Let's say you want to start toying with the idea of developing a humanoid robot, but you have zero (or pretty limited) prior knowledge in that field.

Common sense tells me that you should start with a small body part, like a hand, but what kind of tools and knowledge should you acquire to make it possible? Is FreeCAD + Webots/Blender + Python + human anatomy a correct combination prior to trying 3D printing and (in the far future) open source LLMs?

>>16731759 (OP)
Mechatronics

>>16731759 (OP)

Dunno about programming side but could likely build a hand out of standard parts with some 3D print "skin". If you use pneumatic instead of actuators you could even skip pressure sensors in the fingers, likely could regulate grip strength simply by varying air pressure (guess hydraulics would be overkill for non-heavy duty apps)... more standard parts, very simple control system. Ofc a hand that could actually "feel" would be more fancy...

>>16731759 (OP)
Real question also for the robotics fags in this thread: why do prosthetic robot hands/arms still move like shit all slow and clunky in the year of our lord 2025? Why don't we have luke skywalker level shit yet?

>>16731759 (OP)
Start with microcontroller embedded systems. Pick up a cheap Arduino kit from Amazon so you can get an understanding of how to interface with the various sensors and actuators you'll need. Then get comfortable with whatever CAD software you intend to use. Then learn slicer software for 3D printing.
There is a skillset to 3D printing as well. Practice on some crap you downloaded on thingverse or some shit so you're actually competent in setting up/using the machine.
At that point you're ready for prototyping with more basic "robots." Start with a shitty RC car or something that will get you used to the unique challenges that come along with actually producing a finished product rather than bare-bones circuitry on a test bench. Craftsmanship matters, not just for aesthetics.

Once you're past the theory phase and have become competent in practice, then maybe some anatomy research would help. Can't hurt.
You're right that your goal should be to work on smaller parts first and work your way up. The simpler the better. Make sure every part works correctly on its own before setting them up in tandem.

Because of the awkward nature of the human form, you're gonna want to look into PID Loop Control so the thing can stay upright and not immediately fall over every time it tries to lift a finger.

>>16731759 (OP)
>Let's say you want to start toying with the idea of developing a humanoid robot
Get rich and start a robotics company. There is no other way to be "solely" responsible for the development of a robot, it's a multidisciplinary field, you'd just be one of the cogs in charge of a small aspect of the whole thing. At least as the CEO you hold the rights to the robot and are the de facto owner.

>>16731759 (OP)
https://www.youtube.com/playlist?list=PLkx8KyIQkMfU5szP43GlE_S1QGSPQfL9s

>>16731759 (OP)
Bottom line is: if you have to ask, then you are too stupid to do it.
Even smart individuals at best reach a proof of concept animatronic puppet phase. This shit is simply too complex.

>>16731759 (OP)
>zero prior knowledge
Then go to fucking school for electrical engineering, mechanical engineering, or computer science.
And you shouldn't be building robots at all until you get it working in simulation. Some good ones are ROS + Gazebo, Drake's supposed to have really good contact sim, and nvidia has some nice offerings.
>start with a hand
Oh fuck no. You picked like the most complicated body part. Start with a single actuator for the arms or legs, because most robots are just a frame for actuators. You'll find that making good actuators is really expensive and that anything you can build on your own will be crap unless you have an entire machine shop. No, you can't 3d print gears. Thing is china's starting to make actuators ridiculously fucking cheap(see robstride), so just wait a year and buy those. Or better yet, just buy an entire humanoid. It'll probably be cheaper
>human anatomy
Largely irrelevant
>>16731848
>pneumatics
Pneumatics are shit. They're difficult to control, inefficient, noisy, and break down all the time. If you want to do any fine control, you need fucking expensive valves

>>16731877
>PID control for a humanoid
Lol. PID can't be used directly because humanoids are MIMO and legs are in contact with the ground periodically. Most humanoids today use model predictive control or a reinforcement learned controller that was trained in simulation. But honestly, the more math and control theory you learn the better. Calculus and linear algebra are the minimum. With reinforcement learning, you can be pretty retarded as long as you're good with software.

>>16731848
>pneumatic instead of actuators
Ever done that outside /sci retardation?

>>16731759 (OP)
>a small body part, like a hand,
Hardest part thinkable, buy a set and play around. There are tons of it
>FreeCAD + Webots/Blender + Python + human anatomy
Way to hard to start with. Use models from printables etc. first to see how to design for 3DP.

>>16731877
>Start with microcontroller embedded systems.
Would recommend a servo controller. Lots exist and some have ready made software for robot/puppet move, timeline, presets etc. All for the start w/o a single line of code.

>>16731759 (OP)
>a small body part, like a hand
it might be small but complex as fuck
> FreeCAD + Webots/Blender + Python + human anatomy a correct combination prior to trying 3D printing and (in the far future) open source LLMs?
sounds solid to me, I don't think the tools matter that much, but imo it'd be wise to explore the ecosystem of those software choices of yours to verify how mature and solid are the current toolsets you intend to use, and the bridges between one and the other
I remember something, while I was studying multimedia, one of the teams shinned specially bright, because it implemented parallel development, meaning that as soon as a minimal version of something was implemented, the solution that would be using that implementation was already using that minimal version to roll with it, so when once everything was finished the ball was rolling from end to end, example, the animation integration into the videogame engine was up and running before the model was fully finished and rigged

>>16731948

>you want to do any fine control, you need fucking expensive valves

Thinking Festo, huh. Well if you do it, do it proper. Rather would worry about pressure tank life, sure is less efficient than a battery (unless we assume some fuel cell and run it on grain alcohol).

>Largely irrelevant

Would agree except for social models.

>>16732849
Quite impressive. Now, I don't think we need something as strong and powerful for a humanoid waifu. If it can handle what a human could and is durable enough to last 100 years if you treat her well, that should be enough.