Hacker News Reader

Comment by danbmil99 | Hacker News Reader

This is really core research, not product development. BD and Darpa spin it a bit to keep the funding going, but having worked in this field my guess is we are at least 15-20 years away from serious deployment of legged robotics in the field.

That said, it's amazingly impressive what they have achieved.

I am curious as to why the robot's front legs are so close together in some clips -- maybe approximating a 3-legged device was somehow algorithmically simpler?

View on HN

Quadrupedal animals have multiple gaits; at higher speeds, they'll start bringing the forelegs down together at closer points in time.

Check out, for instance, this video of an actual cheetah running around: http://www.youtube.com/watch?v=KIeXEiJuJUY - you can see the front paws coming down very close to each other in time and space, followed by the wide-spread rear paws making huge leaps forwards.

Also a quick images search for "horse muybridge" may be enlightening - Muybridge's books are super well-known in the animation circles I come from, as they were the first major works of research into high-speed photography documenting How Stuff Moves. You can see that horses have a whole bunch of different gaits available to them; most other animals are similar, but less tractable to a photographer's whims!

  > why the robot's front legs are so close together

Hypothesis: The amount of force required to effect motion changes when you form a small, tightly focused fulcrum, using two legs.

I'm thinking there's some kind of efficiency gained, where, to us, it looks precarious and awkward, but to the machine-calculated algorithms, a trend is detected, where it's easier to stay continuously balanced on a small point while in motion, because it can use it's own inertia and apply smaller amounts of torque and pressure to it's actuators, and use smaller movements, when attempting to stay on it's feet.

When it's at rest, a wide stance is probably safest, but in motion, maybe it's a different story. Given that it's a somewhat rigid machine, with appendages that have alimited range of motion, maybe it targets, the smallest most effective movements?

I would also wonder: What is the net energy consumed from it's power source, to cycle one limb, moving it from fully contracted to fully extended and back again? You would need to test this unloaded in free air, and completely loaded under the weight of the entire robot at rest. And that test would not account for the amount of work it would take to safely absorb the full weight ofthe robot, while it's moving at 20MPH, to bring it to a full stop, with a single limb.

I'm not sure if those sorts of efficiencies are anything more than a simple practical concern in a prototype, given that the only goal is to run some quick, untethered tests in a parking lot, but it might be relevant to a small degree.

This is just my amateur guesswork, though.

I see a lot of four legged animals run like that in assorted nature docs.

I agree that it's core research, but it isn't published anywhere. Their control algorithms are not public; even plenary talks from Mark Raibert don't go into the details.

Excellent point. As I understand it, Darpa does not preclude recipients of funding from patenting their work, nor do they insist on publication. I believe the govt gets some sort of free lunch on the patent side -- they can use anything they want with some sort of IP override (if someone knows more about this please explain).

The way the govt avoids vendor lock-in (in principle anyway) is by farming out the money to multiple vendors, especially as the progress moves towards real practical deployment. They did that with autonomous vehicles, and they're doing it again with legged robotics (http://www.darpa.mil/our_work/tto/programs/darpa_robotics_ch...)

However in this one area (legged locomotion) it seems to me that BD is far, far ahead of anyone else, unless that work is taking place in complete secrecy.

Shall we not forget the MIT cheetah: http://biomimetics.mit.edu:8100/wordpress/

I like the design much better than BD's weighty quadrupeds.

Speculating, one possible reason would be that it can then make balance corrections by pushing (making the leg longer). A wider stance would require making balance corrections by making the leg shorter.

It's what you generally see in nature too when (catlike) animals are bounding: http://cincinnatizoo.org/wp-content/uploads/2012/07/sarah_ru.... Front paws go between the rear ones, etc. Rear ones are used for the brunt of propulsion, front ones for balance and direction etc.

Reply on news.ycombinator.com