SimpleMemVLA
Collection
A Simple but Effective Native-Video Memory for Vision-Language-Action Models • 12 items • Updated • 1
1.6k episodes · 20 fps · 2 cameras · 256×256 av1
SimpleMemVLA Datasets
This repository contains the LeRobot-v3 datasets used for training and evaluating SimpleMemVLA: A Simple but Effective Native-Video Memory for Vision-Language-Action Models.
- Paper: arXiv:2609.05533
- Project page: Hugging Face Collection
- Code: github.com/wadeKeith/SimpleMemVLA
Each dataset follows the LeRobot v3 format and includes per-frame subtask annotations, action labels, camera observations, and normalization statistics.
| Benchmark | Hugging Face Dataset |
|---|---|
| RMBench | yinchenghust/rmbench_lerobot |
| RoboMME | yinchenghust/robomme_lerobot |
| MIKASA-Robo | yinchenghust/mikasa_lerobot |
| RoboMemArena | yinchenghust/robomemarena_lerobot |
| LIBERO | yinchenghust/libero_lerobot |
Datasets are also mirrored on ModelScope under the keithyc account.
Download
Example for the LIBERO dataset:
huggingface-cli download yinchenghust/libero_lerobot --repo-type dataset --local-dir data/datasets/yinchenghust/libero_lerobot
For full training and evaluation instructions, please see the GitHub repository.
- Downloads last month
- 1,179