PDJ Weekly Livestream # 22 - Update + Chat with Jon

This week was a fun one! Lots of little problems came and went, and I've made good progress on two fronts.
🔗 Link to the Ruckig trajectory generator:
ruckig.com/
🔗 Here's a link to the project's Github page:
github.com/Project-DeepBlue-J...
If you're specifically interested in the code used in the throw testing, you can find it here:
github.com/Project-DeepBlue-J...
(The "better timing" one is the one I used in the demonstration)
🔗 And to the project's Zulip site:
pdj.zulipchat.com/
🔗 If you want to help support this project, the best way to do so is through my Patreon, here:
/ harrisonlow
00:00 - Intro
01:25 - CAN-undrums are Solved! (for now...)
02:50 - Why Redesign the Platform?
07:15 - Why am I using the Current Geometry?
11:15 - How to Improve the Simulation?
14:05 - Jacobi-yums
19:44 - What Poses Should We Test?
24:00 - What About Throwing?
38:20 - Jugglebot and Onshape
41:30 - The Teensy's Home
43:30 - IMU vs Inclinometer
48:30 - Chat with Jon
50:00 - Throwing Trajectory
1:06:00 - Real Time Clock - Worth it?
1:15:20 - Back to Trajectory
1:19:30 - Jacobians
1:25:00 - I Love ChatGPT!

[@DennisBouw]

30 degrees as angle inwards seems a bit too steep for the jacobian. Is that a realistic angle? You know the minimal throw height for juggling 3 balls, that throw height gives you the maximum throw angle inwards where it lands in reach of the platform. The more realistic you pick this angle, the better the optimization of the platform

[@harrisonlow]

Yep you're absolutely right! I'll have a closer look at this to get an angle that makes more sense. Tbh I was a bit caught up in just getting the pose generator to work at all that I hadn't even thought about that yet!
Cheers for pointing this out 😊

[@DennisBouw]

No problem, I really enjoy watching your progress and your dedication to this project.

[@yeroca]

Just intuitively, taking the average of the jacobian doesn't sound like a good idea, because it will average out the "peak badness" of certain possition/poses. So maybe the score should have two values, such as average jacobian, and peak bad jacobian.

[@harrisonlow]

I'm not 100% sure how to interpret what you mean by this; there's two ways I'm reading it.
For the first, I'm not planning on averaging the 36 values in each individual Jacobian as part of each pose/geometry 'score'. I haven't exactly landed on exactly how I'll do this yet, but it'll likely be some combination of the condition number and at least one form of symmetry.
For the second way of interpreting your question, I'm hoping to be able to store the Jacobians of each geometry + pose so that once I determine the 'best' I can readily check its results manually, just to be sure.

[@yeroca]

@@harrisonlow Ok, it sounds like you have a good approach.
I don't know what you mean by "condition number" and "one form of symmetry". I was just surmising based on a qualitative understanding (from your description) of how these results at each position/pose would give a set of values describing how much the platform moved/tilted when one actuator is perterbed. Averaging the values across the entire geometry to derive a score seemed like it could mask some problems. But since you won't do that, you can safely discard my comment :D Thanks for reading and your reply.

[@harrisonlow]

@@yeroca Yeah I didn't know what the condition number is before diving into this rabbit hole either.
I'll elaborate on exactly how I'm creating the 'score' when I get around to finalising it (hopefully by next week's stream 🤞)