LePlanner

Official checkpoints for LePlanner: Amortized Iterative Control in Frozen Latent World Models.

[Paper] [Code]

Results

Environment	Success
PushT	98% (rh=1)
Reacher	100%
TwoRooms	100%

Download

from huggingface_hub import hf_hub_download

env = "reacher"  # or "pusht" / "tworooms"

controller = hf_hub_download("nottygian/Leplanner", f"checkpoints/Leplanner Controller/{env}/controller.pt", repo_type="model")
density   = hf_hub_download("nottygian/Leplanner", f"checkpoints/Leplanner Controller/{env}/density.pt", repo_type="model")
world_model = hf_hub_download("nottygian/Leplanner", f"checkpoints/base_world_models/{env}/weights.pt", repo_type="model")

Structure

checkpoints/
├── Leplanner Controller/             # Trained controllers + density
│   ├── pusht/
│   │   ├── controller.pt             # 98% success
│   │   └── density.pt
│   ├── reacher/
│   │   ├── controller.pt             # 100% success
│   │   └── density.pt
│   └── tworooms/
│       ├── controller.pt             # 100% success
│       └── density.pt
├── base_world_models/                # Frozen LeWM world models
│   ├── pusht/config.json + weights.pt
│   ├── reacher/config.json + weights.pt
│   └── tworooms/config.json + weights.pt
└── results/eval/results.jsonl       # Raw evaluation logs

code/                                 # Training & evaluation scripts

Citation

@article{leplanner,
  title={LePlanner: Amortized Iterative Control in Frozen Latent World Models},
  author={[Authors]},
  journal={[Venue]},
  year={[Year]},
}