Closed RobertTLange closed 2 years ago
The MinAtar environments are implemented in plain NumPy and can capture cool dynamics. They should be "jax-able" -- give it a go. Here is a TODO-list:
[x] Asterix
[x] Breakout
[ ] Seaquest
[x] Space Invaders
[x] Freeway
Many of them are non-deterministic/require sampling of objects based on open slots. Come up with a good test for these special cases.
Also add a visualisation wrapper. This can make debugging a lot easier.
Make a decision what the allowed actions are. 'n' for example encodes a no-op. This has to respected!
The MinAtar environments are implemented in plain NumPy and can capture cool dynamics. They should be "jax-able" -- give it a go. Here is a TODO-list:
[x] Asterix
[x] Breakout
[ ] Seaquest
[x] Space Invaders
[x] Freeway
Many of them are non-deterministic/require sampling of objects based on open slots. Come up with a good test for these special cases.
Also add a visualisation wrapper. This can make debugging a lot easier.
Make a decision what the allowed actions are. 'n' for example encodes a no-op. This has to respected!