Page 1 of 1

Sensor Fusion

Posted: Thu Dec 29, 2016 9:57 pm
by dewittpayne
Are there future plans to include sensor fusion to combine GPS, accelerometer, gyro and magnetometer readings in the app to improve the precision and accuracy of data? This sort of thing is done for drone flight control, for example, using inexpensive 9 or 10 degree of freedom sensors and GPS.

Posted: Thu Jan 19, 2017 5:52 pm
by brentp
That's a fine idea - there's probably some existing software or libraries out there to do just that. Got any leads for us?

Posted: Thu Jan 19, 2017 7:17 pm
by dewittpayne
Try some of these:

http://www.starlino.com/imu_guide.html

http://www.starlino.com/dcm_tutorial.html

Direction cosine matrices are better than Euler angles (roll, pitch and yaw), which can suffer what's known as gimbal lock. https://en.wikipedia.org/wiki/Gimbal_lock

http://diydrones.com/profiles/blogs/a-s ... dreckoning

The link to Matrix Pilot in the above is broken. It's here:

https://github.com/MatrixPilot/MatrixPilot/wiki

http://www.pieter-jan.com/node/11 (using a complementary filter instead of a Kalman filter)

http://www.pieter-jan.com/node/7 (integrating rotation rate data to get direction)

The explanation of drift in some of the above is wrong. The problem is that when you integrate noisy data, digitally or otherwise, you get what's called a random walk or a unit root. That means that the range of error in velocity or angle when integrating accelerometer or rotation rate data increases in proportion to the square root of time. The error increases even faster when you integrate accelerometer data twice to calculate position. That's why you need a GPS to provide drift correction and for something like a quadcopter which can hover, a magnetometer to provide a fixed direction reference.

Submarine and missile inertial guidance systems use extremely low noise IMU's based on things like ring laser gyro's or extremely precise. rapidly rotating beryllium spheres so the drift is small. That sort of thing is obviously overkill for logging vehicle data.

Posted: Thu Feb 02, 2017 1:48 am
by dewittpayne
The dead reckoning feature in the new RaceCapture Apex seems to be exactly what I was talking about in terms of sensor fusion. Too bad it's so expensive. Are the rotation rate and acceleration sensors much higher quality than in the RaceCapture?