kharikri / carnd-pid-control-project

PID Controller to maneuver a vehicle

PID Controller

Introduction

This is the fourth project in the second term of the Self Driving Car Nanodegree course offered by Udacity. In this project, I implemented a PID controller to maneuver a vehicle around a simulated track. A PID controller continuously calculates an error value as the difference between a desired setpoint and a measured process variable and applies a correction based on proportional, integral, and derivative terms (sometimes denoted P, I, and D respectively) which give their name to the controller type.

PID Controller Implementation Details

The P, I, and D parameters play different roles in the control system as described below:

• P generates a steering correction proportional to the cross track error (cte). Cross track error is the lateral distance between the vehicle and the reference trajectory. A large P results in the vehicle overshooting the track leading to oscillations. A small P results in fewer oscillations but also means it is slower to respond to corrections
• I controls the drift in the vehicle. A misalignment in the car leads to drift which makes the car drive at an angle instead of in a straight line. To fix the effects of drift the I parameter collects the cte over time
• The D parameter dampens the overshooting caused by P. D is proportional to the changing rate of the cte

In this system, the simulator does not have drift so I keep the I parameter to zero. Essentially this becomes a PD controller which is straightforward to tune manually as explained below. P and D move in opposite directions and the trick is to find optimal values for these parameters to keep the vehicle on the track.

I first assume the D parameter to be zero to tune the P parameter of a P controller. As the steering angle must be between -1 and 1 the P parameter has a maximum value of 1.3 with a cte of 0.7598 obtained from the simulator. The steering angle is related to the cte with the following formula:

steering angle = -P * cte

Setting the values for the parameters P, I, D to 1.3, 0, 0 results in a wobbly vehicle as shown in this video. I progressively half the value of P to 0.65, 0.325, and 0.1625. For P value of 0.1625 the car still wobbles but much less. So, I fixed the P value to be 0.1625.

As mentioned before the D parameter dampens the wobbles seen with the P parameter. I start with a value of 1.0 for D and I get this result. The wobbles have considerably reduced but the vehicle goes off the road after some time. I progressively increment the D value by 1 and at D equals 4 I get a system where the vehicle stays inside the track. See the final result here.

With these parameters (P = 0.1625, I = 0, D = 4.0) I also varied the speed from 30 mph to 70 mph. The vehicle stays inside the track for speeds between 30 mph to 50 mph.

Conclusions

While these parameters keep the car on the track between 30 to 50 mph the car is still wobbly. To reduce the wobble, I tried decreasing P and increasing D further but for these values while the wobble decreases the vehicle goes on the curb a bit because it is not quick enough to respond to corrections. Here is a video for parameters (P = 0.0813, I = 0, D = 4.0). An algorithm like Twiddle to automatically choose the optimal parameter values will reduce the wobble and also enable to run the car at higher speeds.

---

Dependencies for Windows 10

• cmake >= 3.5
• All OSes: click here for installation instructions
• make >= 4.1
  • Windows: Click here for installation instructions
• gcc/g++ >= 5.4
  • Windows: recommend using MinGW
• uWebSockets
  • Run ./install-ubuntu.sh. Check here if you run into issues with websocket installation for windows 10
• Simulator. You can download these from the project intro page in the classroom.

Build Instructions

1. Clone this repo.
2. Make a build directory: mkdir build && cd build
3. Compile: cmake .. && make
4. Run it: ./pid