CarND-Controls-PID

Self-Driving Car Engineer Nanodegree Program

Dependencies

cmake >= 3.5
All OSes: click here for installation instructions
make >= 4.1(mac, linux), 3.81(Windows)
- 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
- Run either ./install-mac.sh or ./install-ubuntu.sh.
- If you install from source, checkout to commit e94b6e1, i.e.
```
git clone https://github.com/uWebSockets/uWebSockets
cd uWebSockets
git checkout e94b6e1
```
  Some function signatures have changed in v0.14.x. See this PR for more details.
Simulator. You can download these from the project intro page in the classroom.

Basic Build Instructions

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

Rubric Points

Your code should compile.

It does! See 'Basic Build Instructions' above.

Note: Twiddle.cpp file was added to CMakeLists.txt.

The PID procedure follows what was taught in the lessons.

PID.cpp contains the PID controller implementation. PID::TotalError() is where I calculate total error using P, I, and D gains and errors. PID::AccumulatedSquaredError() accumulates the squared error during the simulation, I use it during PID parameters tuning. The tuning happens in Twiddle.cpp - the instance of this class stores info about which parameter (P, I, or D) is being currently optimized. Twiddle::Iterate() increases/decreases the value of the current parameter.

Describe the effect each of the P, I, D components had in your implementation.

The best way to describe the components is to observe the extreme points on this video:

P gain is too big - oversteering.
P gain is too small - understeering.
I gain is too big - little effect due absence of persistent errors - slight oversteering.
I gain is too small - no effect in our case since we don't have persistent errors.
D gain is too big - jerky steering.
D gain is too small - cannot compensate P oversteering, oscillations last longer.

Describe how the final hyperparameters were chosen.

Final parameters (P = 0.0837465, I = 2.0899e-06, D = 1.44656) were chosen after running twiddle for many iterations (I've lost the number of iterations, it had been running for about an hour). Each iteration runs 2000 simulation steps (~ 20 secs of simulation), after that the simulation resets to initial state and next iteration of twiddle occurs. I stopped the optimization process after decided current parameters are good enough.

The vehicle must successfully drive a lap around the track.

Here's the video:

istepanov/CarND-PID-Control-Project