The quadrotor has to follow another quadrotor (refered to as target). The target sends its GPS data each 4 seconds.

The idea is to use a Kalman filter as predictor to predict the target's trajectory. The prediction is fed into a PID position controller to improve the quadrotor's responsiveness.

The Kalman implementation for our case is documented here.

It runs the following steps:

1. Kalman prediction steps: the trajectory is predicted using a constant acceleration motion model

2. If there is a new measurement

• Estimate acceleration with Bézier interpolation
• Run the Kalman correction steps: this accounts for the new measurement received to correct the prediction

3. Return the state estimate as waypoint prediction

This takes the predictor's output (waypoint prediction) and apply a PID using the already implemented PID controller.

The role of this PID is to project the prediction to a new waypoint that is further so the quadrotor moves faster.

• Code contains all the source files. We only worked on
  • kalman_predictor.c
  • kalman_predictor.h
  • track_following.c
  • track_following.h
• Documents contains some papers that inspired our work and also some notes we wrote in treating the problem (see directory Notes)
• Presentation contains the presentation in PowerPoint format