Start the simulator and run the program with no arguments to use my parameters or pass the following parameters: p, i, d, max_running_time, max_cte_before_fail
The PID controller has three gains that can be adjusted to get good performance. The p or proportional gain controls how much of a response should be applied in direct proportion to the cross track error. The i or integral gain is used to offset long running errors such as those caused by a bias. It is applied to the integral of the cross track error. Finally, the d gain is the amount to adjust by the derivative of the CTE. It is used to dampen oscillations caused by p gain.
The tuning process went as follows: First I zeroed out all the gains and incrementally increased the p gain first so that the car would stay on the track and then eventually until the oscillations became so violent it would leave the track even on a straight segment. I then increased the d gain to dampen these oscillations. Finally I adjusted the I gain. There was lots of back and forth though, it was not this straightforward.
After this rough tuning manual tuning, the python script auto_tune.ipynb was run to automatically adjust the values using the twiddle optimization method.
p = [p gain, i gain, d gain]
before tuning (baseline): p=[.2,.1,.1], x axis is roughly time
after tuning: p=[0.7308569492165553, 0.1338155374232569, 0.3593786299378923]
The metric for the twiddle method was the RMS CTE for a duration of 3000 control updates (or less if the CTE were to ever go over 3.0, I interpreted this as "off the track"). The algorithm increased the p gain to a much higher value than I had found by hand and relying on strong d gain to keep the car from overshooting.
CTE and steering angle (tuned)
video of the tuned system
- cmake >= 3.5
- All OSes: click here for installation instructions
- make >= 4.1
- Linux: make is installed by default on most Linux distros
- Mac: install Xcode command line tools to get make
- Windows: Click here for installation instructions
- gcc/g++ >= 5.4
- Linux: gcc / g++ is installed by default on most Linux distros
- Mac: same deal as make - [install Xcode command line tools]((https://developer.apple.com/xcode/features/)
- Windows: recommend using MinGW
- uWebSockets == 0.13, but the master branch will probably work just fine
- Follow the instructions in the uWebSockets README to get setup for your platform. You can download the zip of the appropriate version from the releases page. Here's a link to the v0.13 zip.
- If you run OSX and have homebrew installed you can just run the ./install-mac.sh script to install this
- Simulator. You can download these from the project intro page in the classroom.
- Clone this repo.
- Make a build directory:
mkdir build && cd build - Compile:
cmake .. && make - Run it:
./pid.