| --- |
| task_categories: |
| - robotics |
| tags: |
| - manipulation |
| - imitation-learning |
| - visuomotor-policies |
| - short-term-memory |
| --- |
| |
| # ReMemBench: Scaling Short-Term Memory of Visuomotor Policies for Long-Horizon Tasks |
|
|
| [**[Project Page]**](https://shahrutav.github.io/short-term-memory/)   [**[Paper]**](https://huggingface.co/papers/2606.16178)   [**[GitHub]**](https://github.com/ShahRutav/ReMemBench) |
| |
| ReMemBench is a benchmark consisting of eight diverse household manipulation tasks spanning four categories of short-term memory, designed to foster general memory mechanisms in robotic visuomotor policies. Built upon [RoboCasa](https://robocasa.ai/), it provides expert teleoperated demonstrations for training and evaluating policies in long-horizon tasks. |
| |
| ## Task Categories |
| |
| Tasks are organized by memory type. Each task is provided with 50 expert demonstrations for training. |
| |
| | Task Name | Memory Category | Task Variants | |
| |-----------|----------------|---------------| |
| | **Retrieve Fruit**<br/>Remember fruit location (out of view). | **Spatial Memory**<br/>*Recall object locations* | `MemFruitInSinkLeftFar`<br/>`MemFruitInSinkRightFar` | |
| | **Retrieve Oil**<br/>Remember oil bottle location among distractors. | **Spatial Memory**<br/>*Recall object locations* | `MemRetrieveOilsFromCounterLL`<br/>`MemRetrieveOilsFromCounterLR`<br/>`MemRetrieveOilsFromCounterRL`<br/>`MemRetrieveOilsFromCounterRR` | |
| | **Cook Meat**<br/>Remember cooking duration while waiting. | **Prospective Memory**<br/>*Retain intentions over delay* | `MemHeatPot` | |
| | **Cook Meat and Vegetable**<br/>Remember multiple timed actions. | **Prospective Memory**<br/>*Retain intentions over delay* | `MemHeatPotMultiple` | |
| | **Wash and Return to Container**<br/>Remember which saucer (left/right) the fruit came from. | **Object-Associative Memory**<br/>*Recall associations* | `MemWashAndReturnLeft`<br/>`MemWashAndReturnRight` | |
| | **Wash and Return to Original Spot**<br/>Remember original countertop location. | **Object-Associative Memory**<br/>*Recall associations* | `MemWashAndReturnSameLocation` | |
| | **Microwave Breadsticks**<br/>Remember count of breadsticks moved. | **Object-Set Memory**<br/>*Maintain/update sets* | `MemPutKBreadInMicrowave` | |
| | **Relocate Bowls**<br/>Remember count of bowls among distractors. | **Object-Set Memory**<br/>*Maintain/update sets* | `MemPutKBowlInCabinet` | |
| |
| ## Data Downloading |
| |
| You can download the dataset using the `huggingface-cli`: |
| |
| ```bash |
| huggingface-cli download Rutav/ReMemBench-Dataset \ |
| --repo-type dataset \ |
| --local-dir ReMemBench-Dataset \ |
| --local-dir-use-symlinks False |
| ``` |
| |
| ## Dataset Structure |
| |
| The dataset is organized by task name, with each task containing demonstration sessions in HDF5 format. |
| |
| ### File Structure |
| ``` |
| ReMemBench-Dataset/ |
| ├── MemFruitInSinkLeftFar/ |
| ├── MemHeatPot/ |
| │ ├── [timestamp]/ |
| │ │ ├── demo.hdf5 |
| │ │ └── demo_im128.hdf5 # Image version |
| ├── ... |
| └── task_embeds_clip_v3.pickle |
| ``` |
| |
| ### HDF5 File Structure (`demo_im128.hdf5`) |
| - **`actions`**: (T, 12) - [7D arm, 4D base, 1D mode] |
| - **`obs`**: |
| - `robot0_joint_pos_cos` / `robot0_joint_pos_sin`: (T, 7) - Joint position encoding |
| - `robot0_gripper_qpos`: (T, 2) - Gripper position |
| - `robot0_agentview_center_image`: (T, 128, 128, 3) - RGB third-person view |
| - `robot0_eye_in_hand_image`: (T, 128, 128, 3) - RGB eye-in-hand view |
| - **`rewards`**, **`dones`**, **`states`**: Standard simulation signals |
| |
| ## Exploring the Data |
| |
| To visualize demonstrations, use the `replay_dataset.py` script from the [official repository](https://github.com/ShahRutav/ReMemBench): |
| |
| ```bash |
| python robocasa/scripts/replay_dataset.py \ |
| --hdf5_path ReMemBench-Dataset/MemHeatPot/[timestamp]/demo_im128.hdf5 \ |
| --episode_idx 0 \ |
| --render |
| ``` |
| |
| ## Converting to LeRobot Dataset Format |
| |
| A conversion script is provided in the repository to port the data to [LeRobot](https://github.com/huggingface/lerobot) format: |
| |
| ```bash |
| python robocasa/scripts/port_to_lerobot.py \ |
| --dataset_path ReMemBench-Dataset/MemHeatPot/[timestamp]/demo_im128.hdf5 \ |
| --repo_name your_hf_username/MemHeatPot |
| ``` |
| |
| ## Citation |
| |
| ```bibtex |
| @article{shah2024scaling, |
| title={Scaling Short-Term Memory of Visuomotor Policies for Long-Horizon Tasks}, |
| author={Shah, Rutav and others}, |
| journal={arXiv preprint arXiv:2606.16178}, |
| year={2024} |
| } |
| ``` |
