yjang43/lp2-pointmaze
Updated • 19
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Datasets for the PointMaze U-maze navigation task used in LP² (Latent Projection for
Latent Planning), collected in gymnasium-robotics PointMaze_UMaze-v3 with the goal marker
hidden, using our stable-worldmodel fork.
d4rl_pointmaze_u_maze_random.tar.zst — 10,000 random-policy episodes. The world model
and the state dynamics are fit on this set.d4rl_pointmaze_u_maze_expert.tar.zst — 13,210 D4RL expert episodes, re-rendered through
the same environment. Evaluation start and goal pairs are drawn from this set.Extract both into $STABLEWM_HOME/datasets.