CliffWalking-v1
A grid navigation benchmark for comparing short routes with routes that avoid a costly cliff edge.
Published by Farama Foundation
Setup and evaluation
An agent must reach the goal while staying out of cells that send it back to the start. This benchmark is useful for seeing how training behavior and exploration affect the paths learned by algorithms such as SARSA and Q-learning.
Install with pip install gymnasium and create gymnasium.make('CliffWalking-v1'). Compare paths, falls, and episode return when evaluating different policies.
- Framework
- Gymnasium
- Access
- Local Gymnasium environment. No account required.