kriswiner
commented
6 years ago Use 6 DoF fusion and forget about the mag. pitch and roll only need an accel and gyro. Madgwick has a very nice 6 DoF fusion filter for the MPU6050 which should work well here.
https://github.com/kriswiner/MPU6050/blob/master/quaternionFilter.ino
On Fri, Mar 9, 2018 at 3:55 AM, usinjin notifications@github.com wrote:
Hey Kris,
I have a somewhat challenging project: I'm using a 9DOF (FXOS8700CQ + FXAS12002 accel., mag., gyro) to measure the slope of the ground in a 4' radius. The sensor is mounted at the end of a metal (non-ferrous!) "arm" that rotates 360 degrees, measuring the pitch of the sensor every ~3 degrees. I'm running a sensor fusion algorithm using Mahony. I set the gyro trims at the beginning of the run (based on average values) and am using MotionCal to get a hard- and soft-iron matrix for the magnetometer.
When the magnetometer is calibrated, the algorithm works pretty well. When the calibration is somewhat off, the drift becomes absolutely horrendous. Here's the issue: when I have to take the model to demonstrate, it's in a completely different environment (usually inside an office). I don't usually have the option to recalibrate there. The values the arm reports are very far off because of this (boy, does it not look good in front of my project manager!).
Here's my question. Besides the obvious recalibration, is there anything I can do to avoid this issue? Note that we don't need to return the yaw from the algorithm--just the roll (which is the right direction for the orientation on our arm.)
The arm will hopefully one day be a plug-and-play solution, and at the moment, the question of an extra manual recalibration step is out of the question. Is there a quick-and-dirty sort of calibration I can do using only raw sensor values right at the start of the running routine?
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/kriswiner/MPU9250/issues/245, or mute the thread https://github.com/notifications/unsubscribe-auth/AGY1qpgvfJX4IWDTXhf8-S_11XH1FOzGks5tcm3DgaJpZM4SkKSW .
usinjin
Hey Kris,
I have a somewhat challenging project: I'm using a 9DOF (FXOS8700CQ + FXAS12002 accel., mag., gyro) to measure the slope of the ground in a 4' radius. The sensor is mounted at the end of a metal (non-ferrous!) "arm" that rotates 360 degrees, measuring the pitch of the ground every ~3 degrees. I'm running a sensor fusion algorithm using Mahony. I set the gyro trims at the beginning of the run (based on average values) and am using MotionCal to get a hard- and soft-iron matrix for the magnetometer.
When the magnetometer is calibrated, the algorithm works pretty well. When the calibration is somewhat off, the drift becomes absolutely horrendous. Here's the issue: when I have to take the model to demonstrate, it's in a completely different environment (usually inside an office). I don't usually have the option to recalibrate there. The values the arm reports are very far off because of this (boy, does it not look good in front of my project manager!).
Here's my question. Besides the obvious recalibration, is there anything I can do to avoid this issue? Note that we don't need to return the yaw from the algorithm--just the roll (which is the right direction for the orientation on our arm.)
The arm will hopefully one day be a plug-and-play solution, and at the moment, the question of an extra manual recalibration step is out of the question. Is there a quick-and-dirty sort of calibration I can do using only raw sensor values right at the start of the running routine?
Edit: Further info: Using the ESP32, Arduino core (latest toolchain, ESP-IDF, and Arduino files)