This design uses a brushless motor, geared down with a belt and 3D printed gearbox. I wanted to maximise the power output of the joint so that it was strong, fast, accurate and just as important, good looking. After a few years of different drive train options including a series of belt reductions and worm gears, I settled on a planetary gearbox and belt reduction.
The belt provides a 100tooth/15tooth = 6.66...:1 reduction followed by a 5.99:1 reduction from the gearbox yields 39.9:1 reduction. The joint is surprisingly backdrive-able, requiring only 0.825 Nm (300grams on a 275mm arm) of torque to backdrive the unpowered joint.
Another requirement of the design was that the components that make up the knee joint should be the same that make up the pitch axis of the hip joint. This means that all the same G-code can be used to print the parts for the other joint without the need to mirror or redesign any parts.
This update is very late as I finished printing and building the hip joint towards the end of July last year just as I started my first semester of uni and got a bit preoccupied. I am very happy with the way the parts went together and work. Everything feels very sturdy and the joints are both smooth.
I have been tweaking the design over the past month as to improve some aspects I wasn't happy with. I also added some functionality.
- Added AS5048A magnetic encoder to the joint to give absolute position feedback from the joint. This will allow the robot to know the positions of all the joints on startup without homing. (This was always intended to be in the design I only just got around to adding it)
- Swapped all the gear teeth from straight cut to helical cut. I should have started with helical gears as changing them this late in the design messed with my Fusion 360 timeline and I had to do some sketchy stuff to keep the design as parametric as possible. It was well worth it as it reduced the backlash from about +- 3mm at the foot to absolutely nothing and made the gearboxes significantly more smooth with no stiff spots. This came at the small cost of increasing the force required to back-drive the joint but I think its worth it. (Having no backlash makes programming much easier)
- Swapped out the AMT 102-V encoder for an AS5047P magnetic encoder. I was having trouble getting the alignment of the AMT encoder right because I was mounting it upside down to maximise the length of the motor shaft I could get into the collar. This was causing rubbing inside the sensor so I swapped it for a magnetic encoder which was much easier to mount and there is no chance of rubbing.
- Found some new TPU filament for the foot and knee pad which as a bit softer and actually grippy.
I am now pretty happy with the design of the joint so I will begin printing all the parts for the hip joint and should hopefully have it done soon :)