The latest trained models with tensorboard logs

From this directory may be loaded but not automatically saved. Old models should only be replaced by new ones by hand.

numercial observations

Training with an environment (RoboSkate-v0) that has only the numerical observations as well as the optimal steering angle as input. Comparison with A2C, PPO and SAC.

Segmentation

Segmentation dim_2

• large FOV and trained with random crops
• it learned path width and position!!!