With Udacity's open source simulator available, I tried to implement Nvidia's End to End Self Driving research paper. The implementation of the model is the same as that given in the paper.
1 normalization layer 5 conv layers 3 fully connected layers
video link -- https://www.youtube.com/watch?v=Tp0A63jrhsg
Further steps
- Training the network for predicting steering and throttle together. Currently, I have 2 different models for steering and throttle prediction.
- Implement conditional Imitation Learning on CARLA
- Try using RNN on the extracted features and compute the steering values