PID controller
Self-Driving Car Engineer Nanodegree Program
The goal of the project is to implement a PID controller for calculating the steering control input for the vehicle given a cross-track error.
This project should be run in the Simulator which can be downloaded here
Compiling and building
The project has been tested / run on Ubuntu.
- cmake >= 3.5
- make >= 4.1 (Linux, Mac),
- gcc/g++ >= 5.4
Cmake is used as a build system for the project.
In order to compile the code run the commands below in the bash
shell in the root project directory:
mkdir build
cd build
cmake ../src/
make
./pid
Create Eclipse project files:
cd build
cmake -G"Eclipse CDT4 - Unix Makefiles" -D CMAKE_BUILD_TYPE=Debug ../src/
Implementation
The code follows standard implementation of the PID controller.
Given a cross-track error (CTE) at each time stamp,
the ComputeSteering
method computes the control
input based on the Kp
, Ki
, and Kd
gains.
This input is then applied to steer the vehicle in the simulator.
As mentioned in the description, the control input is also bounded to [-1;1]
.
Reflection
- Describe the effect each of the P, I, D components had in your implementation.
P
- magnitude of correction to the target point,I
- removing systematic bias,D
- adds smoothness when correction to the target point.
- Describe how the final hyper-parameters were chosen.
I have used manual version of the twiddle algorithm described in the lecture. So essentially I started with a parameter guess that was shown in the lecture, and then manually tune the parameters until reaching a satisfactory performance.
The problem of applying twiddle, gradient descent or similar parameter space search algorithms
is that the simulator is buried inside the Unity 3D GUI (i.e. one should launch the binary, then click-click-click
,
and only then the track simulator will pop up), and cannot be launched as a binary.
In this respect the manual tuning is faster.