Sokoban dataset with explicit move-only actions.
Easy split: remapped from novastar111/sokoban_v3 (sokoban_short_v0/train|test)
Hard split: remapped from novastar111/swanlake_hard (trajectories/sokoban_hard/train|test)
('move', 1): move up
('move', 2): move down
('move', 3): move left
('move', 4): move right
('stop', 'stop'): end the episode
If a crate is directly in the selected direction and the next cell is free, the… See the full description on the dataset page:
https://huggingface.co/datasets/novastar111/s_v5_move_only.
