Hi rent-crash,
Please remember that the rotation is the key, not the position.
The following approach benefits those who learn best by doing and reflects a hands-on, trial-and-error methodology often seen in creative fields like 3D modeling, animation, and game design. Since I know very little about your background, I might be dead wrong here.
My suggestion is to set up a few joints in a new file. Freely. Then again, try to match them roughly; the key is to be different for this exploration. I even suggest rotating some of them. Copy them for a backup, just in case.
Save the scene. Save iterations.
Set up a cylinder with no caps that fits closely to the joints and has enough density (not too high) in the mesh. Bind it. Name all joints similar to the Mixamo rig that you like to improve.
Then, use the Definition workflow to get a good motion transfer from the source mesh to the target mash( the one with the cylinder).
Save this. Take the definition tag of the source, rotate a few joints, and then define again. Explore how that affects the target rig.
Why do I suggest this? The typical problem I have seen over the past two decades is starting with an important (your robot) model and missing perhaps the playful and destructive exploration of meaning setup. Something that you can trash without any regret. But here is the time investment that lets you focus on more complex things.
If you save this, returning after a while and messing around is easy to refresh your memory.
Sure, that sounds like a lot of time when something bigger and better is waiting for you. But you will gain a solid command of this area faster when you explore it without consequences.
This is just a suggestion. You can toss it, but I don't mind. Perhaps it doesn't fit your learning style, so ignore it. But can 10 minutes do a lot of damage?
Enjoy
