Here is the latest status:
I tested the unit again after installing the vibration dampener. The test didn’t go well. Again, the unit chose a different direction every time I initiated the system and took off. I adjusted the dead band zone during the flight but no luck.
So, I decided that I need to change the position of the accelerometer and place it closer to the center of gravity, to reduce the physical sideways impact of the tilting and tested again. The unit reacted better but it didn’t really hold position. It kind of drifted away. But this made me feel like I am getting closer to the solution.
Here is one of the things with this concept. Because of the way the accelerometers work, when I activate the system, it should basically keep the current speed. So, if it is still when the system is activated, then it should stay still. But if it is in motion at 1m/s for example it should hold that. And you cannot change that with the stick input because it conflicts with the sensor input and the unit gets out of control (I tried this). So, I decided to change the stick input logic when the system is activated. Now, if the unit is moving backwards when the system is activated, when I push the stick forward, it inputs a constant in the output equation to slow the unit down and finally stop it, instead of fighting the equation.
Also, I spent 5 hours to test every variable one by one at home, hand simulating the flight back and forth. I found the combination which I think works the best in the serial monitor, looking at the system output. When doing this, you need to test all combinations, including the ones that doesn’t make sense at all. Many times, you will find that the one doesn’t make sense is the working combination. Which actually was the case in this one. I found a combination which works ten times better than the other combinations that I tested (which I still think make lots of sense).
I took the unit out just before sun down and tested. I think the unit response is way better in terms of “trying to hold the position”. I initiated the system twice, once right in the beginning of the video and once right after the half of the video. The unit reacted erratically because the compensation which seemed normal before is way high with the new setup. Now I assigned the 5th channel to the sensor compensation and will test again to find the right compensation rate with the new setting. You never know but I feel like I am getting close.