tuananhroman
commented
3 years ago - added staged training (class StagedRandomTask in tasks.py), new stages get triggered by callback (staged_train_callback.py)
- adapted reward fnc for continuous action space
- bug fix in obstacle_manager.py on function remove_obstacles()
- most recently trained CNN_NAVREP (trained on 2 maps in total about 1*10⁶ steps)
- better json handling/loading
- better structure in train_agent.py
- more detailed comments