I've stumbled across an issue and was wondering if anyone else has seen anything like that.
A bit of a history. I used to have a quad with NTM2836 750kV motors with 1245 props, AFRO 30A ESCs with SimonK and a Pixhawk. It did fly well, however when I decided to add a little gimbal and a camera my flight times went from 10-15 minutes to 2 and even less. It was time to get new motors, so I got RCTimer 5010-530 motors and 1755 motors.
While waiting for them I read that BLHeli is better than SimonK in some cases so I flashed all my ESCs. Pixhawk got updated from 3.2 to 3.3.2. And here is the problem. All new setup (new motors, props, BLHeli and AC3.3.2) flips randomly in different directions upon take-off. I re-checked all wiring/motor connections/rotation directions many times. I recalibrated IMU and compass few times. No luck.
When holding quad in hand (with motors spinning) and giving it a little tilt to the right for example, it tries to stabilize itself by tilting forward, then left, then back, essentially oscillating in circle like a gyro would. All PIDs were default ones.
Next test I did was strapping quad down to a big heavy reel of wire I have. It's flat on top, heavy and very stable. So, with the quad sitting on top of it firmly attached, I give a little throttle and put a little roll to one side from TX. Two motors start speeding and two opposite ones slow down like they should. However, when the roll input is released, nothing happens. Even more so, all motors start gaining more and more speed. It is soo weird, as physically the quad does not move during the test at all.
Finally, this morning I re-flashed my ESCs back to SimonK and Pixhawk to 3.2. After generic calibrations of TX, accel and compass, with default PIDs (same as in previous tests) the quad took off with no problem at all. It needs some PID tuning, but main problem just disappeared! I did all the same tests as before and quad behaves like it should, stabilizing itself in correct direction and on roll test on stable platform the motors slow down as the roll input goes to neutral.
Has anyone experienced anything like that before? Is it me being stupid and not doing something obvious or is there a bug in software somewhere?