XRL Hexapod Robot Gets a Tail, Learns To Use It

New submitter toygeek writes "In an effort to give various robots more control during free-fall and navigation of severe obstacles, researchers have studied how agama lizards use their tails to retain or correct orientation during leaps and jumps. They've applied the research to both hexapod and wheeled robots, and the results are both astounding, and outstanding!"

Re:Gee, you'd think a cat would be better

[techno-vampire]

Not only that, a cat uses its tail to turn over in mid-air so that even if you drop it with its feet pointing up it will land feet first as long as it has enough falling time to make the turn. I'm not sure, but I think I've heard of them getting turned in as little as 1.5 feet, but ICBW. Still it would be a great improvement over what they've got so far.

Re:Gee, you'd think a cat would be better

[Paul Fernhout]

I just watched the related videos, and it is impressive just what an up and down tail can do. But you're right that a tail that can move side-to-side, or that can curl or corkscrew can probably do even more to maintain a desired 3D orientation. No doubt that will be future directions for this work. So, in the real world, it seems like a common situation that you want to maintain the orientation of 90% of the mass of an object, and are willing to sacrifice the 3D orientation of 10% for a short time to do that (until the tail orientation can be reset at no body-orientation cost when the creature is on a surface again). Brilliant. These robot videos really show how amazing nature can be. I really appreciate tails in a way I never had before watching those videos. I especially found of interest video about a predecessor robot to this which includes video footage of the long-tailed lizard used as a prototype:
http://spectrum.ieee.org/automaton/robotics/diy/dinosaurlike-tails-make-terrestrial-mobile-robots-more-agile [ieee.org]

Been there, done that.

[Animats]

Worked that out back in 1995. See this the "running on rough terrain video. [animats.com] Watch at 1:40.

The next step, which the U. Penn people don't seem to have taken yet, is to solve this as a two-point boundary value problem. Then, instead of trying to maintain attitude during flight, you try to land at a specific time in a specific place with a specific attitude.

Doing that is rocket science. Rocket science is about control of underactuated systems, where the control system has fewer actuators than there are degrees of freedom to be controlled. You want to reach a specified point at a specified time at a specified velocity in a specified attitude. All you have available is the ability to thrust in one direction and to change attitude slowly. But this problem is solveable and there are known solutions. Applying that to robots leads to gymnastics.

The quadrotor people are already doing this kind of thing, but it hasn't been done much for legged machines yet.