Thanks @cbenson! What’s the strongest lynxmotion servo you have? The HT1?
Maybe the robot would need to be a bit smaller. But I am up for a redesign as a joint project. You supply HW and I can build a lynxmotion version (hardware based). The software side I will use ROS and controller package CHAMP.
Of course it will be branded with lynxmotion on all social media etc.
If you are interested just let me know.
For the color scheme and I haven’t quite decided yet on which color it will have. But in any case All plastics parts will be processed and painted buy a painter!
That’s also great news. Do you have any experience in ROS? You might want to get in touch with @cmackenzie who is already quite well versed in ROS 2.0 and is working with us for the SES V2 humanoid: https://www.youtube.com/watch?v=7iLNI5IvHGM (more on his channel)
I’ll run it by the team to see if we can proceed.
Yeah I normally run ROS for my project. All videos above is running ROS with various packages. I run ROS Melodic at the moment, haven’t take the step to ROS2 yet.
Also starting to think about alternatives solutions for the joint. Which is not servos.
Super fast and rough sketch of a joint with two t-motor RI50 which should run in sync to drive a central (lower leg) quasi direct drive with up to 10:1 in gearing. To reduce the width even further I am thinking of ILM 50x14 frameless motor.
With this solution I could get quite good numbers around 12nm and 32nm peak torque:-)
Since you’re effectively 3D printing all parts, their complexity doesn’t quite matter, though plastic gears at that torque might be problematic. If you opt for two frameless motors like the RL50, why not choose one like that the RL100 which provides more than double the torque and would make for a thinner design? Understand the symmetry and having the gears at the center of the leg segment though, and having two does create some redundancy, though there’s also the chance they might fight each other if not properly aligned?
The other “catch” seems to be the high current draw, at up to 14A per actuator peak, so multiply that by 24 actuators if the motion requires it. A LiPo can do it, but a 13S which can provide 168A peak with enough capacity… curious to see the weight.
hi! this design is just…impressive!! it’s been a while since I started playing with reinforcement learning and quadrupeds and I was looking for an open source reliable design like this are you planning to open source the URDF? I’d more than happy to play with it with my library! (https://github.com/nicrusso7/rex-gym)
Thanks for the good inputs @cbenson!
Here is some answers and thoughts.
The RL100 would be quite big to direct mount on the knees, that is why I am looking on the RL50.
The RL100 would also require a totally different design.
The max current at 14A/motor would maybe be a problem. But the main reason I am looking for this peak torque provided by two RL50 is to be capable of jumping. Which would require 2 or 3 x torque of the body weight.
I am thinking of buying the gears in metal to be able to handle the forces. Which maybe not 3dprinting would be able to handle. Even tho I have some good experience with printing gears in alloy910. (Could work for a concept at least).
Yeah alignment would need to be correct!
I think a bit more about it. But I like the form factor of the RL50 which would make a quite nice compact joint.
Hey @NicRusso. Thanks! At the moment it is closed source. But send me a Skype at : raess1
And we can discuss it. I could probably share it if you want to help and test the RL lib on the Urdf. But it would still remain closed for others in the current state.
Btw what factors does your lib take in to considerations? Current and torque?
Haven’t calculated that yet since I don’t have so much current data while it on it’s legs.
I tested a bit load on front legs with body weight, while holding the back by hand. Highest individual servo load was 250mAh then. But we will see
Update:
Hip -> leg bearings in place! Works perfect!
bearings arrived today! Printing bottom cover with battery placement!
are you planning to open source the URDF? I’d more than happy to play with it with my library! (
