roninfiertze wrote:Wow, they really need to work on the lip syncing on those. Looks like she is chewing bubble gum while making a constant noise which makes zero sense.
I'm more on the code side rather than hardware, so if this is caused by loose tolerances or sticky joints in the hardware then I have no idea.
Lots of servo chop is apparent that can likely be fixed in code... not sure if their program loop is excessively long or hanging on a piece of code but if they added a better smoothing algorithm and cut their loop down to a few milliseconds they'd have a lot of that fixed, regardless of the servo or controller brand. Even dirt cheap $1 servos can be run smooth. I don't know what controller they are running on but I have made even a 16MHz Arduino Nano smooth on multiple simultaneous servos without it lagging. That's on hacked together 3d printed parts driving linear gears etc. and I would hope their hardware is far better designed.
Addressable I2C serial controllers on a separate circuit to the common ground are a godsend. I usually overbuild with ~20v batteries and then dedicate a buck converter for the Arduino and another for the serial servo board so you never end up with the entire system being drawn down for some reason.
A lot of servo noise can also be a result of not fine-tuning PWM frequencies for the exact servo model being run. Every different model of servo I have worked with needs to be tuned properly or they get loud and choppy like this.
Just my 2 cents as a completely unprofessional robot builder and completely professional coder.
Getting el cheapo servos to run smooth is actually easier. The internal, often analog and proportional-only, control loop will respond poorly to minute position increments and take some time to center on a final position. The poor signal resolution of updating at, say, 10 Hz (that's about how fast I got my arduino nano to do a dozen servos, but I'm sure you can do better) is masked by the even poorer control resolution. All of that is topped off with a sloppy gear train, further "enhancing" the effect.
In one of their earlier vids you can see the label on a servo they were using at the time. I looked it up, and it was a high end brushless one. I think it was about $150. Something like that is going to respond very quickly to any quirky signals given by the controller.
No idea why they would have problems with their controller keeping up. I would love to comment on the "right way" to do such a thing but that's well outside my skill set
Honestly, if I had that kind of budget I would be ditching the RC servos altogether and looking into "quasi-direct drive" (robotics buzzword for low gear ratio) BLDC drives. Getting rid of the ~200:1 gearbox in RC servos yields much smoother motion and also eliminates the biggest source of noise. Having direct access to the motor controller is beneficial as well, RC servo PWM input is inherently low-bandwidth and modern micros are probably happier pumping out serial data.
These kind of drives are already being used in Engineered Arts "mesmer" animatronic system, and they work very well. I imagine Realdoll avoided them at least for now due to cost (~$100/actuator is bottom-end pricing for such a setup), physical size and weight, and development costs. Perhaps they're working on a new drive system right now, in which case hats off to them
PS: I don't mean to dunk on the Realdoll team too hard, but I mean check out this Elvis head:
https://www.youtube.com/watch?v=hw74noM5I_8
Released in 2007, MSRP was originally around $300, retailed much lower. I would have hoped after 15 years almost and for ~30 times the price, you would get way better animation. I'm sure a lot of it is just down to economics of scale, but I wanted robot girls by now, dammit!
