TODO: Complete README.md

Combined Dynamic Autoencoder (CDAE)

Dynamic Autoencoder for Learning Autonomous Driving Task

Prerequisites

The implementation is based on python3 and PyTorch. The requirements are included in the requirements file:

pip3 install -r requirements.txt

Datasets

The following datasets are used:

1. Simulated using CARLA simulator.
2. Udacity dataset with additional post-processing

TODO: Upload simulated dataset

TODO: Explain additional post-processing

Model Overview

Combined Dynamic Autoencoder consists of three main parts:

1. Autoencoder
2. Gated Recurrent Unit
3. Imitation / Reinforcement learning controller network (multilayer perceptron)

Training Procedure

TODO: Complete explanation

1) Train the autoencoder

python train_ae.py --config config/model/carla_ae.yaml

The default config is provided in the example.

2) Train the GRU

python train_rnn.py --config config/model/carla_rnn.yaml

Note that autoencoder is not frozen and fine-tuned as a part of GRU

3) Train the imitation learning network

python train_imitation.py --config config/model/carla_imitation.yaml