robot-control
duobench_raw
DuoBench Raw Dataset
This is the raw dataset collected for the FR3 Duo Benchmark DuoBench.
The format is a raw parquet format that includes additional information such as robot state (real) and sim state (sim) which are filtered out for training.
If you are looking for the converted hugging face dataset ready for training, see this dataset.
You can also manually convert this raw dataset to the hugging face format with the following command:
# install rcs:… See the full description on the dataset page: https://huggingface.co/datasets/RobotControlStack/duobench_raw.rcs_real_sim_async_all_500
Simple Mixed (sim&real) Franka Pick-Cube Dataset in LeRobot Format
This repository contains the franka dataset in parquet format used to train PiZero in the paper Robot Control Stack: A Lean Ecosystem for Robot Learning at Scale.
This dataset is in LeRobot format and thus ready to use for training OpenPi, lerobot and other models which support this common format.
The Robot Control Stack (RCS) is a flexible Gymnasium wrapper-based robot control interface made for robot learning… See the full description on the dataset page: https://huggingface.co/datasets/RobotControlStack/rcs_real_sim_async_all_500.duobench
DuoBench LeRobot Dataset
This is the dataset collected for the FR3 Duo Benchmark DuoBench in LeRobot format ready for training.
The dataset is a stripped down version from the raw dataset and only contains data required for VLA training.
Checkout the raw dataset if you are interested e.g. in franka states (real) and sim states.
The raw dataset repo also shows how to convert the data into this lerobot version.
Project page: https://duobench.github.io/
DuoBench Code:… See the full description on the dataset page: https://huggingface.co/datasets/RobotControlStack/duobench.rcs_utn_green_box_v1
Simple Teleoperated Franka Pick-Cube Dataset
This repository contains the franka dataset in parquet format used to train Octo, OpenVLA and PiZero in the paper Robot Control Stack: A Lean Ecosystem for Robot Learning at Scale.
The Robot Control Stack (RCS) is a flexible Gymnasium wrapper-based robot control interface made for robot learning and specifically Vision-Language-Action (VLA) models.
It unifies MuJoCo simulation and real world robot control with four supported robots:… See the full description on the dataset page: https://huggingface.co/datasets/RobotControlStack/rcs_utn_green_box_v1.rcs_sim_async_green_box_tactile_v1
Simulated Franka Pick-Cube Tactile Dataset in LeRobot Format
The dataset was generated using the Robot Control Stack (RCS). It is in LeRobot format and thus ready-to-train. The task is a simple "grasp the cuboid".
It is created by a hardcoded trajectory planner.
It contains tactile images from digit sensors attached to the fingers, rendered by tacto.
RCS is a flexible Gymnasium wrapper-based robot control interface made for robot learning and specifically Vision-Language-Action… See the full description on the dataset page: https://huggingface.co/datasets/RobotControlStack/rcs_sim_async_green_box_tactile_v1.rcs_utn_green_box_lerobot
Simple Teleoperated Franka Pick-Cube Dataset in LeRobot Format
This repository contains the franka dataset in parquet format used to train PiZero in the paper Robot Control Stack: A Lean Ecosystem for Robot Learning at Scale.
This dataset is in LeRobot format and thus ready to use for training OpenPi, lerobot and other models which support this common format.
The Robot Control Stack (RCS) is a flexible Gymnasium wrapper-based robot control interface made for robot learning and… See the full description on the dataset page: https://huggingface.co/datasets/RobotControlStack/rcs_utn_green_box_lerobot.