Question about not having odometry data and using just accelerometers and GPS

[evanbekt]

As the title mentions, i have only GPS and accelerometer data. Even though it's pretty clear on how to use GPS data, when i combine GPS and IMU data that i have (which do not include odometry) my results are messy and completly inaccurate. Is there a clear path on how to use just acceleretion, pitch and yaw instead of velocity?

[betaBison]

Hi @evanbekt. Can you be as specific as possible with what sensors you're using so I can recommend some resources?

Broadly speaking, I would recommend adding linear velocity and possibly pitch, roll, yaw to the extended kalman filter states and then updating your state transition matrix accordingly (because velocity is the integral of your acceleration measurements).

Also, if your GPS receiver is giving you latitude and longitude (and not the raw pseudoranges and GPS satellite positions), I would not recommend basing your project off of this repository. This repository was for a class project for a GPS class and wasn't mean to be easily adaptable. If you tell me your sensor outputs, I may be able to recommend some other repositories that would be a better fit for you.

Some resources that have been helpful for me.

• Chapter 3 of Probabilistic Robotics by Sebastian Thrun et al about the theory of Extended Kalman filters
• Chapter 8 of Small Unmanned Aircraft: Theory and Practice by Beard and McLain which includes an Extended Kalman filter implementation using GPS, Accelerometers, and Gyroscopes.

[evanbekt]

Thank you for your immediate response and I will make sure to study your sources. I don't know the types of sensors since I'm just getting data that have already been collected. Most of the sources and implementations I found use linear velocity (which sadly i do not have). My sensor outputs are

• GPS (Lon, Lat, Alt)
• Acceleration, Yaw and Pitch (X,Y,Z)
• Altitude pressure for a barometer

I was hoping to succeed on tracking the devices the sensors are placed using your repository since it seemed the best one for this kind of task.