Goal output when training with reward shaping wrappers

[Yutai Zhou]

I am training agents using both the scoring and checkpoints wrapper. I would like to understand how the different wrappers used can impact the reward logging.

In the multiagent setting, the default behavior is that every agent shares the same reward. This means if the team advances 10% of field length, every agent gets 0.1 point, if the team scores a goal, every agent gets 1 point. Is that correct?

If checkpoints wrapper is used, does the logged reward include the checkpoints rewards? If so, how can I log a goal-based only reward metric in addition to a goal+distance based reward metric?

[stanczyk]

Please follow https://github.com/google-research/football/issues/210