Why do bipedal robots have crouching gait?

Bipedal robots such as Atlas in this youtube clip https://www.youtube.com/watch?v=rVlhMGQgDkY always seem to adopt a somewhat comical crouching gait where they never straighten their knees. Compare this to the human at 55s into the above video. Why the difference between man and machine?

Humans are poorly designed for an upright stance. Most other bipeds (which is pretty much birds) keep their knees flexed.

That robot can put up with just so much… After getting pushed and shoved around, It finally walks off the job.

Not really. Sure, the spine and other bits of the skeleton have been awkwardly adapter from our quadrupedal ancestors, but the legs have ended up as pretty good ways to get around. Humans can walk long distances very efficiently; perhaps not as efficient as a lot of quadrupeds but at least as good as any other biped. Humans are also some of the best and most efficient medium-distance runners (however it’s worth noting that without modern running shoes, most people will run on the balls of their feet and use their ankles as “reversed joints” like other bipeds).

My WAG is that walking like a human, essentially teetering on inverted pendulums without actually falling, is very difficult to control. One of the advantages of our form of locomotion, where we use very low muscle force, is also less important for these early generations of robots where it’s easier to use compact and powerful motors.

Improved stability by lowering the center of gravity.

Humans don’t straighten our knees when walking either. See this video, for example. (I’m not sure if that’s meant to be anatomically accurate, but it looks realistic to me.)

Our muscles need to expend energy to hold position even when not moving. Locking the leg straight allows it to bear weight without expending much energy, since the load is being carried straight through the bones and the muscles of the leg aren’t doing much. It takes much less energy to stand with your legs straight than to crouch.

These robots have hydraulic actuators that can hold a position indefinitely without expending energy, so it’s no different to stand crouched than with the legs straight. And making fine adjustments for balance is easier when the legs are slightly bent.

The robot seems to completely lack foot and ankle dynamic compared to humans.

Here the sequence of events that occur at the foot and ankle level in a human:

  • during stance, the foot lays flat on the ground while the other foot lifts off, swings forward, etc.

  • in late stance, the heel lifts off the ground and the foot lays on the digits. At this moment, the leg starts to bend (knee flexion). This mechanism allows to initiate the swing period of the cycle with minimal energy consumption. Indeed, the configuration of the joints is such that the heel will rise and the will knee flex when you move the pelvis forward, and this requires very little muscle contraction. This movement starts to raise the foot and stores angular momentum in the flexing leg.

  • the foot continues raising and the leg flexing by inertia, leading to foot lift-off.

  • the knee remains flexed as the leg swings forward as a pendulum.

  • the knee extends (a bit) in late swing.

  • and there is another nice mechanism: the heel rocker. The heel hits the ground first and then the entire foot rotates in order to go lay flat on the ground. Because the ankle is a couple of inches forward relative to the heel, this movement pulls the ankle forward. This forward motion is transmitted by the leg to the pelvis and gets re-used as forward momentum.

The dynamic of the foot and ankle mean that:

  • the leg can be lifted (in order to swing) by using the momentum of the entire body rather than muscle contraction.
  • the shock of the leg hitting the ground at touch-down is converted into forward momentum rather than being wasted.
    This makes the leg apparatus extremely efficient and allows us to walk for miles and miles on a firm ground without tiring.
    Note also that we get a chance to push the body forward in late stance, after heel lift-off, by contracting Achilles tendon.

When you look at this robot (and most bipedal robots), you will see that the foot remains completely horizontal at all times. All the mechanisms I just described are omitted for simplicity (it is hard enough to design robots that walk safely). This contributed a lot to the “alien” look of these robot’s gait.

In general, the knee does flex in mid-swing and extend in late swing; but the amplitude is moderate (30-40 degrees).

See here for instance