How you might build an IMU + GPS fusion algorithm suitable for unmanned aerial vehicles (UAVs) or quadcopters.

Estimate the pose (position and orientation) of a ground vehicle using an inertial measurement unit (IMU) and a monocular camera. In this example, you:

MATLAB Mobile™ reports sensor data from the accelerometer, gyroscope, and magnetometer on Apple or Android mobile devices. Raw data from each sensor or fused orientation data can be obtained. This examples shows how to compare the fused orientation data from the phone with the orientation estimate from the ahrsfilter object.

Estimate the position and orientation of ground vehicles by fusing data from an inertial measurement unit (IMU) and a global positioning system (GPS) receiver.

Combine robot odometry data and observed fiducial markers called AprilTags to better estimate the robot trajectory and the landmark positions in the environment. The example uses a pose graph approach and a factor graph approach, and compares the two graphs.

Reduce the drift in the estimated trajectory (location and orientation) of a monocular camera using 3-D pose graph optimization. Visual odometry estimates the current global pose of the camera (current frame). Because of poor matching or errors in 3-D point triangulation, robot trajectories often tends to drift from the ground truth. Loop closure detection and pose graph optimization reduce this drift and correct for errors.

Monocular visual-inertial odometry estimates the position and orientation of the robot using camera and inertial measurement unit (IMU) sensor data. Camera-based state estimation is accurate during low-speed navigation. However, camera-based estimation faces challenges such as motion blur and track loss at higher speeds. Also monocular camera-based estimation can estimate poses at an arbitrary scale. On the other hand, inertial navigation can handle high-speed navigation easily and estimate poses at world scale. You can combine the advantages of both types of sensor data to achieve better accuracy using tightly coupled factor graph optimization. For good execution time performance only a small portion of the entire factor graph containing only most recent measurements is optimized at every optimization step. This variant of factor graph optimization is popularly referred to as Sliding Window or Partial Graph Optimization.