Migrate env to racecar gym

[rohanjaya]

Changes

We are working on getting the Dreamer v3 to work with the racecar_gym environment with dream.py as the entry point. Currently it reaches the evaluation phase (line 264 in dream.py), which calls simulate() from tools.py. An error occurs when calling logger.write(), at the end of the simulation, that must be debugged.

Refer to dream.py Main() and tools.py simulate() as the primary, chronological logic.

Motivation and Context

Ensure that we are able to begin training our agent using a reinforcement environment. Once completed, we will be able to port to the vehicle.

How Has This Been Tested?

This has primarily been tested by calling and debugging dream.py.

Screenshots (if appropriate):

Needed changed

• Debug issue with the logger in tools.simulate.
• See if the training step works
• Find out how to use the logger to monitor training performance
• Determine which observation keys need to saved to cache (save_episode function)

[waterkimchi]

good