- End-to-end neural network
- Run the simulator with the "perfect" driver. Collect the images and the steering angle.
- Create a network based on VGG
- Error function.
- Training
- Actor/Critic
- Segmentation
- Simulated Car Racing Championship Competition Software Manual
- Gym-TORCS
- Deep Reinforcement Learning approach to Autonomous Navigation
- End-to-End Deep Learning for Self-Driving Cars
- Teaching a Machine to Steer a Car
- Community Steering Models
- Reinforcement learning with docker and torcs
- If you want to compile for yourself, use /home/gym_torcs/vtorcs-RL-color/src/libs/raceengineclient/raceengine.cpp
- to get from little screen to big screen: down, down, enter; down, enter; right, right, enter; enter
- run with command line --random option to enable random track selector
- image augmentation, road segmentation
- test network and solver
- adjust hyperparameters