The results of the last test (and blog post) were clear, the SEA design is not working as I had hoped. I maintain my believe that the compliancy in the SEA design is needed (superior) to a static approach. However, this current design will not work with the relatively weak pull of the 9g servos. At least not in the current configuration.
In my desire to move the prototype along, I have considered some other approaches. In my Mojo3 #6 blog post, I was considering using a hinged approach similar to a 4 bar mechanism. With such an approach this would allow for the servo to directly lift the 'arm' off the ground. This would provide the movement needed (ability to lift the foot when moving the leg forward), however, a different thought came to mind as I was adding the servo gear to the hip joint.
Why not use a servo gear to raise the 'arm'?
|Mojo3 - Robot Dog - Alternative leg approach (OpenSCAD)|
The gear driven servo would provide small additional torque (without the servo horn) and less backlash. It will also be significantly easier to tune, being able to reset the location of the gear. Also, the Inverse Kinematics would be much easier to calculate. The downside is that I would lose the compliancy in the lifting mechanism. Compliancy would need to be moved to another location on the leg. (perhaps the 'arm' or foot). Note, the above picture uses "shorty" arm, just a very quick sketch to see if the concept was plausible.
|Mojo3 - Old and New leg designs - "shorty" on the right side.|
Testing the new design
|Mojo3 Robot Dog - Test for Kinematic path|
The new design proved to work well. It was easy to adjust and tune. Most important, I was able to get around 12mm of upward movement in the Kinematics. With experimentation, I learned that increasing the delay or slowing down the motion, provided a very different Kinematic path. The servos had more time to get to their set points.
|Mojo3 - impact of timing differences in the Kinematics|
Putting all this design work together
I am pleased with the current design. This approach will be faster to get to a walking quadruped. The gear design will allow the robot to be adjusted much easier. I will need to have some compliant mechanism added to it in the future. The new robot frame should look something like this:
|Mojo3 - A Quadruped Robot Dog (OpenSCAD - exploded component view)|