DARwin AX
by Bullit
Wed Mar 23, 2011 6:40 pm
Thanks SJ for those clarifications.
It seems to me that the rx-28m holds its position more finely then the other Robotis servos. When I think about this, it makes sense given the precision of the potentiometer and the fact its read by a 10bit A2D on the ATMega8 in the servo. One can imagine there is no way that the Robotis potentiometer servos achieve true 0.29 degree precision. The rx-28m's on the other hand use a 12bit magnetic encoder and a digital SPI interface to the sensor therefore the output is over-sampled and we can imagine that it achieves the 0.29 degree precision.
The Robotis ex-106+'s have an interesting arrangement of potentiometer on the final output and a magnetic encoder on the motor. The magnetic encoder there is likely only quadrature of the motor and gives some finer resolution between steps from the potentiometer. This solution has some problems in that the motor position can be finely controlled but the output includes gearing backlash. I have found the ex-106+'s have the greatest backlash of the Robotis servos. Robotis also struggled quite a bit getting calibration of the ex's to work well as one can imagine given the arrangement of potentiometer and motor encoder.
If you look at the specs for the Murata potentiometers used by Robotis in the servos its easy to understand why they wear out and why they are noisy.
It seems to me that the rx-28m holds its position more finely then the other Robotis servos. When I think about this, it makes sense given the precision of the potentiometer and the fact its read by a 10bit A2D on the ATMega8 in the servo. One can imagine there is no way that the Robotis potentiometer servos achieve true 0.29 degree precision. The rx-28m's on the other hand use a 12bit magnetic encoder and a digital SPI interface to the sensor therefore the output is over-sampled and we can imagine that it achieves the 0.29 degree precision.
The Robotis ex-106+'s have an interesting arrangement of potentiometer on the final output and a magnetic encoder on the motor. The magnetic encoder there is likely only quadrature of the motor and gives some finer resolution between steps from the potentiometer. This solution has some problems in that the motor position can be finely controlled but the output includes gearing backlash. I have found the ex-106+'s have the greatest backlash of the Robotis servos. Robotis also struggled quite a bit getting calibration of the ex's to work well as one can imagine given the arrangement of potentiometer and motor encoder.
If you look at the specs for the Murata potentiometers used by Robotis in the servos its easy to understand why they wear out and why they are noisy.
Last edited by Bullit on Thu Mar 24, 2011 9:43 am, edited 1 time in total.
by i-Bot
Wed Apr 13, 2011 11:48 am
A couple of intersting updates are on sourceforge.
The actual actual firmware for the CM-730 is now there.
Also Robotis now made the step from 1024 to 4096 resolution on the RX28-M servos. Will be good to here from Bullit if this makes a noticable difference to perfomance.
Are Robotis or Bullit taking the DARwin to Robogames ?
The actual actual firmware for the CM-730 is now there.
Also Robotis now made the step from 1024 to 4096 resolution on the RX28-M servos. Will be good to here from Bullit if this makes a noticable difference to perfomance.
Are Robotis or Bullit taking the DARwin to Robogames ?
by Bullit
Mon Apr 25, 2011 12:55 pm
I've now updated my Darwin to the new 4096 firmware. The new firmware changes the working angle of RX-28M's from 300 degrees to 360 degrees so where the number space used to be 1024 steps over 300 degrees it is now 4096 steps over 360 degrees. The other big change in the RX-28M firmware is the change from CW/CCW slope to PID parameters. Thus far the Darwin code only has default P with no I or D terms used. This change seemed to impact a couple of my walking parameters a bit and required some re-tunning of the holonomic gait to get it back where I had it previously.
Overall after re-tuning I think the changes in performance are very subtle. I think the one thing I noticed being different is the range at which the gyro gains where useful. I'm not really sure why this is yet but it seems I can push the gyro gains even higher than before without getting oscillation. For instance with the new firmware / software I have been able to get my Darwin to walk quite well on soft carpet which required significant gyro gain given the surface variability. These settings although they worked well on soft carpet did not work that well on short carpet or a hard surface. Darwin would still walk on the hard surface but the settling time made the walk seem a bit odd as he over stabilized planting each footstep.
Overall after re-tuning I think the changes in performance are very subtle. I think the one thing I noticed being different is the range at which the gyro gains where useful. I'm not really sure why this is yet but it seems I can push the gyro gains even higher than before without getting oscillation. For instance with the new firmware / software I have been able to get my Darwin to walk quite well on soft carpet which required significant gyro gain given the surface variability. These settings although they worked well on soft carpet did not work that well on short carpet or a hard surface. Darwin would still walk on the hard surface but the settling time made the walk seem a bit odd as he over stabilized planting each footstep.