Boston Dynamics Atlas Dynamic Duo Tackles Obstacle Course

Atlas robot jumps over a gap

Historically, the capabilities of real world humanoid robots have trailed far behind their TV and movie counterparts. But roboticists kept pushing state of the art forward, and Boston Dynamics just shared a progress report: their research platform Atlas can now complete a two-robot parkour routine.

Watching the minute-long routine on YouTube (embedded after the break) shows movements more demanding than their dance to the song “Do You Love Me?  And according to Boston Dynamics, this new capability is actually even more impressive than it looks. Unlike earlier demonstrations, this routine used fewer preprogrammed motions that made up earlier dance performances. Atlas now makes more use of its onboard sensors to perceive its environment, and more of its onboard computing power to decide how to best move through the world on a case-by-case basis. It also needed to string individual actions together in a continuous sequence, something it had trouble doing earlier.

Such advances are hard to tell from a robot demonstration video, which are frequently edited and curated to show highlighted success and skip all the (many, many) fails along the way. Certainly Boston Dynamics did so themselves before, but this time it is accompanied by almost six minutes worth of behind-the-scenes footage. (Also after the break.) We see the robot stumbling as it learned, and the humans working to put them back on their feet.

Humanoid robot evolution has not always gone smoothly (sometimes entertainingly so) but Atlas is leaps and bounds over its predecessors like Honda Asimo. Such research finds its way to less humanoid looking robots like the Stretch. And who knows, maybe one day real robots will be like their TV and movie counterparts that have, for so long, been played by humans inside costumes.

[via Ars Technica]

35 thoughts on “Boston Dynamics Atlas Dynamic Duo Tackles Obstacle Course

        1. Now that’s an interesting thought. Maybe it could be possible, similar to a slip ring in electronics.

          Or some form of symbiosis, where the “wheel” is a loosely coupled organ, which cells are too, after all. Energy exchange could still work through chemicals/osmosis on the periphery.

          Animals/humans do have joints, they don’t rotate 360° but there are sill parts that are not nurished by blood vessels, only by the fluid they are contained in.

          So it should be possible. Bigger issue would be that nerves wouldn’t work across such joints, so you would need maybe different kind of communication, that is slower, as in chemical or electric fields.

          All of this is not impossible, but I assume the lack of evolutionary pressure prevents development of such a significantly different (even if based on already used principles) solutions.

    1. Yes, it is utterly terrifying that it took a lab full of people to pre-program a series of steps into a computer that directs a stupid pile of hardware to appear to run parcour.

      All that robot does is to follow instructions where to step in what order. The only mildly interesting part (technology wise) is that it balances itself to adjust for uneven footing.

      All the rest is pre-programmed. It isn’t as though they’ve built a parcour runner that can look at a bunch of obstacles, compute a path through them to a goal, then adapt to changing conditions as it traverses the course and discovers things that weren’t obvious from the starting point.

      It is a demonstration. A carefully staged demonstration.

      Outside of the lab, with no programmers or keepers, it would stand there stupidly until its batteries died and it collapsed like the expensive pile of spare parts that it is.


      For all of that, it is impressive and cool to watch.

      1. And the cheat by keeping the hips and knee joints in a permanently bent position. Humans try to straighten their legs for minimal energy expenditure, at the cost a much reduced range of balancing motions.

      2. I think Boston Dynamics has done enough to show the realistic expectations of their robots. This is far better than the console gaming companies that would show up to E3 with ‘in game’ footage that is actually running on high end PCs under the table.

        They have never hidden the fact that these lab tests are just that, lab tests. They are an R&D company.

  1. As someone that used to practice “the martial art of rubbing away”… I was largely unimpressed until it backflipped.

    Admittedly, stakes are a lot lower than putting a human neck on the line, but flips are spicy!

    1. Flips are easier for the robot than for humans, because they can be pre-computed and repeated exactly the same every time. There are no unpredictable external forces that would apply while the robot is in the air – only reaction forces caused by its own motions – so it doesn’t have to do anything smart to pull off a flip. Just land with its feet down.

      1. lol, what do you mean by easier? Seems like trying to compare apples and oranges. Well, more like comparing apples and motorcycles. I can safely say I will never backflip with only a 2′ drop, so I guess yes, it would be easier for me to make a robot backflip that doing a backflip myself.

        Robots aren’t just computer programs that run the same way every time. Hell, even a lot of modern programs don’t run the same way twice due to all their external influence. In the real world there are a ton of external factors that these robots need to see and feel, and take into account.

        The first robot I ever made was a super simple open loop rover. Go forward for 2 seconds. Turn right for 1 second. Go forward for 2 seconds. Etc etc. I expected it to make a nice little square indefinitely. Didn’t take long to realize the 1 second turns didn’t always give a 90 degree angle. Different friction at different parts of the floor lead to different amount of angle achieved. As the battery died, max RPM lowered and also reduce the angle.

        Look at HD-01’s left foot at 0:37 when jumping up on the box. Something sticks and it does not move as intended. Instead of flopping over and ending the run, the sub routines kick in and it saves itself. As grip level changes, temperatures change, battery state of charge changes, hardware sticks or moves slower than new, everything needs to change. There is no such thing as ‘exactly the same every time’.

        1. While the robot is in the air, it’s in a well-defined state: it’s a free falling object. It has far fewer unknown variables than when it’s walking or just standing – it’s just hanging on for the ride. A simple mechanical toy can make a back-flip – that in itself is not impressive.

    1. I thought the same thing. It seems unlikely to me that this frame, out of the entire video, was chosen at random that just *happens* to show what looks like the robot hand-traversing that beam. :-)

      A good video nonetheless. :-)

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.