> So for "learning robotics" I would personally recommend starting with something like assembly lines.
That depends on what you want to learn. In assembly lines the solution very often is to constrain the inputs and limit variability on conditions. If that is what you train on that is what you will learn. You will have 101 tricks on how to QA the inputs and how to maintain the same grippyness on surfaces and so on. Which is great if that is what you want to do.
But if what you want to learn is how to make robots which can adapt to varying circumstances your best bet is to try that. In examples where failure is less costly and you can iterate quickly.
Also "robotics" is not a single thing. Even just the computer stuff has many subfields (localisation, perception, calibration, tracking, prediction, planning, control, etc etc). When learning your best bet is to pick a few of these and assume that the other things are either solved or pick a challenge where they are not needed. :) Not talking about all the "sibling subject" like electrical engineering, mechanical engineering, thermal engineering, fluid dynamics, etc. Cool cutting edge stuff usually requires innovation in more than one of these.
When you look at the latest Boston Dynamic video we all marvel at the exquisite control they demonstrate. But they also have to cool their actuators the right way otherwise the whole thing will just melt.
As a learner your best bet is to pick a stable platform made by people who know what they were doing and colour inside the performance envelope of that machine. What you make will probably won't break the state of the art but you will still learn a ton.