For the last few days, the outdoor temperature dropped to the teens from the twenties. As a result, she didn’t get off the ground; she just jittered around. The graph of double integrated (accelerometer – gravity) shows she believed she’s was climbing right at the point the IMU temperature dropped.
This looks awfully similar to this.
The problem is that in these cooler temperatures, once the props start spinning, the IMU cools, and the temperature sensitive accelerometer output shifts as a result. Because I take a snapshot of gravity prior to the props spinning, net acceleration (accelerometer – gravity) is wrong, so integrating it twice to get distances is very very wrong.
The problem showed up this time because to add the salad bowl lid, I had to remove her Pimoroni Tangerine Dream PiBow, thus exposing her to the elements. For now, I’ve just moved the initial read of gravity to immediately before take off, but I’m also working out how to reintroduce the PiBox case or similar in a lower profile version to that the salad bowl still fits neatly on top.