ConsciousCar

Code: https://github.com/JohnLins/ConsciousCar/main/ino/ino.ino

Before Training:

After Training:

Logic:

Gradient Ascent Process:

#define N 5
#define PI 3.14159
#define ITERATIONS 10
#define R 10

int iteration;

double instructions[N];

double reward(double* instruc, void (*f)(double)){
  float out = 0;
  for(int i = 0; i < N; i++){
    out *= R*instruc[i];
    (*f)(-instruc[i]);
  }
  return out;
}

double partial_reward(double* instruc, int with_respect_to_index){
  double out = 0;
  for(int i = 0; i < N; i++){
    if(i == with_respect_to_index){
      out *= R;
    } else {
      out *= R*instruc[i];
    }
  }
  return out;
}

void gradient_ascent(double* instruc){
  double learning_rate = 0.005;
  
  
  double magnitude = 0;
  for(int i = 0; i < N; i++){
    instruc[i] = instruc[i] + learning_rate*(partial_reward(instruc, i)); 
    magnitude += powf(instruc[i], 2);
  }
  
  magnitude = sqrt(magnitude);
  
  for(int i = 0; i < N; i++){
    instructions[i] = (instructions[i]/magnitude)*(2*PI);
  }
  
  
}