Posted by hakonjdjohnsen 1 day ago
I imagine an alternative design with single motor mounted horizontally controls 4-wings like the dragonfly in Dune. It would eliminate the spinning.
I want to build a robot that does one simple thing - follow my voice and deliver something to it's original location. That means I need to learn batteries and power loads, electronics, audio processing, directional acoustics, more programming, soldering, and a half dozen other things.
AI + deep thinking about the problem.
The goal is not the objective but to learn everything along the way.
Even with the amount of open source software for this sort of application out there it will still take you at least a month but more likely two or three to get something that can be controlled.
If you want a nice place to start with building then I think Dhremflight's videos are an amazing resource as is liftoff on Steam because you can get some flying skills before you actually control something irl.
- How to correctly hang a pre-framed door (shim it, tack it in place with 16 gauge nails, and use one wood screw in the top 2 hinges to prevent the door from sagging)
- How to properly adhere MDF trim to the ceiling to hide a closet door track (screw a furring strip or a header board to the studs above the drywall, shim it so it’s level in case the ceiling is uneven, and then nail through the face of the trim into the strip)
- What the difference is between the 27 DAP putties and spackles in the paint aisle (turns out Painters Putty 53 is typically used to hide finish nail holes, because you tend to make a lot of them and can just smoosh it in and level it off with a palette knife!)
But I do agree about using LLMs for learning new things - it's great to just get those answers, hopefully the LLM doesn't hallucinate too much and cause you to destroy your project in the process ;) On the flip side, it removes the useful skill of learning to communicate with fellow human beings, and atrophies the skill of how to ask questions, which often can lead to answers just by phrasing the question correctly. It also removes the chance to join a community of like-minded individuals. It's a completely soulless interaction. Maybe it's a boon for anti-social types, but I think it might also be creating more anti-social behaviors in humans, in a time when IMHO we should be learning to become more social and accepting of each other.
I went down a similar rabbit hole soon after covid started to build a small, eh, 'hand sanitizer factory' (that's my story and I'm sticking to it). By the time it was done I had a fully automated distillation setup complete with load cells, six different temperature readouts, solid state current control, flow/temp control for liebig condenser recycle, hierarchy of pid loops largely managing output flow and an attempt at live specific gravity testing. That with a nifty little dashboard to show live progress and compare metrics from various runs. Spent a couple grand on it in total and learned a LOT in the process.
If you have some time available definitely worth it.
Exploration and tinkering! But you have to have passion and deep interest in that topic/problem
That's actually quite doable. The problem is that when the environment changes you will drift and then it becomes very hard to distinguish between actual heading changes and external influences. But if you can make a full spin you can (usually!) obtain magnetic North again. For drones this is a problem because they will drift and spin all the time. When you're on the ground and you have encoders on your axis you can do a lot better than with a drone.
RTK GPS is very nice to have but may not always be available (GPS itself may not always be available). I have some stuff in the works just for that. It does well in the lab but I still have (serious) problems in the field.
gps also won't give you orientation really anyway, and indeed, is mostly outdoors, and rtk is needed for this resolution but also rather inaccessible to amateurs.
If you're working on indoor navigation, that's very cool, when I used to do it, in my context it was easier because we controlled the environment of the robots (markers on walls of a known colour type stuff). We also explored passive radar type arrangements for picking up wifi repeaters, but it became complicated to eg. ignore bluetooth.
Though in the case of this video, the wii sensor bar style approach seems fine, assuming you control reflections
It's insanely complex stuff, I've had to do a lot of re-studying math and this whole project has given me more respect than I already had for the engineers behind modern positioning systems.