CARL Mario Environment¶
The ToadGAN environment provides context-guided procedural generation inspired by Nintendo’s super Mario Bros. The agent has to guide a character through levels including jumps, coins and enemies generated according to context definition. The level is solved as soon as the agent arrives at the end. Its actions include moving forwards or backwards and jumping.
The generation can be controlled by the following context features:
Context Feature |
Default |
Bounds |
|---|---|---|
level_index |
0 |
(None, None, ‘categorical’, array([ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13])) |
noise |
None * |
(-1.0, 1.0, <class ‘float’>) |
mario_state |
0 |
(None, None, ‘categorical’, [0, 1, 2]) |
mario_inertia |
0.89 |
(0.5, 1.5, <class ‘float’>) |