MultiAgentLearning / playground

PlayGround: AI Research into Multi-Agent Learning.
https://www.pommerman.com
Apache License 2.0
765 stars 217 forks source link

Bomb Lifetime is faulty #198

Closed dist1ll closed 5 years ago

dist1ll commented 5 years ago

After a player plants a bomb, the bomb gets ticked in the same step!

That means a bomb with lifetime 1 will explode immediately in the step where it's been planted. An easy fix would be to just set the bomb lifetime after planting a bomb to BOMB_LIFETIME + 1.

It would also be possible to put the bomb calculation at the start of the step method, however that complicates the bomb kicking movements a lot. I think the former solution is fine

cinjon commented 5 years ago

This is reasonable to me. Thanks for bringing it to our attention. Adjusted in 9e4d78f.

cinjon commented 5 years ago

Heads up @m2q that this change causes agents from the prior competitions to not be as good because they were built with a different setup. We'd like to maintain their efficacy for the next competition, and so I think a reasonable middle ground is to additionally change DEFAULT_BOMB_LIFE to be 9. Could you chime in if you see something wrong with that? Thanks.