Control and Trajectory Tracking for Autonomous Vehicle

Proportional-Integral-Derivative (PID)

In this project, you will apply the skills you have acquired in this course to design a PID controller to perform vehicle trajectory tracking. Given a trajectory as an array of locations, and a simulation environment, you will design and code a PID controller and test its efficiency on the CARLA simulator used in the industry.

Installation

Run the following commands to install the starter code in the Udacity Workspace:

Clone the repository:

git clone https://github.com/udacity/nd013-c6-control-starter.git

Run Carla Simulator

Open new window

• su - student // Will say permission denied, ignore and continue
• cd /opt/carla-simulator/
• SDL_VIDEODRIVER=offscreen ./CarlaUE4.sh -opengl

Compile and Run the Controller

Open new window

• cd nd013-c6-control-starter/project
• ./install-ubuntu.sh
• cd pid_controller/
• rm -rf rpclib
• git clone https://github.com/rpclib/rpclib.git
• cmake .
• make (This last command compiles your c++ code, run it after every change in your code)

Testing

To test your installation run the following commands.

• cd nd013-c6-control-starter/project
• ./run_main_pid.sh This will silently fail ctrl + C to stop
• ./run_main_pid.sh (again) Go to desktop mode to see CARLA

If error bind is already in use, or address already being used

• ps -aux | grep carla
• kill id

Project Instructions

In the previous project you built a path planner for the autonomous vehicle. Now you will build the steer and throttle controller so that the car follows the trajectory.

You will design and run the a PID controller as described in the previous course.

In the directory /pid_controller you will find the files pid_controller.cpp and pid_controller.h. This is where you will code your pid controller. The function pid is called in main.cpp.

Step 1: Build the PID controller object

Complete the TODO in the pid_controller.h and pid_controller.cpp.

Run the simulator and see in the desktop mode the car in the CARLA simulator. Take a screenshot and add it to your report. The car should not move in the simulation.

Step 2: PID controller for throttle:

1. In main.cpp, complete the TODO (step 2) to compute the error for the throttle pid. The error is the speed difference between the actual speed and the desired speed.

Useful variables:

• The last point of v_points vector contains the velocity computed by the path planner.
• velocity contains the actual velocity.
• The output of the controller should be inside [-1, 1].

2. Comment your code to explain why did you computed the error this way.

3. Tune the parameters of the pid until you get satisfying results (a perfect trajectory is not expected).

Step 3: PID controller for steer:

1. In main.cpp, complete the TODO (step 3) to compute the error for the steer pid. The error is the angle difference between the actual steer and the desired steer to reach the planned position.

Useful variables:

• The variable y_points and x_point gives the desired trajectory planned by the path_planner.
• yaw gives the actual rotational angle of the car.
• The output of the controller should be inside [-1.2, 1.2].
• If needed, the position of the car is stored in the variables x_position, y_position and z_position

2. Comment your code to explain why did you computed the error this way.

3. Tune the parameters of the pid until you get satisfying results (a perfect trajectory is not expected).

Step 4: Evaluate the PID efficiency

The values of the error and the pid command are saved in thottle_data.txt and steer_data.txt. Plot the saved values using the command (in nd013-c6-control-refresh/project):

python3 plot_pid.py

You might need to install a few additional python modules:

pip3 install pandas
pip3 install matplotlib

Answer the following questions:

• Add the plots to your report and explain them (describe what you see)
• What is the effect of the PID according to the plots, how each part of the PID affects the control command?
• How would you design a way to automatically tune the PID parameters?
• PID controller is a model free controller, i.e. it does not use a model of the car. Could you explain the pros and cons of this type of controller?
• (Optional) What would you do to improve the PID controller?

Tips:

• When you wil be testing your c++ code, restart the Carla simulator to remove the former car from the simulation.
• If the simulation freezes on the desktop mode but is still running on the terminal, close the desktop and restart it.
• When you will be tuning the PID parameters, try between those values:

Project report

Once we begin the project and run the CARLA simulator as described in the main README file (without implement contrls and another stuff) the vehicle is fully stopped.

Throttle Control

As can be seen in the throttle graph, an attempt is made to eliminate the error. However, it cannot be completely eliminated, at least with these parameters.

Additionally, you can see how the proportional parameter generates that "mirror" effect between the output and the error, provided by a scalar value.

Steer Control

On the other hand, in the steering error graph it can be seen how values close to the constant trigger the output due to the integral parameter of the controller.

• How would you design a way to automatically tune the PID parameters? I think that nowadays there are many ways to automatically calculate these parameters, even in the industry nowadays PLCs do it very easily. Now, focused on the course the Twiddle method could be of great help when optimizing these coefficients.

• PID controller is a model free controller, i.e. it does not use a model of the car. Could you explain the pros and cons of this type of controller? Advantages of the PID Controller include; simplicity and ease of tuning the parameters, as these are only 3 and in general terms a PID controller works well in most of the systems. However, many of the variables to which a vehicle is exposed in these cases vary according to its speed and various other parameters, increasing the complexity of the system (some being non-linear) and rendering a controller as simple as the PID unusable.

• What would you do to improve the PID controller? Integral anti-windup tactics can be used to avoid overloading the integral part of the PID at all costs. In addition, fast-forward techniques can be explored to decrease the reaction time of the controller.