Thanks for the great write-up. I really liked the clear breakdown of how the optimal transmission ratio shifts between stance and swing across motor sizes and ambulation speeds. The way you explained it with such simple, intuitive plots was really elegant. We see a very similar tradeoff in robotic prosthetics as well.
To me, this is a strong argument for variable transmissions: if the best transmission ratio changes so much with task and speed, then varying the ratio over time may be better than committing to a single fixed design point.
We explored something along those lines in a recent paper on a knee prosthesis with torque-sensitive actuation, where the transmission ratio varies continuously with output torque. When applying high torques, like during stance or running, the ratio increases, while at low torques, like swing or slow walking, it decreases. That lets the actuator operate much closer to the “optimal” transmission ratios you showed in your plots.
I thought this might be relevant to the actuator design problem you’re mentioning here:
Thanks for sharing, that's an interesting design and I can tell how complex it must have been to design and control. I had some experience with similar intuition-based continuously adapting transmission ratios in this project (paper linked from post: https://www.avikde.me/i/182198523/nonlinear-transmission-design). However, that was a much simpler activity (flapping) with much less unpredictable environmental interaction.
Holy crap, that engineering team is immense. Not to bash our beloved field of robotics, but even the 747 plane had 4,500 engineers (source: https://www.smithsonianmag.com/air-space-magazine/a-amp-s-interview-joe-sutter-14734609/). I feel like aerospace has more (safety) requirements and tech specs to hit than we do for tasks like running. I wonder what the breakdown of what each engineer did in Honor looked like.
On another note, I've been itching to going back to building robots. In my naive past I've usually just kinda guesstimated/back-of-the-napkin'd the specs I'd need for the actuators given a specific task, like walking. But I'm curious, in your experience what's the best design approach? Some people I've seen (and what I'm looking to try) is to guesstimate/assume your motors are strong enough for most basic tasks and then CAD -> Sim -> Refine -> Sim -> Build Hardware.
Here's a basic recipe for motor sizing to build robots:
- Start with a catalog (e.g. the TQ one referenced)
- Follow the method in this article to size the gearing for a few motors
- Get a sense of the total power for the best gearing for each motor. Typically larger motors will be better, but they will of course weigh more. Use judgment to trade off motor weight vs. total power
This is only the first pass, you'll get secondary constraints like voltage, if that gear ratio is attainable using a gearing technology you can use, etc. If this topic is interesting, I can try to write more about it in the future, but this should be a starting point!
Thanks for the great write-up. I really liked the clear breakdown of how the optimal transmission ratio shifts between stance and swing across motor sizes and ambulation speeds. The way you explained it with such simple, intuitive plots was really elegant. We see a very similar tradeoff in robotic prosthetics as well.
To me, this is a strong argument for variable transmissions: if the best transmission ratio changes so much with task and speed, then varying the ratio over time may be better than committing to a single fixed design point.
We explored something along those lines in a recent paper on a knee prosthesis with torque-sensitive actuation, where the transmission ratio varies continuously with output torque. When applying high torques, like during stance or running, the ratio increases, while at low torques, like swing or slow walking, it decreases. That lets the actuator operate much closer to the “optimal” transmission ratios you showed in your plots.
I thought this might be relevant to the actuator design problem you’re mentioning here:
https://journals.sagepub.com/doi/10.1177/02783649251383018
Thanks for sharing, that's an interesting design and I can tell how complex it must have been to design and control. I had some experience with similar intuition-based continuously adapting transmission ratios in this project (paper linked from post: https://www.avikde.me/i/182198523/nonlinear-transmission-design). However, that was a much simpler activity (flapping) with much less unpredictable environmental interaction.
Holy crap, that engineering team is immense. Not to bash our beloved field of robotics, but even the 747 plane had 4,500 engineers (source: https://www.smithsonianmag.com/air-space-magazine/a-amp-s-interview-joe-sutter-14734609/). I feel like aerospace has more (safety) requirements and tech specs to hit than we do for tasks like running. I wonder what the breakdown of what each engineer did in Honor looked like.
On another note, I've been itching to going back to building robots. In my naive past I've usually just kinda guesstimated/back-of-the-napkin'd the specs I'd need for the actuators given a specific task, like walking. But I'm curious, in your experience what's the best design approach? Some people I've seen (and what I'm looking to try) is to guesstimate/assume your motors are strong enough for most basic tasks and then CAD -> Sim -> Refine -> Sim -> Build Hardware.
Here's a basic recipe for motor sizing to build robots:
- Start with a catalog (e.g. the TQ one referenced)
- Follow the method in this article to size the gearing for a few motors
- Get a sense of the total power for the best gearing for each motor. Typically larger motors will be better, but they will of course weigh more. Use judgment to trade off motor weight vs. total power
This is only the first pass, you'll get secondary constraints like voltage, if that gear ratio is attainable using a gearing technology you can use, etc. If this topic is interesting, I can try to write more about it in the future, but this should be a starting point!