Nothing is wrong. This is the expected behavior. The accelerations due to gravity and those due to other forces are equivalent.
You can't directly measure tilt (for example) while you are accelerating, but with some considerable mathematical effort, it is possible to estimate tilt in such a situation.
I don't know about the MPU6050 but you're at least supposed to have an absolute external reference, e.g., 3D Compass values.
Anyway, the MPU6050 (and even the MPU9150 providing a comapss) are outdated.
What do you mean by considerable math effort?
Do you know how to proceed (or links, posts, articles)?
You need a good filter (either the complementary filter or a Kalman filter) to estimate absolute orientation. The theory is taught in advanced engineering and aerospace navigation college courses. My favorite textbook is Dan Simon's Optimal State Estimation.
A filter that works well with your IMU and the Arduino is RTIMULib. As pointed out above, you will need a magnetometer too.
Or you can buy the BNO055 sensor, which is self contained, with a built in ARM processor doing essentially the same thing as RTIMULib.
I still have the same proble; Roll measure are still influenced by "turning acceleration".
I'm using the NAxisMotion library (from arduino)
here is part of the code
setup:
NAxisMotion mySensor;
I2C.begin(); //Initialize I2C communication to the let the library communicate with the sensor.
//Sensor Initialization
mySensor.initSensor(); //The I2C Address can be changed here inside this function in the library
mySensor.setOperationMode(OPERATION_MODE_NDOF); //Can be configured to other operation modes as desired
mySensor.setUpdateMode(MANUAL); //The default is AUTO. Changing to MANUAL requires calling the relevant update functions prior to calling the read functions
Here is some data( data on xls sheet and graph on next sheet)
I logged the yaw, roll of a (farm) tractor ridding at (around 8km/h) on field.
The captor is well secured at the cab floor.
Data are logged each 100ms.
The followed path is ridding 200m then turn left (about 180°) in order to go back again 200m then turn left about 180 ° and so on.
So each time I turn 180° , we can see a very great yaw variation (normal) but also a greater roll variation (around 15°) compared to the "mean" roll which is bouncing between -4/+4° (due to terrain roughness).
I looked at the yaw versus roll plot, and the data do suggest the built-in filter may not be working as well as it could.
If so, the problem may be due to incorrect choice of the relative weights of the gyro data versus the accelerometer data. During a sharp turn, the gyro data should be weighted more heavily than the accelerometer.
It is likely that the filter built into the BNO055 is a Kalman filter, but of course the system model would be wrong for your particular test case. To implement an optimal Kalman filter, the entire system (including the target vehicle and the vehicle's expected motion in 3D) must be modeled correctly.
If you want to optimize a Kalman filter for your application, you need an open source version such as RTIMUlib.
So I guess that you're using another bno055 library, which one? Would this library give better result for my application?
For which application do you use an imu?
You said "...the entire model must be modeled correctly...".
In my particular application, a roll is always accompanied with a "lateral" acceleration (see picture).
I don't use a library. I access the BNO055 directly with I2C routines and just read the Euler angle registers.
The BNO055 will not function correctly if you have not performed the built in calibration routine (with the BNO055 in its final resting place), and you also have to monitor the calibration status to see that it remains calibrated. If you have not done so, you cannot trust the data it provides.
If you want to take a look at the routines I use, I've appended BNO055 autopilot code that I used for an autonomous electric boat. This code is intended for Atmel Studio 4 and not Arduino, but it would not be difficult to adapt it to the Arduino environment.
The BNO055 works as well as any other consumer grade IMU I've tried, including those from CHRobotics. If it doesn't work for your application (when used properly) you may need to write a more sophisticated Kalman filter. The reference is in post #5.
i will try based on your project, it will take some days (elapsed) in order to implement and test it.
Anyway I will come back on the forum in order to give you some feedback.
I did some calibrations and the measure are better when cailbration's status are " 3 3 3 3 " (System, Gyro, Accle, Magn).
Unfortunately these calibration's status don't remain at that good level (3333) for a long time. After a few minutes, it drops to System:0 G:1, A:2: M:1 (not always these values but each time very low figures )
I have to "shake" it to get again a good/better status, but again lost after a 2-3 minutes.
With this bad status, roll measure (the one which is important for my project) is untrustable!
I'm quite disappointed and I don't know what/how to do to have a good roll measure.
Yes, the calibration procedure is a mystery, although my unit does not seem to be as badly behaved as yours.
Are you exposing it to different magnetic fields or nearby iron objects AFTER calibration?
If the BNO055 doesn't work in your application, you might try one of the modules from CHRobotics. The owner, Caleb, is responsive to questions and might work with you.
No, the sensor is kept at the same place (tractor cab).
My application needs accuracy and precision (at least 0.5° of precision) for the roll measure in the range -5° to +5° at 10Hz.
Is the BNO055 suitable for this purpose?
The consumer grade IMU sensors that I know of are accurate to about 1 degree at best, and then only after very careful calibration of both the accelerometer and magnetometer and frequent recalibration of the gyro offsets (which depend strongly on the temperature).
If you need reliable measurements more accurate than that, your best bet may be a commercial grade IMU. But they are very expensive.
I suggest to communicate with the very knowledgeable owner of CHRobotics and discuss your particular application.
