Welcome to the Advanced Kalman Filtering and Sensor Fusion Simulation exercise project. In this project, you will be developing the source code for a number of different types of Kalman Filters which are used to estimate the navigation state of a 2D vehicle problem; such as that would be found on a self-driving car!
This README is broken down into the following sections:
- Setup - the environment and code setup required to get started and a brief overview of the project structure.
- The Tasks - the tasks you will need to complete for the project
This project will use the Ubuntu 64 20.04.2.0 LTS VM C++ development environment that is setup for this course. (Refer to the Setting up the Development Environment Lesson) Please follow the steps below to compile the simulation.
- Install the dependencies
sudo apt install libeigen3-dev libsdl2-dev libsdl2-ttf-dev
- Clone the repository
git clone https://github.com/technitute/AKFSF-Simulation-CPP.git
- Setup the cmake build
cd AKFSF-Simulation-CPP
mkdir build
cd build
cmake ../
- Compile the code
make
- You should now be able to and run the estimation simulator
./AKFSF-Simulation
There are 6 main files of interest:
- kalmanfilter_lkf_student.cpp
- kalmanfilter_ekf_student.cpp
- kalmanfilter_ukf_student.cpp
- kalmanfilter_lkf_answer.cpp
- kalmanfilter_ekf_answer.cpp
- kalmanfilter_ukf_answer.cpp
To use one of the filter files, replace the kalmanfilter.cpp file with the one of interest.
When carrying out the Linear Kalman Filter task, use the kalmanfilter_lkf_student.cpp as a base and develop all the exercise on top of each other. The file kalmanfilter_lkf_answer.cpp is a working version that you can use to compare your solution too. The same thing applies to the Extended Kalman Filter and Unscented Kalman Filter exercises with the corresponding files. You can use any of the files as a base for your capstone project.
The simulation can be run with different motion and sensor profiles to test the different scenarios and evaluate how the filter performs with different conditions. The different profiles can be activated pressing the number keys 1-9,0, to active the corresponding profile.
- 1 - Constant Velocity + GPS + GYRO + Zero Initial Conditions
- 2 - Constant Velocity + GPS + GYRO + Non-zero Initial Conditions
- 3 - Constant Speed Profile + GPS + GYRO
- 4 - Variable Speed Profile + GPS + GYRO
- 5 - Constant Velocity + GPS + GYRO + LIDAR+ Zero Initial Conditions
- 6 - Constant Velocity + GPS + GYRO + LIDAR + Non-zero Initial Conditions
- 7 - Constant Speed Profile + GPS + GYRO + LIDAR
- 8 - Variable Speed Profile + GPS + GYRO + LIDAR
- 9 - CAPSTONE
- 0 - CAPSTONE BONUS (with No Lidar Data Association)
The tasks for this simulator can be broken down into 4 different areas:
- Linear Kalman Filter
- Extended Kalman Filter
- Unscented Kalman Filter
- Capstone Project
Starting with the kalmanfilter_lkf_student.cpp and replace the kalmanfilter.cpp file. Initialize the filter on first prediction step and then implement the 2D Vehicle process model and Linear Kalman Filter Prediction steps. Test and verify good performance with profile 1.
Continuing on from Exercise 2, Implement the GPS update step using the Linear Kalman Filter Update equations. Test and verify good performance with profile 1. Play around with tunings and see results for profiles 2,3 and 4.
Continuing on from Exercise 3, Modify the filter to initialize the filter on first GPS measurement rather than prediction. Use the flag INIT_ON_FIRST_PREDICTION = false to enable this functionality. Test and verify good performance with profile 1 and 2.
Starting with the kalmanfilter_ekf_student.cpp and replace the kalmanfilter.cpp file.
Implement the 2D Vehicle process model and Extended Kalman Filter Prediction steps. Test and verify good performance with profiles 1-4.
Continuing on from EKF Exercise 2, Implement the LIDAR update step using the Extended Kalman Filter Update equations. Test and verify good performance with profiles 1-8.
Starting with the kalmanfilter_ukf_student.cpp and replace the kalmanfilter.cpp file.
Implement the 2D Vehicle Process model and Unscented Kalman Filter Prediction steps. Test and verify good performance with profiles 1-4.
Continuing on from UKF Exercise 2, Implement the LIDAR update step using the Unscented Kalman Filter Update equations. Test and verify good performance with profiles 1-8.
Starting from any filter base replace the kalmanfilter.cpp file. Program a filter to provide the best estimation performance for profiles 9,0 (It should also work on any other profiles aswell).
This project was developed for the Technitute Course - Advanced Kalman Filtering and Sensor Fusion. Developed and produced by Dr. Steven Dumble.