Atlas robot does parkour
blog.bostondynamics.com
blog.bostondynamics.com
I want the source code to firefighting bots to be public so I know whether my town buys the ones that prioritize babies over elderly. Or stuff like it’s easier to save 5 dogs than 1 human.
Nurses too have great power for good and harm. Currently when nurses decide on mercy killings [0] it requires lots of deliberation to figure out and it’s just one person. If a robonurse is programmed to put lethal morphine into a patient under certain conditions, that’s a big deal.
[0] https://www.npr.org/templates/story/story.php?storyId=122054...
Also, I looked a while ago but is there anything up and coming in the anti-facial recognition space? I know that some guy created an almost life-like replica of his own face for sale, and then there were some sort of reflectors, but anything that cannot be outlawed, like a full face covering, would be cool.
https://www.theverge.com/2019/4/23/18512472/fool-ai-surveill...
Robots should be able to swarm and restrain uncooperative suspects so there should be less need for lethality unless the perpetrator is an active threat against other people -a hostage situation, mass shooting, etc.
It used to be cops could make their own decisions on the spot about a subject (good or bad), now with cameras there is much much less ad-hoc decision making. Everyone gets booked.
It’s a double edged sword. Depends on what we want.
Or, do you want more enforcement, period. Automation can scale a lot more than human LEO - and since everyone breaks some law pretty much daily, that limitation is a feature in my book. Didn't turn on your blinkers for long enough before switching lanes on a deserted street (by 0.2 seconds)? Jailerbot 2000 deactivates your car, detains you and sends you to the fully automated jail below the city until the judge sees you on Monday.
Here's what I'm thinking the good future looks like: Let's say I have a big back yard and would like to have that be a productive farm. AI bot comes over, surveys the land, does soil tests, orders seeds, fertilizer and garden equipment off Amazon. Plants seeds, sets up permaculture, sets up irrigation, installs monitoring, goes next door, repeats for every house in the neighborhood. That's the kind of AI future I want to see. Software codifying knowledge of an experienced practitioner to manipulate the physical world in a way beneficial to humans.
It hadn't occurred to me that drones let despots wage high-tempo modern wars (at least, partially) without the associated risks (of devolved control) and costs (of very expensive training and equipment for your soldiers)—but they totally do. That's a pretty big deal. Wonder if we'll see 2nd-tier, regional powers gain influence through this effect, as they can afford quite good drone programs while weaker neighbors might not be able to.
The drones will be effective for a while, then it’ll become impossible to fly them in airspace controlled by a 1st or 2nd tier power, and you’ll be back to needing foot soldiers (if only to take out the drone jammers).
This is just a semi-autonomous vehicle that is bipedal or quadrupedal instead of having wheels. That's it.
Agreed but I’d also like to see human rights protected from the new decentralized governments. Not sure how that could work.
Here's a real world contemporary attempt to implement some approximation to that. I'm not claiming that it's anywhere near perfect... but it seems to be working, so far:
https://pdfhost.io/v/Hrr2IgtuS_SocialContractoftheDemocratic...
The weird thing is that the more I study history, the more I think that the real dystopia started a century (and perhaps a couple millenia) before I was born, and we're just so adapted to it that we think it's normal. Take the "[Ro]bot" out of your headlines and those would be headlines from today, and if you ignore that the media never bothered to report on those a century ago, they'd be headlines from a hundred years ago. We were depersonalized by the industrial revolution, and then large numbers of us were killed in the wars that followed.
In some ways, this gives me a weird sort of hope that our descendants will welcome their robot overlords. In others, there's a melancholy foreboding that our descendants will welcome their robot overlords.
> According to anarcho-primitivism, the shift from hunter-gatherer to agricultural subsistence during the Neolithic Revolution gave rise to coercion, social alienation and social stratification.
As though sexual coercion, social ostracism and strict social hierarchy didn't exist beforehand. Take, for instance, the Chimpanzee.
The Polity are these benevolent AI that rule humanity after "The Quiet War". It is called that because not a single shot is fired because they just take control of all hightech weapons and technology one day.
The books would be considered utopian scifi if the rest of the universe wasn't so gritty. Which is why it is my favorite series.
Asher world builds literally better than anyone I know - his aliens feel truly alien.
The Hooder is straight out of nightmare fuel.
The Polity universe is technologically comparable to the Culture but in almost every other way darker, grittier and dirtier.
There are quite a few parallels between them though - enough that if you love one you'll at least like the other.
The AI-ruling-humanity idea took a bit to warm up to. They do seem to be presented as benevolent overlords.
I have some sympathy for this view, but there's also the fact that at this point we've basically solved hunger from a societal point of view.
Deprivation of calories from a population now happens because of war or intentional. At the current time we no longer are at the whim of natural causes of hunger.
As I^HYuval would say, those are the things that help start the long journey to the mess we're today.
So, I would agree with OP in that it started with invention of agriculture, in a sense that that gave rise to rigidly hierarchical societies, and centralized states that could extract surplus from their citizens and spend it on warfare and other forms of oppression. But we didn't have to take that route, and it doesn't mean that agriculture per se is bad.
We were depersonalized loooong before that - read up on the 30 years war, or any other conflict before the Industrial Revolution.
“Soldiers” of the time were mercenaries, many of whom were pressed in to ‘service’ when their village/town/farm was destroyed by said mercenary company. These mercenaries fought at the whims of princes and kings who squabbled over whims and privilege, and the princes and generals are the only ones who are remembered.
the current estimate is that there have roughly been 105 billion birth since 50 000BC, with our current world population at almost 8 billion... about 8% of all humans that have ever lived are currently still alive.
The misery is pretty localized in Africa and some places in the middle east, which is less then 20% of the population.
Life could be better in a lot of places, but it's incomparably better to life under the churches in the dark ages for example.
Given the number of countries with mandatory military service, and the looming threat of the draft in countries that don't, this maybe isn't capturing the whole story.
We're DONE!
The ultra rich and the Jeff Bezos of the world will buy them and return us to a world of Kings and Queens. Good luck trying to fight that.
They might select the cream of the crop, wipe the rest off the planet to restore it to pristine condition and live in infinite wealth forever.
That would be much harder in feudalism. Not outright impossible - if they were bored with you, they wouldn't object - but if the king wanted to keep you, it would be risky to defy him.
One of the reasons why Late Medieval and Renaissance Italy was such a hotbed of intellectual activity was that local nobles had only very limited reach. If you got into conflict with one prince, you could move 20 miles away and be under protection of somebody else. And that could be done almost overnight.
The rich never needed robots to buy the police. There is more than enough people willing to be cruel and inhuman in exchange for lunch money and the feeling of superiority.
And the robo-donkey thing was trialled and rejected by USMC for use in the cargo carrying role - too noisy, needed to much direct control and other things
To live in a world perfectly capable (in a purely physical and logistical context) of providing necessities of living for everyone regardless of their output or decisions, but chooses not to, is pretty dystopian to me.
Living standards are way higher today than a century ago.
You are comparing against a non-existent Utopia.
>Living standards are way higher today than a century ago.
How much higher would they be without capitalism? Probably a lot.
>You are comparing against a non-existent Utopia.
I realize it doesn't exist, that doesn't mean it shouldn't, or that the reasons it doesn't aren't immoral and unethical.
We have a lot more examples of failed capitalist countries than failed communist ones. Failure cannot be directly linked to a country's economic system. It happens in both for a wide range of reasons.
I recommend you do more research on systems that have tried to eliminate money, private property and trade. While there has been some US intervention to disrupt them, as in the case of US backed coup attempts, funding of political and armed opposition, and sanctions, the totalitarianism and dysfunction they create is so extreme that there is doubt that anti-market systems are unnatural and unworkable.
>>We have a lot more examples of failed capitalist countries than failed communist ones.
Look at China before and after its 1978 market reforms, or East versus West Germany, or South versus North Korea.
We have socialists in the West, like Jacobin magazine founder Bhaskar Sunkara defending the Berlin wall, that was used to prevent people in East Germany from fleeing, and even the brutal murder of Czar Nicholas II's little children, to avoid admitting that socialism is harmful. Beware of who you believe.
OTOH how many people are aware of Rojava or the Zapatistas, and their track record?
Maybe. But that's not different than human cops who already show implicit bias and poor facial recognition and kill people they have no right to kill.
We won't trust AI with decision-making, but we will trust the status-quo police forces to dish out violence from the safety of an office.
But for now, the sensing, state estimation, and energy density problems preclude the use of these robots in the real world. There's far too much uncertainty in the structure of the world or the modelling of unknown areas to do parkour outside -- unless I've missed something huge that boston dynamics has done without talking about.
As someone said below, however, if you can develop a hybrid control scheme with a human controlling a smart-ish legged vehicle, you're on a roll. That's the future for legged vehicles that I could see: something like mechwarrior or titanfall, but without the AI (remote operated). Or Dreadnaughts. But small (as someone else pointed out).
Much depends on scale. BD is working towards a robot-donkey, something that can follow soldiers over rough terrain. At scales much larger than a horse, such robots already exist and are already much more mobile than soldiers in most terrains. These robots are called tanks. Or trucks.
Once a vehicle is carrying a large enough load that it will not fit between the trees and so cannot follow soldiers into forests, legs are not very useful. Wheels and tracks are better than legs, particularly in soft/wet terrain. So large walking robots are likely never going to happen. The MechWarrior fantasy will have to live on in things not much larger than a horse, things that can fit between trees and along footpaths.
But, I read somewhere that only 20% (ish) of the earth's surface is accessible by wheeled / tracked vehicles. I guess the important part is that's where the people are.
Moreover, of the places that are accessible to tracked vehicles, many are accessible only to small tracked vehicles, smaller than a typical car, and not at all to tank-sized vehicles.
That was done years ago, as the Legged Squad Support System.[1] It sort of worked, but was too expensive and too noisy, being powered by a small IC engine. The USMC abandoned the project in 2015.
Boston Dynamics' big accomplishment was to get enough funding to actually do anything useful in mobile robotics. Most university robotics projects were a professor and a few grad students, and took forever to get anything. BD spent upwards of $100 million of DARPA money over a decade and got some working demos. Then they were funded by Google, and now Softbank.
Big walkers have been built. The Timberjack, from John Deere, is probably the best example.[2] This is a large six-legged off-road machine equipped with a long arm with a chainsaw. It worked but was not cost effective vs. wheeled systems.
The message is that with enough money you can do legged locomotion, but so far, there are few profitable applications.
[1] https://en.wikipedia.org/wiki/Legged_Squad_Support_System
Have you ever seen your cat/dog stare at a look-alike version of stuffed toy and become completely bamboozled? Humans are particularly magnetized by anything that has 2 legs and moves like itself. No one bats an eye on a sentry with a gatling gun, but how dare you put 2 legs under it - sudden dystopia.
(side note : ios slide to type will not allow the spelling of murder, try it.)
Edit: I think from a seek and destroy perspective we are much more likely to go the cloud of small drones route than these. These would serve as an armed / peace keeping / law enforcement. presence but would not be nearly as effective in a real combat situation as the drone swarm.
1) carrying enough energy
2) getting the price point down to what people are prepared to pay
for telepresence.
These are very tough indeed.
After all, the whole point of all this motor modelling is to enable the robot to do things humans do, like noticing when its foot lightly touches its own tether, and then using its hands to unhook the tether from its foot, temporarily pull it lightly aside, step around it, and then drop it. (Think: vacuuming a room with a tethered vacuum cleaner.) That's a very hard thing to teach a robot to do; but a human that realizes they're "plugged in" would do it almost intuitively.
A few meters of high-current cable weighs a LOT. Exerts large forces in weird directions and adds to your motor requirements.
On humans and tethers:
Do you like wifi? Why not stick to Ethernet?
SCUBA is diving without a tether. It’s popular.
Climbers use ropes so they don't die, not because ropes are fun. Bouldering is popular.
Singers and guitarists tend to move to wireless for performance as soon as they can afford to (purists excepted).
Because tethers suck. Suuuuuuck.
Look up saturation diving. The link from the diver to their dive bell — now that's a tether. One critical to human life, unable to be lost for even a moment, and yet which humans freely do freaking oxyacetylene welding around. And yet it still works out fine, in 99.999% of dives. Humans are good at managing tethers!
But to be clear, my point isn't that tethered robots would be preferable to or even as good as untethered robots. Just because humans are good at tethers doesn't mean they like them. So of course untethered telepresence robots would be better, if they were practical.
My point is that tethered human-equivalent telepresence robots — in combination with human operators who can map their intuitive management of their own body-plan into management of the robot and its tether — will be practical in an absolute sense, long before untethered human-equivalent telepresence robots become practical. We'll have tethered telepresence 'bots doing real, useful stuff out there in the world (decomissioning toxic/irradiated structures in non-polluting ways, say—or replacing those sat divers!), long before we manage to make batteries dense enough to power untethered telepresence 'bots for the same purposes.
And by the time the CapEx would get low enough to use telepresence for such things, we'd probably have the general strategic AI to not need it.
Also I believe 50 firemen robots is unlikely the best way to fight a fire :p Like autonomous cars: impose autonomous cars to lower car accidents in cities and save lives©? Just design the streets differently and prevent the cars to drive so fast…
[1] https://www.amazon.fr/Merci-changer-m%C3%A9tier-Lettres-robo...
... and then be ashamed of the lives lost because emergency responders couldn't get to the scene in time. :(
Everything is tradeoffs.
edit: that’s not to say they’ll never be an issue. They almost certainly will, but fighter jets can’t hold ground and neither can any robot.
I think we'll find states that fail to do that, suffer. The other way out is rejection of technology to a large extent, but I doubt many will try that model as it'd cause a large & swift hit to quality of life.
(why, yes, I do have a fairly large technological-determinism and social-groups-experience-evolution-like-anything-else-does streak, why do you ask?)
Centralization of power invites corruption and leads to social fragility.
A bit like the GitHub AI coding controversy. Yes, we maybe could AI ourselves out of a job, maybe? We can decide not to tho.
We have complete control over what we DO and DO NOT.
If humanity could organise well enough, we could collectively decide a weaponless utopia.
But alas, we are stupid.
Keep it humanoid so it can work in the same space as me, then it can fold itself into a box or something when not working.
Amazon currently pays people to fill boxes for a good economic reason.
The robots/technologies are just tools. You have opreasive govenments spanning generations without robots and you have democracies as well.
I certainly want more robots around to do the boring and risky jobs.
Even if wars are conducted using robots(which is pretty innevitable) I see it as a good thing. The best tech wins instead of the most barbarian or the better propaganda machine.
To me, I like to see when they stutter or slightly-fail, since that to indicates the dynamics and real-time part of the thing, that it is not a 100% scripted thing (although the blog post mentions the engineers fine-tuning the celebratory arm-pump, to what extent is that scripted then?)
1. [EDIT: IGNORE THIS POINT] They've done this before. [1]
2. To get more specific, look at the robot's legs around 0:09-0:16. They don't seem to be moving normally.
I don't have professional video analysis chops, but I'd put some money on this being fake.
Happy be to be proven wrong, though.
EDIT: It appears I remembered this story incorrectly, and posted a link without reading it. I retract my first statement. I'll leave the link for posterity, though.
[1] https://www.theverge.com/tldr/2019/6/17/18681682/boston-dyna...
Your link says nothing of the sort.
This behind-the-scenes video says it is choreographed and scripted. It also contains instances where things go wrong as well.
Here's my logic / assumptions: How does the robot "know" (have sufficient model information to predict the response) that the board will hold it when it does a vault? It doesn't -- it is told to vault by placing one limb on top. As we've seen with self-driving cars, where the sensing is 99% of the problem, that's the barrier between a closed set and the real world.
Don't get me wrong. The fact that Atlas / Boston Dynamics is in a state where the sensing is _mostly_ what remains a problem is astounding.
In general, I'm an AI researcher and an AI skeptic. After 10 years in robotics, I'm learning that the Human Is Cheaper and where Human Is Expensive, we prefer toasters, not terminators. (just enough hard-coded / hard-wired adaptivity to get by with tons of supervision)
At which point you're back to an automated assembly line which needs very little dynamics or intelligence. The whole point of developing such advanced robots is so that you don't need to worry as much about that can on the floor.
It's pretty hard to get a horse to careen off a cliff or straight into a tree. The largest risk for riding a horse drunk is that it'll inadvertently scrape you off on a low-hanging branch.
The # of important people who died in a car after a driver error is much higher.
> Ten army corps were observed over 20 years, giving a total of 200 observations of one corps for a one year period. The period or module of observation is thus one year. The total deaths from horse kicks were 122, and the average number of deaths per year per corps was thus 122/200 = 0.61.[1]
I think there's also a tendency to go "You just need linear regression" a lot when actually, well, there is a reason people are excited about new AI techniques: they make things that were very difficult much easier, and they make some things possible that never were before.
Is deep learning / reinforcement learning actually useful here? Maybe. I would be surprised if Boston Dynamics wasn't at least looking into it.
And then there's where it takes off. There isn't a single human that can come anywhere close to the language knowledge base AI algorithms can build up. Even a team of 100 humans would not have the breadth of languages algorithms have these days. Yes, context understanding humans still have the edge, but it's shrinking every day. But so many language aspects, from translation to grammar, machines outperform all but the very best humans, and 99.9% of all humans even within one language.
I don't think I've seen discussion about how you would stop one in an adverse situation.
Hogtie cowboy style? Taser? Paint splashed over sensors?
I was thinking some high strength fishing line fashioned into a bolo sling would be cheap.
Or maybe just some “invisibility cloak” made out of astronaut blankets.
Doesn't seem particularly hard if they're in range.
Now if they're sniping you from a mile away, that's a bit worse.
Humans snipers regularly make kills at that distance or farther, so that's not really a new threat.
I'd be more worried about the development of supersonic AGMs that are smaller and cheaper. Right now, it doesn't make much economic sense to fire a $10K USD missile to kill random people (though the USA does it anyway), but if it only costs $1K... now you're starting to get to WMD levels of murder.
A robot, maybe with drone support, could be really accurate at launching grenades or unguided mortar rounds. Or hell, attach it to an artillery gun. Much cheaper than missiles.
Or I’m wondering if it looks like both feet are off the ground when really only one is and supported by the other?
Alas, I can't help but keep imagining a Terminator climbing and jumping over barriers, instead of those smaller, less-threatening robots:
https://www.youtube.com/watch?v=wSXl_XKXZAI
I'm rather surprised no one else has mentioned James Cameron's robot-from-the-future nightmare so far.
And clearly the two robots should fight.
I feel like we've been seeing increasingly sophisticated demos like this for over a decade now and yet nothing has really made it to market that I can tell.
Maybe there just literally aren't many good use cases for the types of robots made by Boston dynamics?
In perfect conditions if you run it 100 times this is the result.
In the real world they have no application as they fail pretty much always in real conditions. But this is research, pushing things in controlled conditions help you learn what you can do in less controlled conditions. Though sometimes it's just a deadend.
The exact same thing is true of YouTuber stunt videos. Any very impressive stunt was failed hundreds of times but they show you the footage where they pulled it off. The thing to remember is that media is largely fake or deceiving, presenting the very best and not the reality.
They do have support pages and user manuals online.
My understanding is that the majority of research was completed within the past few decades, in comparison to the relative nascency of modern deep learning techniques.
Even once robots can do amazing things, there needs to be an economic use case that's compatible with what they can do, how they're powered, and how long they can work for.
Similar to quadcopters in a way - they have changed the world in some limited ways, but so far they haven't scaled up to carry people or out to deliver parcels, so they remain niche. If someone figures out one day that robots can deliver burritos/assassinate dissidents/stack shelves efficiently, it could be a quick takeoff.
https://youtu.be/tF4DML7FIWk?t=49
Also are the layers of foam over the landing surfaces they backflip onto needed for dampening?
https://www.youtube.com/watch?v=Dx-vUsShk9U (Robot fight in “Outside the Wire”).
For the war fighting it's pretty much always more effective to use things like tanks or drones rather than dealing with walking motions, that's still pretty inefficient.
Nice job, nonetheless.
I'm not sure I understand why? HN is usually pretty tolerant of healthy skepticism. And it's not as if we're saying "this feat is impossible, Boston Dynamics must be lying". We are (or at least I am) saying "the thing I'm looking at doesn't look real".
Too cynical?
Sweeping tunnels?
Construction?
https://blog.bostondynamics.com/atlas-leaps-bounds-and-backf...
No, they’re not “humanoids”, and it’s chilling to hear them call it that.
But more simply than that: this will be used for war and human surveillance and response. It is already underway.
TL;DR: Why?
Why allow "useless eaters" to roam the planet, dealing with them coughing on your parfait when you can have one of these sanitary machines do it? And of course, when we tell the parfait makers that they're out of a job, they'll freak...now we have the perfect, emotionless authority to keep them in line.
Definition of humanoid
(Entry 1 of 2) : having human form or characteristics
// humanoid dentition
// humanoid robots
Seriously, is there something simple that I can put in my glovebox of my car to use in case these robots start getting scary? Is there a simple household good that will work well against these, until they start explicitly shielding against stuff?
Edit: Like this:
https://www.youtube.com/watch?v=Es2rVHLKS3g
https://www.youtube.com/watch?v=WVy5Vm43X_A
There's also bulk rodent glue, if you could figure out how to shoot it :) https://www.domyown.com/catchmaster-bulk-glue-can-p-15693.ht...
I’ll be looking for some Holi festival type gear.
Is there a paintball shotgun?
Yes, there are [0]. But it seems like they shoot “slugs” and not a spray of paint in a cone.
Less destructively, a good net launcher seems like it'd stand a decent chance at fouling things but they are usually single shot and inaccurate at any kind of range.
Presumably most systems will use multiple sensors and vision systems, LiDAR, RADAR, etc. The key would be to disable enough of each to put the machine in 'safe mode'.
Vision systems are the easiest to attack. A cheap paintball gun, spray can, or fire extinguisher would be more than capable of disabling a camera system.
LiDAR would likely also be fairly easy to disable. Most LiDAR systems also depend on visual light, and have a revolving sensor of some sort (though there are some fixed units as well) which have their own weaknesses. A reasonably powerful pellet gun, or most rifles or pistols would be enough to disable a revolving lidar sensor, and a few shots with paint would disable a fixed unit, much like a camera.
It's almost certain the designer of such robots will think of these attacks and build around them. Think windshield wipers or compressed air to clear debris. So, the ideal material would be very thick, very sticky, and completely opaque. It would also need a high pressure container and a nozzle that produces a laminar stream to hit the sensors from at least 10m away.
Or Reason. Whichever available. Aim for torso.
Obviously would need to modify to generate more energy.
I’m typically like this for everything, but for robots that can jump around and stuff (and will likely soon do police and military actions) this is novel enough for me to explore this for the first time.
I think similarly about facial recognition software for how to evade and confuse. And also for biometrics, locks, etc. Any of these things are interesting to me and I want to know how to do stuff with them.