Udacity Self-Driving Car Nanodegree - Extended Kalman Filter Implementation

This project consists of implementing an Extended Kalman Filter with C++. A simulator provided by Udacity generates noisy RADAR and LIDAR measurements of the position and velocity of an object, and the Extended Kalman Filter[EKF] must fusion those measurements to predict the position of the object. The communication between the simulator and the EKF is done using WebSocket using the uWebSockets implementation on the EKF side.

The project has the following dependencies (from Udacity's seed project):

• cmake >= 3.5
• make >= 4.1
• gcc/g++ >= 5.4
• Udacity's simulator.

For instructions on how to install these components on different operating systems, please, visit Udacity's seed project.

These are the suggested steps:

• Clone the repo and cd to it on a Terminal.
• Create the build directory: mkdir build
• cd build
• cmake ..
• make: This will create the following executable
  • ExtendedKF : EKF implementation.

From the build directory, execute ./ExtendedKF. The output should be:

Listening to port 4567
Connected!!!

The simulator connect to it right away.

The simulator provides two datasets. The difference between them are:

• The direction the car (the object) is moving.
• The order the first measurement is sent to the EKF. On dataset 1, the LIDAR measurement is sent first. On the dataset 2, the RADAR measurement is sent first.

Here is the simulator final state after running the EKF with dataset 1:

Here is the simulator final state after running the EKF with dataset 2:

The code compiles without errors.

The EKF accuracy was:

• Dataset 1 : RMSE <= [0.0964, 0.0853, 0.4154, 0.4316]
• Dataset 2 : RMSE <= [0.0726, 0.0965, 0.4216, 0.4932]

The EKF can be modified to use either RADAR or LIDAR data also. To do that we have to uncomment either line 65-67 or 68-70 in src/FusionEKF.cpp

The EKF accuracy with only LIDAR data was:

• Dataset 1 : RMSE <= [0.1838, 0.1542, 0.5748, 0.4893]
• Dataset 2 : RMSE <= [0.1673, 0.1567, 0.6566, 0.5001]

The EKF accuracy with only RADAR data was:

• Dataset 1 : RMSE <= [0.2310, 0.3355, 0.5076, 0.7001]
• Dataset 2 : RMSE <= [0.2439, 0.3375, 0.6035, 0.8174]

The Kalman filter implementation can be found src/kalman_filter.cpp and it is used to predict at src/FusionEKF.cpp line 149 and to update line 160 to 175.

The first measurement is handled at src/FusionEKF.cpp from line 84 to line 112.

The predict operation could be found at src/FusionEKF.cpp line 149 and the update operation from line 160 to 175 of the same file.

Different type of measurements are handled in two places in src/FusionEKF.cpp:

An example of this calculation optimization is when the Q matrix is calculated src/FusionEKF.cpp line 134 to line 147.