YouTube seems to be back on the air, so I’ve FINALLY managed to upload the clips from the last wk-10 days.

With the uploading holdup (the Google video site doesn’t seem quite up to the standard of YouTube, BTW) I haven’t managed to make much more progress than the "disaster" in vid #7.

As I propbably mentioned elsewhere, dealing with extended surfaces on which a bot balances now seems beyond the previous trick of selecting a few points (e.g. the corners) on the surfaces that are expected to be in contact with the floor, and then calling those points "feet".

This was the trick that worked with all prev simulated bots, but is not accurate enough (modulo various numerical techniques used to solve the equations of motion) to have the bot balance.

Vid #7 shows a first run of the current animation s/w on the Humanoid. It’s a bit facical — the s/w can’t always tell whether the bot is balanced or not.

The major discontinuities where the bot seems to fall on its face and then become upright again within a couple of frames (30 fps at present) are due to a "recovery" module I put in to right a bot even when it (simulated) fell over.

But the "righting" method (also)seems crazy when viewed at the current frame rate.

Previously I was animating at 10 fps, but this has been upped to 30 so the movements of the simulated Robotis AX12+ servos will relate to real time and not 1/4 or 1/2 of real time.

Anyway, some special-purpose code for extended surfaces in contact will now have to go in. And maybe the numerical methods used to solve the motion will also need tweeking to make things more relaistic for this trickiest of robots.

---

On another front, I’ve been tinkering with some tactics for humanoid "kung fu".

Looking at vid of some prev comps it seemed that a bot that stayed upright — regardless of whether it landed any blow on an opponent — was already 1/2 way to winning the comp.

So one of the tricks I was planning in the "fighting" version of the Humanoid was a special "tetrahedral" stance. I.e. arms and legs would point to the vertices of a tetrahedron, thereby making the bot 100% stable from any attack.

With a flick of a couple of limbs, the bot can even roll around in an uncontrolled way.

This would then be a primary attack metehod — a stance that could topple an opponent yet not topple the humanoid.

And — of course — using the continuous-rotating Robotis servos can also be an advantage. E.g. windmilling of arms, or a 720 deg roundhouse kick are pretty easy of a bot.

I don’t know whether typical contest rules allow for move that are impossible for a human.