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