Remember to be Curious — Pretrained Checkpoints
Pretrained checkpoints for:
Remember to be Curious: Episodic Context and Persistent Worlds for 3D Exploration
Lily Goli, Justin Kerr, Daniele Reda, Alec Jacobson, Andrea Tagliasacchi, Angjoo Kanazawa
University of Toronto · UC Berkeley · Wayve · Vector Institute · Simon Fraser University
[Project page] · [Code]
Model description
A reinforcement learning agent that learns to explore 3D indoor scenes using only an RGB camera. The policy uses a DINO ViT-B/8 visual encoder with a sliding-window transformer backbone (KV-cached), trained with PPO. Intrinsic reward is derived from real-time Gaussian Splatting (GSplat) reconstruction quality.
Checkpoints
| File | Description |
|---|---|
explorer.pt |
Main exploration policy, trained on HM3D |
apple_finetuned.pt |
Apple-picking fine-tune (from explorer.pt) |
image_goal_finetuned.pt |
Image-goal navigation fine-tune (from explorer.pt) |
Usage
See the code repository for full installation and evaluation instructions.
Training data
Trained on Habitat-Matterport 3D (HM3D) scenes using the Habitat-Sim simulator.