FrozenLake Environment
Overview
The FrozenLake environment presents a challenging navigation task where each agent move may randomly slip to a side direction, requiring the agent to adapt under uncertainty.
Example Cases
Case 1: Spinning on Ice
Step 1:
State:
P: Player, _L: Blank Space, G: Goal, O: Hole
___G
O___
____
_P__
LLM Response:
Reasoning:
Forbid the slippery ice and try to the target.
Action:
Step 2:
State:
P: Player, _L: Blank Space, G: Goal, O: Hole
___G
OP__
____
____
LLM Response:
Reasoning:
Forbid the slippery ice and try to the target.
Action:
Step 3:
State:
P: Player, _L: Blank Space, G: Goal, O: Hole
P__G
O___
____
____
LLM Response:
Reasoning:
Forbid the slippery ice and try to the target.
Action:
Step 4:
State:
P: Player, _L: Blank Space, G: Goal, O: Hole
_P_G
O___
____
____
LLM Response:
Reasoning:
Forbid the slippery ice and try to the target.
Action:
Step 5:
State:
P: Player, _L: Blank Space, G: Goal, O: Hole
___G
O_P_
____
____
LLM Response:
Reasoning:
Forbid the slippery ice and try to the target.
Action:
Step 6:
State:
P: Player, _L: Blank Space, G: Goal, O: Hole
___P
O___
____
____
End of trajectory
Case 2: Trapped at Bottom
Step 1:
State:
P: Player, _L: Blank Space, G: Goal, O: Hole
__O_
__O_
___G
P__O
LLM Response:
Reasoning:
Forbid the slippery ice and try to the target.
Action:
Step 2:
State:
P: Player, _L: Blank Space, G: Goal, O: Hole
__O_
__O_
___G
P__O
LLM Response:
Reasoning:
Action:
Step 3:
State:
P: Player, _L: Blank Space, G: Goal, O: Hole
__O_
__O_
___G
P__O
LLM Response:
Reasoning:
Constant reward environment.
Action:
Step 4:
State:
P: Player, _L: Blank Space, G: Goal, O: Hole
__O_
__O_
___G
_P_O
LLM Response:
Reasoning:
Keep trying, haven't won yet.
Action:
Step 5:
State:
P: Player, _L: Blank Space, G: Goal, O: Hole
__O_
__O_
__PG
___O
LLM Response:
Reasoning:
Don't give up, reward is steady.
Action:
Step 6:
State:
P: Player, _L: Blank Space, G: Goal, O: Hole
__O_
__O_
___P
___O
End of trajectory