Inverse Kinematic Library Suggestions?

@krooger I don't know what kinematics is, but it sounds like he same math a game engine uses, which is just matrix arithmetic. Is that close?

@krooger

I haven't started collecting kinematic library links ... yet. But I've found that James Bruton  (James Bruton - YouTube) often has viable code to fill the gap. Adding a generic library can be a PITA in the prototype stage. I have a BittleX quadruped and the kinematic code there is tied up in knots. Others I've seen have the high requirements you mention. I'll pipe up if I come across something interesting.

What is the platform you're using for the bot? Are you using a ESP32 or a Pi?

EDIT: you did say ESP32.  Sorry. drain bamage.

@krooger

On second thought, I'm not sure you need to be concerned about the 3rd coord. It defines a plane, similar to the base rotation of the Fabrik2D example. That's the offset you mention.  So you use the 2D to get to (x,y) and the offset is the delta from the current z pos.

@krooger

What I'm thinking of is run the solver twice.

1. for (x, y) fixed z.

2. for (x, z) with the y from 1.

I think that should work.

When I get a chance, I'll run a few tests of the code.

Did you try this one? https://github.com/kousheekc/Kinematics

Thanks for the replies, I've been AFK.

Posted by: @zander

↑

same math a game engine uses

Indeed that math would work, however it has been over a decade since I've studied Linear Algebra and I'm trying to avoid relearning. I failed that course the first time through haha.

Posted by: @tfmccarthy

↑

What I'm thinking of is run the solver twice.

Yea I was kind of thinking along the same lines, however I might have thought up a way to solve the problem from scratch just using trig. I'm thinking I can calculate the Z (hip abduction) angle independently which basically tilts the plane, then project my other joints onto this new plane and then solve a 2D problem.

I think this should work and I can re-use the 2D solver I already wrote.

Unfortunately I return to my day job now so might not get to try anytime soon.

Posted by: @aliarifat

↑

Did you try this one? https://github.com/kousheekc/Kinematics

However I'll definitely have to look into this library.

@krooger

Posted by: @krooger

↑

Yea I was kind of thinking along the same lines, however I might have thought up a way to solve the problem from scratch just using trig. I'm thinking I can calculate the Z (hip abduction) angle independently which basically tilts the plane, then project my other joints onto this new plane and then solve a 2D problem.

I did a snap review of the Fabrik library using Google Test and I now think it should work. I have to put a test arm together to really verify it. I will need to get the correct servo axial you use but I can work without it for the moment.  I was going to have to do this anyway for my bluesky project, so it will serve two purposes.

FYI, an example of this is the Petoi OpenCat robot. However, it doesn't calculate the angles on the fly and so doesn't use kinematics. They use a separate tool to record the iterative positions and save them in EEPROM for later playback.  

@krooger 

I'm working on a test harness for the FABRIK library and need your help to verify that I have the servo layout correct. The picture has 3 servos, (A, B, C) which would correspond to

• A = shoulder or hip
• B = elbow or knee
• C = wrist or ankle.

The layout gives the rough idea of the servo orientation and the linkage between B and C.

Does this match your robot layout? I think you have the C servo moved closer to servo B and use a ridged rod to control the joint but that should be mechanically the same, I think. (My mechanical engineering is, um, ... primitive.)

I also need to get a better idea of how you did the A (shoulder) bracket and how you mounted that to servo B. I don't have a 3D printer so I'm fabricating all of this with popsicle, chopsticks, paper, paperclips, screws, and glue. (Yeah, McGyver lives on.)

If you could provide a better view of the servos, etc., I'd be grateful.

Posted by: @tfmccarthy

↑

If you could provide a better view of the servos, etc., I'd be grateful.

This model isn't really setup to do a good animation but this should give a pretty good idea of how it is setup.

My design does not have a wrist/ankle joint, so the first 2 joints you see are both hip joints so that it simulates a Ball Joint and can move in 3 Dimension. The last joint will control the Knee flexion.

Posted by: @tfmccarthy

↑

I think you have the C servo moved closer to servo B and use a ridged rod to control the joint but that should be mechanically the same

I have all my servos bunched up by the hip to reduce the torque on the leg by keeping the weight close to the pivot. So I use a pushrod (not modeled in the video) to transmit the force but it is functionally the same as if I placed the servo at the knee joint. Lots of designs do this but I see it as just additional mass far from the pivot and can put more load on the servos.

The issue I was having with the FABRIK library is that my "arm" reaches parallel with the base rotation whereas the FABRIK library has it so the arm moves perpendicular to the base rotation. I was thinking of a way to adapt the library to work, but I'm pretty sure I'm just going to write the code from scratch when I find a bit more time.

@krooger 

This helps lots. Thanks.

I now have to work out the test stand to keep the rig upright.

Will post what I end up with.

@krooger

@tfmccarthy

I have seen working examples of inverse kinematics on those toy arms without the need for a library?

Posted by: @robotbuilder

↑

I have seen working examples of inverse kinematics on those toy arms without the need for a library?

Perfect, this is almost exactly what I intended to do and also more elegant I'm sure.

I'll definitely follow this framework. 

@robotbuilder 

Way cool! This reminds me of James Bruton's solution.

Now I gotta try this one and go for his 6DOF arm.

Great find! (XMas came early this year!)