anasmatic/CarND-Term2-Project5-MPC

udacity CarND term2 Model Predictive Control

CarND-Controls-MPC

Self-Driving Car Engineer Nanodegree Program

---

In this project I decide the steering angle and speed of the simulator car , using MPC (Model Predictive Control)

default project installation :

1. mkdir build
2. cd build
3. cmake ..
4. make
5. ./mpc

my own Build Instructions

I was using Windows 10 and VisualStudio17

-to build this project using Bash for window :

navigate to projet
write cmd : mkdir build
then navigate to build
write cmd : cmake .. -G "Unix Makefiles" && make
write cmd : ./mpc

MPC

The Model

MPC is a model that optimizes the controls of a vehicle ( steering and speed ) by reducing the cost (error comparing to desired trajectory)

Any model cannot be 100% perfect, but MPC is close to perfection because it keeps refreshing the predection of the controls

MPC only excute the first set of controls, and then repeat the process to predect a new state.

1. Setup the duration of the trajectory T = N * dt while N = number of steps, and dt is constant delta-time of each step.

2. Setup the model : a. find intital [ x , y , psi , v , cte , epsi ], this is the initial state b. predect the next x , y , psi , v , cte , and epsi

       // Recall the equations for the model:
       // x_[t+1] = x[t] + v[t] * cos(psi[t]) * dt
       // y_[t+1] = y[t] + v[t] * sin(psi[t]) * dt
       // psi_[t+1] = psi[t] + v[t] / Lf * delta[t] * dt
       // v_[t+1] = v[t] + a[t] * dt
       // cte[t+1] = f(x[t]) - y[t] + v[t] * sin(epsi[t]) * dt
       // epsi[t+1] = psi[t] - psides[t] + v[t] * delta[t] / Lf * dt
   

   c. find the constrains : tune "delta" & "a" , which are related to stearing

3. Feed the resulting values to the cost function, which will result #N sets of controls

4. Only apply the first set of contols , and repeat the process again

Timestep Length and Elapsed Duration (N & dt)

we have to tune N & dt taking in concedration that more N means more processing. also less dt means more proccessing, and the aplication will run slow.

so we need the best value of N and dt that will cause cross track error to decrease to 0

I tried diffrent values of N, ex: N = 20 cause the car to take wrong decisions in straight lines, and green debug line was crasy ! Finaly I set these values to N = 8, and dt = 0.1

Polynomial Fitting and MPC Preprocessing

Third degree polynomial is required to set the cte and epsi values.

Model Predictive Control with Latency

Latency in simulator = 100ms , that was set to simulate natural conditions of course this will effect the car dicisions, because there's 100ms delay between predection and sending it to the server.

to handel this I multiply initial x , y , psi, and v by the latency value

px = px + v*cos(psi)*latency;
py = py + v*sin(psi)*latency;
//psi = psi + v*delta / Lf*latency;//Drives the car crazy !
v = v + acceleration*latency;

I hade troubles making psi works, so I ommitted it.

results :

[youtube video](https://www.youtube.com/watch?v=MgXbPQHJcXE)

car can take many loops around the track,

gif animation :

it can handle sharp turns and recover after them

gif animation :

only one problem , that is starts wiggeling at the start of the track, but stop wiggle right after it approches the first turn

gif animation :

video of one lap :