Your robot is great, you should share some day how to build it. the idea of using a cell phone as its microncontroller is awesome :) I would like to use Redmi Note 7 or 8 or 9.
I’m really enjoying watching your videos, I’m going to start making my own hexapod soon, I’ll be using MG90S servos. A word of warning though, those MG995 servos are supposed to come with metal gears inside, sometimes the come with plastic instead and they just don’t last, it might be worth checking if you have proper metal gears. Also have you considered adding positional feedback to the servos, that way the controller can know where the servos are at all times. Also which stereo camera are you using?
Yes it seems only one of the gears is metal in MG996, and it is causing some issues already. I was really hoping that these will work with the new hexapod but doesn't seem like it! :( The camera is really very generic. Just look up "Synchronized Dual Lens Stereo USB Camera".
@@MakeYourPet thats the exact thing that happened to me, it seems it is a bit of a lottery whether or not you get the fully metal ones especially when all the ones I've seen have been advertised as having all metal gears. It maybe wouldn't be as bad if they were all plastic gears but I found if you try to turn the servo by hand, even just a couple of times, the metal gear just rips through the plastic one, I'm sure that would happen pretty quickly too if you put any force on the servo. I don't know if there are any reliable sellers on AliExpress that do sell genuine all metal gear ones but that might be worth looking at.
Hi, friend! Great job. But you understand that for a real project you need to use feedback from the motors. It will be good if the neural network controls the motors and receives feedback from them. It is necessary that the neural network takes the force on the motor and the position. I am currently working on this very issue. Do you have any thoughts on this?
Hello friend! Your project seems very interesting. And yes, you need some soft of feedback if you are trying to train a neural net for the movements. But for this project which uses regular RC servos I just skipped that for simplicity, because I had to add wiring and use 18 input channels to read the data. The idea of training a deep net to control all the movements based on the feedback is very tempting, but expect it to be very challenging. I know of multiple similar attempts by big tech companies which all kinda failed at the end. For example see this ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-gn4nRCC9TwQ.html