Catch up on stories from the past week (and beyond) at the Slashdot story archive

 



Forgot your password?
typodupeerror
×
Robotics

Boston Dynamics Teaches Atlas Robot Parkour (pcmag.com) 66

An anonymous reader quotes a report from PCMag: Boston Dynamics taught its robots to dance last year, now one of them can complete a parkour course. The robot company, which is owned by Hyundai, has released two new videos today. The first shows off how well the Atlas robot can perform parkour, while the second video takes us behind the scenes and explains how Atlas works. It's actually the first time Atlas has managed to complete the complex obstacle course flawlessly, and Boston wants to celebrate that.

After the dancing video last year, we all expect Boston's robots to do amazing things now, and the parkour video doesn't disappoint. But what it doesn't show is all the hard work, missteps, and problems that need fixing in order to make these demonstrations possible. That's what the second video reveals, and it's the first time I've seen these robots breakdown and leak fluids.

There's a hardware team, a software team, technicians, and operations staff required to make this whole operation work. Each Atlas requires three onboard computers in order to process the sensor data and run the algorithms allowing this 5ft, 190lbs robot to move with 28-degrees of freedom without falling over (very often). And even though we're asking these robots to do the same actions as humans, it's a very different challenge.

This discussion has been archived. No new comments can be posted.

Boston Dynamics Teaches Atlas Robot Parkour

Comments Filter:
  • When do they get mated up to a humanoid body?
    • It is humanoid. However, a human form body is at least a decade away while a lifelike human form body is 20 or 30 years off at best.

    • When do they get mated up to a humanoid body?

      Atlas is a humanoid body. Perhaps you're thinking of one of their dog-like chassis like BigDog or Spot. They tried to teach them parkour, but they were too old to learn these new tricks... :-)

      • When do they get mated up to a humanoid body?

        Atlas is a humanoid body. Perhaps you're thinking of one of their dog-like chassis like BigDog or Spot. They tried to teach them parkour, but they were too old to learn these new tricks... :-)

        Well you know - maybe human-like would be a better description. A few weeks back I saw a Japanese robot - it was an amazingly lifelke thing- other than it had a blue plastic "body". But the demonstration was more for facial movements, voice, and skin quality on the face. The body didn't do anything, and was just there to avoid the Marie Antoinette effect.

        There are some "adult toys" that are pretty realistic looking. Mate up the Boston Dynamics robot, with the skin quality and looks of that Japanese robot

        • "Real Doll"?

          They will get there, for now, the power source and the robotics are too big to fit in that package, perhaps in the future that will change though.

          • "Real Doll"?

            They will get there, for now, the power source and the robotics are too big to fit in that package, perhaps in the future that will change though.

            Yeah. I suspect that miniaturization of components will come a long way. The tittlization (I think I just made a new word) market plus the mobility market will drive that aspect.

            While the realdoll market niche is kind of obvious, the mobility market could be simply amazing. Not only for the obvious, like quadriplegic/paraplegics, but there are a lot of people out there with balance and muscle weakness issues. Or just older folks, to eliminate falling.

    • Like a mechwarrior?

      • Like a mechwarrior?

        Why not? I was thinking more along the lines of human companion contraptions - think along the lines of an escort droid for women or men. Where in the present age, women could design their perfect man, and men who aren't of the desired demographic can still get companionship. Win/win situation.

        But a mechwarrior - or general augmentation of human abilities for particular heavy or dangerous work would be pretty awesome.

        Or a mechanical contraption that allows people with various movement problems to walk

        • Yeah I can see this being a futuristic wheel chair of sorts.

          • Yeah I can see this being a futuristic wheel chair of sorts.

            And in some cases, might be a therapeutic accessory. There may be nerve grafting in some cases, and the mech will keep the muscles in shape while the new nerves relearn how to work normally.

  • You've got to yell "Parkour!" while doing parkour or it's not really parkour.

  • What can it do besides useless human tricks like dance moves and stepping and jumping? It still only wanders around the lab and only when you run a program to do it. Can it take out the trash or fetch lunch? Anything useful?

    • Re: (Score:3, Insightful)

      by MrKaos ( 858439 )

      What can it do besides useless human tricks like dance moves and stepping and jumping? Can it take out the trash or fetch lunch? Anything useful?

      Not yet, but the day will come when it will be able to chase you down and catch you whilst a human casually walks up and arrests you. Knowing Parkour won't help you escape.

      You need to ask *why* are certain skills being robotized.

      • At what point does it have enough skill to be useful for anything other than advertising videos? Its been in development something like a decade.

        • by MrKaos ( 858439 )

          At what point does it have enough skill to be useful for anything other than ? Its been in development something like a decade.

          As I said You need to ask *why* are certain skills being robotized. Doing something useful is a conglomeration of skills. For example impulse control and extrapolation are skills that when used in combination allow you to get a reasonable idea of where a certain scenario is headed.

          Try looking ahead to later versions of this machine and what can be *done* with it. Right now it's still only useful for advertising videos so that it keeps attracting funding.

    • Re:Useless (Score:5, Insightful)

      by timeOday ( 582209 ) on Tuesday August 17, 2021 @05:32PM (#61702849)
      You're not wrong, but it's sad to see that attitude when somebody engineers something at a level never accomplished before. It's hard to say when this will reach the threshold of profitability or what other technologies might need to be invented to combine it with before that will happen. But the progress they have made is real, in fact it is amazing.
      • Its been a decade and they have videos of dancing and jumping. They made it walk and jump a long time ago, these videos seem like sequences of things it already does, its not a big step forward from the last video, or the one before that, or the one before that.

        • Re:Useless (Score:4, Informative)

          by timeOday ( 582209 ) on Tuesday August 17, 2021 @05:51PM (#61702919)
          No, not jumping around on uneven surfaces like that. Untethered, so quiet and controlled!

          You just have a short memory. Here's something from 9 years ago:

          https://www.youtube.com/watch?... [youtube.com]

          • You made my point exactly. They made the basic moves a long time ago. What is really new in this 2021 Parkour video from the 2018 Parkour Atlas video?
            https://www.youtube.com/watch?... [youtube.com]

            I'm having a hard time measuring the progress in 3 years. Still just gimmicks. Call me when it can take out the trash.

            • Could you, personally, do everything you see in the video from 3 years ago?

              OK, now, how about the new one? (Watch all the way to the end before answering).

              Be honest now.

              • Why would I need to do back flips? What is your point? How would my life improve if I could do back flips? How would my life improve if I had a robot that could do back flips?

                Atlas could work as a pretty amazing ladder/lift hybrid, it can get in tight places like a ladder but move around like a scissor lift. There is something useful I can't do, act like stable ladder while moving.

            • I'm having a hard time measuring the progress in 3 years.

              That's because you judge a company's work based on 2x 1min long marketing videos of their successes. If you saw their failures, the problems they had achieving this, or even had a first year university level of understanding of robotics you would see those 3 years of progress as an amazing accomplishment.

              For one at no point did the robot in your video 3 years ago need to account for using momentum to stay on a surface as they do in this new video. It is an order of magnitude harder problem using momentum to

              • That's because you judge a company's work based on 2x 1min long marketing videos of their successes

                Also, yes, we see only a 1 minute marketing video. So there's a good chance it doesn't actually work like that 90% of the time IRL. I could make a 1 minute marketing video of myself which would make me look like the coolest dude ever too. But the reality is, well, different.

                • If you would have bothered to take a look at the second linked video in TFS, there they show you some the failures and technical problems they experienced.

                  Sorry, I clearly was not accustomed to the idea that people don't have enough attention span to even read the summary anymore. Hope you managed to get to the end of the first video or is your brain already too far TikTok'ed ?

                • And yet their marketing videos have been turned into actual products. I have had the opportunity to use Spot, and he worked *exactly* like the 1 minute video we saw, in our hands, on our premises, on a surface he'd never seen before.

                  I don't get the comments here. One person says the video shows no progress after 3 years, the other says they are lying and the robot can only work in a highly contrived example.

                  Which is it? Find the frigging middle ground.

                  FYI, the entire point of Boston Dynamics is R&D into

          • jumping around on uneven surfaces

            Uneven in a very even sort of way. Mind you, I'm still impressed, but it's still a considerable distance away from natural terrain. I suspect the scene geometry was heavily preprocessed. But I'm still impressed. Maybe in five more years they will be able to do Olympic level floor exercise, and goodbye shin splints. Hello battery fires.

          • The reason the new ones are untethered is because they now have an isolated room to test in instead of a test rig in a shop. The umbilical was always to supply power. They switched that out for an ICE. The ICE also means it doesn't qualify for military or NASA contracts. Without those early adopters, there's no market for these things. They're quiet not because they're actually quiet but because of the sound engineering done on the video.
            • by aitikin ( 909209 )

              The reason the new ones are untethered is because they now have an isolated room to test in instead of a test rig in a shop. The umbilical was always to supply power. They switched that out for an ICE. The ICE also means it doesn't qualify for military or NASA contracts. Without those early adopters, there's no market for these things. They're quiet not because they're actually quiet but because of the sound engineering done on the video.

              Source?

              Everything I'm seeing is still saying electric motors [reuters.com], but nothing about engines.

              • The motors are electric. The power source is a small ICE. You can see into parts of it. The only place to keep the power source is on the back. It's roughly a 30cm by 20cm area. It couldn't be more than a few cm thick. Even lithium polymer batteries of that size couldn't power several dozen servos, three beefy computers, and a lidar array for more than a 5-10 minutes. OTOH, a 1kW ICE would be about the size of a snow blower motor.
    • by jcdick1 ( 254644 )

      Why climb Everest? Or go to the moon? Or send probes to Mars?

      • Why not make Atlas do something useful? It can do back flips and jump over your desk but it can't bring you a cup of coffee or wash the coffee cup. What do I need a robot solider for?

        • What do I need a robot solider for?

          I think you just answered your own question. Put a gun in the claws of that parkouring robot and someone will find it useful.

    • What can it do besides useless human tricks like dance moves and stepping and jumping? It still only wanders around the lab and only when you run a program to do it. Can it take out the trash or fetch lunch? Anything useful?

      I feel sorry for you. You must lead such a boring and angry existence.

      • Because I'm not impressed by a robot warrior that doesn't do anything useful after a decade of development? There is 1 new movement from the 2018 video. For all the movements Atlas can make why can't it pick up a garbage can and dump it?

        Call me sad, I call the progress and usefulness of Atlas sad.

    • Some of Boston Robotics quadruped models already commercial uses (and are being sold) - e.g. for security patrol or police work going into dangerous situations. In the future we won't need heroic seniors to sacrifice themselves to go into places like the failed Fukishiima nuclear plant - we can use robots instead.

      As far as the bipeds such as Atlas go .... We live in a world designed for humans, from steps to doors, elevators, escalators, sidewalks, etc, etc. Anything capable of being able to do useful stuf

      • Wouldn't a video of the robot getting a tray from a table while stepping over the dog be more impressive?

        I'm not against progress. I'm just not impressed with what look like programmed moves, mostly the same moves we've seen before. Let's see it in a new situation it has not seen before. The content of these videos does not impress me for a decade of progress. Let's see it open a door and step over the dog, then dump the trash - honestly seems capable. Instead more jumping and walking on an incline.

    • Apparently IT can't actually do SHIT without a bunch of help and special casing. If you can't just give a set of waypoints and say "go forth and follow them" then the robot isn't doing the job, you are collaborating with it at best.

    • I suspect that DoD and other military agencies are happy with that ability to handle complex terrain. If it can do that, what's keeping it from carrying ammo across a scarred battlefield, or running into a burning building?

      Those aren't exactly something fun to show in a public promo, so I'll be happy admiring the backflips.

      What I'd like to know is how much of it is hard-programmed? Are they telling the robot to run a certain path, and then it recognizes the obstacles and manages accordingly, or is it close

    • Not sure if it can take out the trash, but it can probably take out the Taliban.

  • It's not useful until (if) they attach dextrous hands to it. Jumping around with bent legs like a constipated clown without toes is useless. Plus you don't see in the video that it has as high as 50% fall rate for certain maneuvers. I don't think Hyundai is forward thinking enough to invest the $1 billion more needed to make it useful.

    • Please tell us what you have achieved today. Video, or it didn't happen.
      • Nobody handed me $1.1 billion to develop a robot, did they? As for what I achieved today, I posted a useful comment on slashdot pointing out the uselessness of the boston dynamics robots. By pointing out the flaws and what it needs to be cool, I hope it will inspire the next generation of slashdotters to get off their ass and do something useful, like posting something insightful.

      • He made a valid point.

    • Thanks for your insights into what it will take to mechanise the field of industrial parkour. It is a vast untapped market.

  • by Anonymous Coward on Tuesday August 17, 2021 @05:21PM (#61702827)
    If Boston Dynamics makes a robot that posts to slashdot when it's supposed to be working we are all fucked.
  • Obviously a fake ..
  • seems like the perfect thing to send to mars

  • Comment removed (Score:4, Informative)

    by account_deleted ( 4530225 ) on Tuesday August 17, 2021 @08:54PM (#61703317)
    Comment removed based on user account deletion
  • This may be good incremental work, but it was more like walking on irregular surfaces (as a human would do on a nature trail) than parkour. Plus backflips, which the Boston Dynamics robots have been doing for 3-4 years now.

  • This is so cool, but ...can't help it. Unlike previous Boston Dynamics demos this fits right there in the middle of the uncanny valley, for me at least.

  • Pretty sure all this has happened before, and it will happen again, if they ever put Caprica re-runs back on the air.

    Hyundai - Robots That Make Sense!

  • So it can chase me down through complex terrain.

    Super.

  • That wasn't parkour. That was combat drills. Let's get that straight right out the gate.

    What you need to appreciate about the Boston Dynamic bipedal robots is that they're a scam. They're not useful for combat, construction, or anything else really. The human form is not the most efficient for those tasks. The reason humans do those tasks now is that those tasks require a human to do them. If you want a robot to go into combat, you don't design it like a person. You don't give it legs, you give it stair-cl

  • Won't be long now and the robots will be doing American Ninja Warrior.
  • I'm waiting for them to learn hand-to-hand combat or knife fighting. Imagine one of those suckers with a baseball bat!

  • Don't get me wrong, what they do is great but when will they be profitable enough to stop getting handed off and sold like a cheap hooker?

You will have many recoverable tape errors.

Working...