[RL-baseline] Model v5, experiment #3

[ziritrion]

Action set #2 was chosen for this experiment:
[0.0, 0.0, 0.0], # no action
[0.0, 0.8, 0.0], # throttle
[0.0, 0.0, 0.6], # break
[-0.9, 0.0, 0.0], # left
[0.9, 0.0, 0.0], # right

The Running Reward oscillated between 200 up until the 20k episode mark, where it suddenly dropped. Since the other experiments had experienced long intervals of low reward during training, I decided to train up to 30k episodes to see if the RR could improve, but the result was disappointing.

Final Running Reward was 72, with an achieved max of 379 around the 4k episode mark.

Tensorboard screenshots below:

Sample video below:
https://user-images.githubusercontent.com/1465235/113551628-44479480-95f5-11eb-9a6f-dc3a6e61ced5.mp4