Happy New Year everyone!
If you haven’t seen it yet, I hope you enjoy this video from Boston Dynamics. These are mechanical marvels in so many ways, and what is depicted in this video reflects years of work on top of the already impressive constructions that Boston Dynamics has shown off throughout the years.
Advertisement
I enjoyed the video for what it is – robots following a choreographed dance sequence, and then watched it again. Count the degrees of motion in each of the joints, such as the ankle. Observe how the bots are able to balance. Take note of the motions. Whoa.
There are some robotic products in consumer and professional sectors, such as Husqvarna’s robotic lawn mower from 10 years ago, Makita’s iRobot Roomba-like floor cleaner, and the new remote-control-driven Hilti Jaibot that drills overhead holes.
Consider all of the innovations that have come to quadcopter “drones” in recent years. How long until robots start following you around the jobsite, moving materials or even doing work?
It’ll be interesting to see how things evolve, if they do.
The SPOT launched last year, but I haven’t seen or heard of any jobsite-related applications or real-world usage examples for it yet. At this time, its main selling point to construction and utility industries is in remote inspection capabilities.
Think about what this tech might enable if it trickles into tool and equipment sectors. Imagine a tool box that didn’t roll but walked over uneven terrain, following you or moving to a predetermined location as it charts its own path.
But, maybe none of this tech will trickle into the tool industry. I mean, consider all of the sensors built into smartphones these days. And yet, most cordless power tools still lack anti-kickback safety tech.
Tom D says
Safety features like antikickback have to be really well done to be usable – otherwise they become an annoyance and the safety gets bypassed.
Or it’s a tool so large that the operators have to learn how to use it without triggering the safeties.
Norman Huizenga says
The Boston bomb Squad used spot for some time last year:
https://www.wbur.org/news/2019/11/25/boston-dynamics-robot-dog-massachusetts-state-police
Shane Hester says
Saw it on the news this morning….pretty cool!
Francois Caen says
When I’m hiding under a pile of rubble, during the long night of the post-nuclear winter, as T-1000s hunt me down and the fire from their lasers flies over my head, I will remember this video fondly.
Jon Schmidt says
For sure! I made a similar comment after showing my 12 year old the video. Something about robot overlords.
However this engineering feat is truly amazing. STEM all the way.
Stacey Jones says
You have 15 seconds to drop your weapon!
PETE says
green tips will object
Scotty. says
the dancing robots make me incredibly uneasy. Can we trust that programmers (or hackers) have an ethical base to use this technology the right way?
How about a brigade of them marching down Main Street programmed to enforce Covid lockdown dictates?
Stuart says
For one, you can’t just run out to Best Buy and purchase something like this – Boston Dynamics leases the robots to select companies as they further develop the tech.
One of Boston Dynamics’ conditions is that their robots cannot be weaponized.
If you’re concerned about ethical use of new technologies, there are a lot of other things to worry about over robots being reprogrammed for nefarious tasks.
Rx9 says
“One of Boston Dynamics’ conditions is that their robots cannot be weaponized.”
Stuart, I respect you highly, but I disagree with you on this point. Considering that DARPA makes up a considerable portion of the Boston Dynamics’ customer base, that statement is pure PR spin. Every major military is working on these, including our own.
Truth be told, though, you can’t hold BD entirely accountable for what happens to the tech. The building blocks (MEMS, li-ion cells, brushless motors, machine learning algorithms, etc.) fell into place and BD, along hundreds of others have been trying to put them together. Anything that can be used as a weapon inevitably will be at some point, so if BD doesn’t do it, someone else surely will.
The video is nice though.
Stuart says
That’s not an opinion, that’s what they said in the context of the Boston PD use article that Norman linked to above.
Perhaps only the SPOT isn’t to be weaponized?
Over the years, Boston Dynamics’ various robots have only been demoed as equipped with sensor suites or hardpoints for cargos such as backpacks.
They could very well have different variants aimed at civilian and military use, where the civilian and LEO-use robots have different clauses than those being developed for military use.
Before iRobot sold off their military division, they had separate robots aimed at different industries – the PackBot and similar platforms for military use, and the Roomba market for civilian markets. If Boston Dynamics has government research contracts (how else are they paying for all that R&D?), they likely have similar differentiation.
Still, with the Spot only available under lease arrangements, and with non-weaponization clauses, we’re unlikely to see armed robotic canines anytime soon.
PETE says
I’d bet money that Boston Dynamics has an disassociated company that is being handed the Boston Dynamics tech to do the weapon development for DARPA.
Of course there is no way for us to know about that…..
Tom D says
I just realized that this would have been more impressive if the camera itself was on a robot – revealed by a mirror at the end.
Hammy says
But really the robot was inside of us all along…
Chris I says
The real question here we need to be asking is this: who did it better — the Boston Dynamics robots or the wife dancing in the kitchen (https://youtu.be/9DioivdeAP0)?
You’re welcome.
Brian Collins says
What would be more impressive is one of these robots playing fetch with one of their robot dogs…forget the bad dancing
Stuart says
Object detection, such as identifying, tracking, and retrieving a red ball is actually pretty easy. There are consumer products that can handle the brains of this, such as the PixyCam (https://www.amazon.com/Charmed-Labs-Pixy2-Vision-Sensor/dp/B07D1CLYD2/?tag=toolguyd-20).
SPOT: Locate, retrieve, and return object. Repeat.
ATLAS: Identify and pick up object when returned, throw (or insert into ball-launcher attachment). Repeat.
Paul says
I have done a lot of automation projects. Three rules that I usually go by: (1) except in rare cases a machine is not going to be able to do something a human can’t do. Don’t expect it to solve problems we can’t do ourselves. (2) Don’t expect labor costs to decrease. They might or might not because what happens next is not necessarily what you expect. There was an old myth that when you bought a PC, you eliminated a secretary in the office. Oh and something about paperwork reduction. Neither one happened, much to the enjoyment of the paper industry in the 1980’s.
And (3) you need to be able to write down step by step in very mechanical terms exactly what you want the machine to do. In very exact language (when the sensor indicates full, shut off the pump, mix for 10 additional minutes). Any time there is a “judgement call”, good luck with that. Don’t try to “mix until smooth”.
With regards to point 3 though I have to go a bit further. There is a bunch of software out there called “AI” (Artificial Intelligence). Get over the idea that it is “intelligent” because it’s not. People working in this area are working with 40 year old technology first invented in the 1980’s. After 40 years and nearly unlimited money spend, it is little improved from the 1980’s efforts. AI is best on “neural networks” which are software simulations of how we think that neurons work. So we feed a neural network a set of thousands or these days, millions of patterns and the “correct answer” to train it, then feed it “data” and get a result back. This pattern matching system works very well dealing with cases where there are literally thousands or millions of inputs such as image recognition but it is not “intelligent”…it doesn’t “learn”. You can feed it 1 million images of a “hot dog” and it will recognize a “hot dog” but feed it an image of a pickle if that’s not in the training data, and that’s a hot dog, too! And despite years of research we cannot predict what the optimal or best configuration of a neural network is. It is all trial and error. We have no theories on why any of it works so we haven’t significantly improved upon the technology from 40 years ago. The current “solution” is…wait for it…hardware acceleration! That’s right we now need to add an “AI chip” to your cell phone so that you can run pattern recognition faster. As an example one of the state of the art systems is ResneXT-101, a state of the art image recognizer, that has around 90 million parameters and achieves an 82% recognition rate on an annual image recognition contest. That’s right….it misses 18% of the time. Current “champions” (Facebook, Google) manage to increase this to 85%. This would be great for a “google search” if it came back with the right answer in 4 out of 5 links.
But think about how happy you would be if 1 out of every 5 products you bought was defective. Or here’s another example. Tesla actually believes that they can use AI for the purposes of driving a car and use it as part of their “auto pilot” feature. So a couple months ago a doctor was “driving” home late at night in his Tesla in the local area. I can’t really call it driving though because what he was doing was watching a movie on his phone while the car was driving. This is a relatively deserted 4 lane divided state highway so very little chance of anything going wrong, except for the accident scene that was going on. The car got confused and plowed into a police officer at full speed. Obviously the officer was just one of those 15% errors that happens some of the time, but I don’t think the officer saw it in quite the same way. We can also see it in all the Fakebook “fact checkers” in use where AI tags postings that Fakebook doesn’t like in what amounts to hilarious ways.
Does that mean that we can’t have more automation? Certainly not. As a simple example, twenty years ago only very sophisticated mechanical maintenance crews outside of a shop environment had air impact wrenches. Now all you hear all day long is ratta-tat-tat from brushless battery impact guns and the air lines and air compressors have all but disappeared. Certainly electronics have dramatically changed both the weight and skill level required from welders. I would consider both of these “automation” on a low level.
As an example of a current ongoing project I’m doing the controls for a new steel (car) shredder. The plan was that one operator would run the crane feeding it and run the shredder itself, based on operating a smaller shredder at a sister plant. 10 minutes into startup it was immediately obvious that the crane operator is a lot more busy feeding it and that the feed mechanism requires constant adjustment in ways that infinite computing power from Google just won’t solve. Too much anticipation of what is going on, too many judgement calls and guesses that are not easily written down in a concrete set of instructions, and where a 15% error rate just won’t do.
As a “better” automation example when I was working at a plant making ductile iron water pipe they had one guy responsible for making “batches” of material to melt down to make into pipe. After automation productivity increased from 45 TPH to over 60 TPH and quality improved to the point where we pretty much cut out all use of extra cost ingredients in the batches, making pipe from almost 100% shredded cars. The “job” was eliminated but since production increased by over 30%, the guy doing the job was just moved to a different position in the same department. We could make pipe better, faster, and cheaper. So the guy that was probably impacted worked at the competitor’s plant, making our employees more secure in their jobs.
As a final example, consider what has happened in the agriculture business. Farming productivity in the 20th century increased somewhere over 10,000%. I’m not making this up. Farming went from the #1 occupation with around 70%+ of the population in the 19th century to less than 3% of the population. Yet we have to question then why did unemployment not skyrocket to over 50% during that time? Farmers made more money and food production costs dropped…which impacted everyone. A hundred years ago food was a much bigger part of our family budgets than today, which is how we can afford to spend money on basically everything else, which drives up the need for labor making cell phones that didn’t exist even 50 years ago.
As a society though we have to get used to change. There is no such thing as “job security”. Things grow and change so quickly that this is no longer possible.
Stuart says
Thank you for sharing this and for your insights!
In one example, Boston Dynamics says that their SPOT robot has saved one company 20 hours of employee time per week on site inspections.
But I would agree that I don’t see these robots replacing human workers. Similar to how an impact driver allows for faster completion of fastening tasks, more advanced robotic tech will lead to easier/faster/less risky execution of certain tasks, similar to Hilti’s overhead-drilling Jaibot system.
With respect to self-driving cars, I would expect for greater integration of assisted driving and object avoidance tech, and perhaps later on we might see greater automation once vehicles can communicate with each other to help further decrease the frequency or chance of collisions.
Tom D says
The thing a robot can do is stuff that would be hazardous to humans – think enclosed tank inspection for example.
The problem comes that many things that are hazardous to humans are also hazardous to robots – and the robots are expensive!
Stuart says
Robots are usually cheaper to repair and replace parts in the humans.
How much to repair a robot joint than say a person’s wrist, elbow, or shoulder?
Rx9 says
Excellent post, Paul.
Much like the snake oils and wonder liniments of the 1800’s, AI currently enjoys such a powerful position in the public mind precisely because so few understand how well it actually works, or even how it works in general.
I’m even more critical of the Tesla “FSD” tech now than I was 5 years ago. That they’ve achieved such a market cap speaks to the incredible power of perception management. That, and quantitative easing.
MFC says
The bi-pedal robots are absolutely amazing. Watching their very precise and almost human movement gives me hope that they will be launched into the mainstream soon.
Mike McFalls says
I found them being able to ‘jump’ the most impressive of all the movements displayed. If you stop and think about the intricacies in the movements and the dynamic ways these robots move and the fluid was of the joints, etc, you have to walk away at least a little impressed.
Stuart says
You might be interested in some of Boston Dynamics’ other videos, such as with their two-wheeled robot jumping onto elevated surfaces or rolling down stairs.
There’s a video with the wheeled robot, or an earlier version, rolling down an incline. I wonder if it’s able to stop on the incline, maybe if it ramps the speed down instead of a hard stop.
Thom says
I for one welcome our new robot overlords!