lower than hiring humans, companies will buy in. You can get tax breaks on the capex for the robots as you devalue them. Robots also shield your opex (right now), because nobody is clamoring that the robots should get paid more. Your labor costs for humans could double at any point; even $15/hr isn't keeping pace with inflation if you look back to the 50's or 60's.
> Lots of comments about the limitations of the suction grabber, but I wonder if anyone here has an educated guess as to how far we are from the robotic hand dexterity needed for near human-level manipulation of relatively simple objects like boxes.
We already have have the means to do human-level manipulation like that, it just doesn't use a hand. I have a buddy that works in the space and have seen their robots go. If you want your box up on it's side, the robot will do that for you.
I don't think we're likely to see a big push towards human hands. Hands are great generalist manipulators, and are good when you don't know what kind of stuff you'll have to handle. That doesn't really apply to robots, though. There are almost always more specialized designs that are cheaper to make, cheaper to repair, and better at handling the item in question.
Just thinking about boxes, for example, a forklift seems infinitely better at handling boxes with potentially weak bottoms than a human is. Our hands cannot support the full length of the box.
Looking at surgery, almost the entire point of human hands is to manipulate tools which were designed to be manipulated by human hands. Just attach the tool directly to an arm on a platform that can spin and angle itself like a wrist. No need for overly-complex fingers that might drop a scalpel on someone's face.
I don't see human hands being a useful design for a robot until we get general artificial intelligence. If I can ask the robot to go perform a novel task, then it makes sense to equip it in a way that it can handle any kind of generic object. Until then, the robot was designed for a purpose, and it makes sense to choose specialized manipulators.
I know this is of cause done because it’s an easy target for robotics, and proves a value stream. But you still have to wonder just how big of a societal change we could achieve if these efforts where target at empowering and improving the lives of the masses instead of removing the “annoyance” of having to pay minimum wage workers as part of the costs of huge insanely profitable corporations.
Stretch from Boston Dynamics - https://news.ycombinator.com/item?id=26619523 - March 2021 (62 comments)
"its mobile base means it can slot into spaces designed for humans"
Sounds like there would be the temptation to have it running near humans. It looks pretty heavy and powerful.
What would help at least me is to see this dogs actually used for something useful where they are the best solution, what I mean used in production not demos - because I can't imagine that many scenarios where are a good solution over a drone or a robot with wheels.
To be realistic. Spot (the dog) is not autonomous currently either. In fact it doesn't even use "AI", all of its movements are electrical and mechanical control theory on steroids ;)
The BD customer can integrate an AI to direct movements if desired but it's not what BD ships. BD instead offers a human operated remote control with camera system.
I might be missing something though, a scenario where an robot with legs can do better then a flying or wheeled one.
There, I made you afraid of cheap $100 quadcopters off Amazon and raspberry Pis.
Now you can buy larger quad-copters than I had built off the shelf.
Things change , people get used to it, opinions morph.
When robotic dogs and bipeds become more common place, people will become less threatened by them just due to sheer anxiety fatigue, if not more friendly reasons.
My take : when I see a robotic dog I think "That's a corporation with a spending budget!"
In another case, small CoTS RC planes did significant damage to a Russian airbase in Syria [1].
[0]: Death From Above: The Drone Bombs of the Caliphate https://www.bellingcat.com/news/mena/2017/02/10/death-drone-...
[1]: Russia Offers New Details About Syrian Mass Drone Attack, Now Implies Ukrainian Connection https://www.thedrive.com/the-war-zone/17595/russia-offers-ne...
More advanced "loitering munitions" (essentially, drones that fly around until they're sent to slam into a target) were used in the recent Nagorno-Karabakh conflict.
I believe that all these cases were human-in-the-loop for various reasons, but the possibility for automated targeting seems reachable soon if the armies would really want it (which is not certain).
That's considered reasonably effective by the militaries, and fully within the capabilities of many cheap quadcopters.
Wait, so they reserve the right to void your warranty if you use their product to make art?
I remember there was a case where a company bricked someone's device because they left a bad review[0], but having companies judge the artistic merit of their customers' use cases seems somehow even more dystopian.
[0] https://www.techrepublic.com/article/iot-company-bricks-cust...
That actually sounds even more awesome :)
Almost reminds me of the last DARPA challenge for humanoid robots where KAIST cheated with a robot which simply rolled from one place to another & did super slow stairs one leg at a time, which was exactly what the competition was supposed to overcome.
Can you explain more about how they "cheated"? it looks like they were awarded 1st prize.
You see it all the time in ML stuff. "Only a true AI could X" (with an implicit "and that would imply Y") and then people figure out how to build a system that does X but doesn't imply Y.
Turns out it's amazingly hard to build good standardize tests that don't fall into "When a measure becomes a target, it ceases to be a good measure."
1- Decide what you want made and how you want it made (by choosing ingredients and cooking methods from a UI)
2- Place your order
3- Automated Restaurant queues your custom order, picks ingredients, cooks according to your preferences, and packs it
4. Drone (or human) delivers it rapidly, while still warm
Better to fill a big warehouse-size restaurant with robots than to fill a million kitchens with robots.
user: cookingrobot
created: July 28, 2009I'd much rather have robots in my kitchen and my garden that grow fresh vegetables on my soil, adapted to my preferences, and turn them into dishes that match my taste, than the end-product of a ginormous industrial farming machine.
60-70% of the worlds food supply comes from family farmers, yet most pesticide and fertiliser usage is in industrial farming. Factory farming is only cheaper because artificial fertiliser is so cheap, but if you look at the "costs" on the entire system/ecosystem it's a lot more expensive.
If you only view food as a source of nutrition and energy, then this might seem reasonable, but it completely ignores the socio-cultural aspects of cooking.
Having a robot in your kitchen allows you to share in the work, to learn from it, and teach it. Maybe you cut the tomatoes while it peels the onions. Or you start cooking, and it finishes for you. Maybe you cook your lunch, but because you want to wake to the smell of freshly baked croissant, the robot does it.
There is an infinite spectrum of interaction, learning, and joy that could come from this, that goes beyond merely eating the prepared dish.
Maybe small restaurants in the style of old Japanese eating houses combined with community gardens would be a good middle ground though, but having industrial take-out everyday sounds like a dystopia to me.
Still missing the ingredient prep though.