yamoling / lle Goto Github PK

State: At the moment, the only kind of reward that exists is a collective team reward.
Observation: It could be useful in some cases (adversarial ones, for instance) to know the reward of each agent.

Idea: enable the user to choose which kind of reward to use

When an agent dies, it should no longer block the lasers of its own colour.
The test test_dead_agent_does_not_block_the_laser tests this behaviour and is currently not passing.

State: today, when any agent dies, world.done is set to true.
Observation: in adversarial setups, such behaviour makes no sense. Even in cooperative scenarios, we could want the agents to respawn upon death.

Idea: Implement a "death strategy" that can be chosen by the user.