Create dataset card for MemoryRTBench / MemoAct
#1
by nielsr HF Staff - opened
README.md
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
task_categories:
|
| 3 |
+
- robotics
|
| 4 |
+
tags:
|
| 5 |
+
- robotics
|
| 6 |
+
- embodied-ai
|
| 7 |
+
---
|
| 8 |
+
|
| 9 |
+
# MemoryRTBench
|
| 10 |
+
|
| 11 |
+
This repository contains the dataset for **MemoryRTBench**, introduced in the paper [MemoAct: Atkinson-Shiffrin-Inspired Hierarchical Memory-Augmented Policy for Robotic Manipulation](https://huggingface.co/papers/2603.18494).
|
| 12 |
+
|
| 13 |
+
[Project Page](https://tlf-tlf.github.io/MemoActPage/) | [Paper](https://huggingface.co/papers/2603.18494)
|
| 14 |
+
|
| 15 |
+
## Dataset Description
|
| 16 |
+
|
| 17 |
+
MemoryRTBench is constructed based on RoboTwin 2.0, comprising 6 manipulation tasks designed to systematically evaluate policy memory capabilities across three key dimensions:
|
| 18 |
+
- **Sequential memory**
|
| 19 |
+
- **Spatial memory**
|
| 20 |
+
- **Episodic memory**
|