Actuator Mechanical Design (v1)
Status: Current Project (Jun 2020)
Brushless Motor Selection
Knowing the torque requirements from the basic calculations made in Robot Dog Mechatronics, I could start looking for a suitable brushless motor to base the actuator around. Since torque per amp is roughly proportional to motor radius squared, I looked online for the largest diameter motor available.
The largest motors I could find were tiny 6007 motors (stator of 60mm diameter, 7mm tall). After holding it in my hand and realising how tiny it was for its very high price, I looked at what motor Ben Katz used in his paper [1] to get a rough estimate. On page 24, figure 2-1 shows the final actuator. Firstly by matching the rotor shape, rough dimensions and the font used, I narrowed it down to the "eX8108 105KV" motor [2]. The site also confirmed that they supplied the motors for the infamous back-flipping Cheetah Mini, confirming my internet detective skills.
Now armed with the stator dimensions and KV rating, I started looking on Alibaba for motors since I would need a much larger motor than a 8108.
TMotor 6007 BLDC motor.
Large BLDC motor options. Torque is indirectly proportional to KV, directly proportional to diameter^2 and stator height.
Relative torque measured against the 6007 motor.
The ideal motor would have a low KV, large diameter and large height as well as not being too heavy and not cost too much.
The Turnigy motors were very appealing however I did not trust the stated weights as well as being impossible to find. In addition, they use the 9225 to mean a rotor diameter of 92mm instead of the stator which confused me for a while.
In the end, the most ideal motor was the generic 8313 100KV. They are widely available from Chinese sellers.
Standard BLDC stator sizes from Chinese manufacturers? From [3].
One interesting thing I discovered talking to a seller on Alibaba was that there appears to be standard sizes of stator. The OD, ID and Number of slots is fixed due to the stamping moulds however the height can be easily customised.
The table of stamping moulds can be found on [3].
Ballin motors.
Gearbox Design
For robots who will interact in environments with humans, having the ability to backdrive is very important. This allows the robot to feel external forces instead of being completely rigid, lowering impact forces and giving the robot a sense of force.
This essentially rules out SEA (series elastic actuators) since they are non-backdrivable without advanced force sensors which only work with power. High gear ratios (>10:1) also make backdriving harder as well as increasing the reflected inertia. Too high of a gear ratio also requires the input shaft (motor) to spin very fast, wasting energy accelerating and decelerating.
Planetary Gearbox
Following Ben Katz [1], I attempted to design a 9:1 planetary gearbox that could fit inside the stator ID. Unfortunately numerous problems came up.
In order to fit the gears inside, a very small module (essentially tooth size) was required, necessitating use of very tiny endmills to cut the radii in-between the teeth. I could manufacture the gears with a gear hobbing setup however it would be too expensive. It also does not allow manufacture of internal gears, necessitating even more custom tooling.
Looking for gears online also did not yield many useful results. No existing gears existed that would work and custom gears would cost too much. In addition, the small tooth size required use of hardened steel to handle the torque without snapping off, further increasing cost.
Cycloid Gearbox
One seemingly overlooked gearbox solution is the cycloid gearbox. Its major benefits over planetary gearboxes are:
Large contact area since half of all the lobes are in contact.
Easily customised gearbox ratio.
Relatively easy to manufacture just with a CNC router.
Unfortunately it also comes with some disadvantages:
Some designs may not be backdrivable.
Many parts that require tight tolerances.
Complex geometry.
My major inspiration for moving to cycloids was looking at Paul Gould's work [5].
To create the geometry, the parametric equations were generated using this website [6].
Unfortunately my first cycloid discs were not backdrivable and I had to redesign the geometry which can be seen below.
How a cycloid gearbox operates [4]. Green is input shaft. Purple is output shaft. Gear ratio is the number of lobes on the yellow cycloid disc.
First cycloid design.
Left is new geometry, right is old.
Fresh new cycloid discs.
Actuator top with driver PCB and heatsink.
Assembled actuator.
Assembled actuator.