Algorithm 1 Pseudocode for Predicting the Steering Angle

#Nvidia Model

Lambda: Output shape: 400 × 600 × 3

Image normalization to avoid saturation and make gradients work better.

#2D Convolution Neural Network for handle features.

Convolution1: 5 × 5, filter: 24, strides: 2 × 2, activation: ELU

Convolution2: 5 × 5, filter: 24, strides: 2 × 2, activation: ELU

Convolution3: 5 × 5, filter: 48, strides: 2 × 2, activation: ELU

Convolution4: 3 × 3, filter: 64, strides: 1 × 1, activation: ELU

Convolution5: 3 × 3, filter: 64, strides: 1 × 1, activation: ELU

#Dropout avoids overfitting

Drop out (0.5)

#Fully Connected Layer for predicting the steering angle.

Fully connected 1: neurons: 100, activation: ELU

Fully connected 2: neurons: 50, activation: ELU

Fully connected 3: neurons: 10, activation: ELU

Fully connected 4: neurons: 1 (output)

model.compile(Adam(lr = 0.0001), loss=’mse’)

return model