ShareRobot Dataset

ShareRobot, a high-quality heterogeneous dataset that labels multi-dimensional information, including task planning, object affordance, and end-effector trajectory, effectively enhancing various robotic capabilities.

Overview of ShareRobot Dataset

For planning, we have 51,403 episodes and each with 30 frames. In the process of data generation, we design 5 different templates for each of the 10 question types in RoboVQA [1]. In the process of data generation, we randomly select 2 templates of each question type to generate question-answer pairs for every instance. This process transforms 51,403 instances into 1,027,990 question-answer pairs, with annotators monitoring data generation to maintain the dataset’s integrity.

For Affordance, we have 6,522 images and each with affordance areas aligned with an instruction.

For Trajectory, we have 6,870 images and each with at least 3 {x, y} coordinates aligned with an instruction.

Dataset Sources

ShareRobot dataset contains 23 original datasets from Open X-Embodiment dataset [2], 12 embodiments and 107 types of atomic tasks.

Raw Dataset for Planning

Raw Dataset	Number of Raws
nyu_door_opening_surprising_effectiveness	421
bridge	15738
dlr_edan_shared_control_converted_externally_to_rlds	63
utokyo_xarm_pick_and_place_converted_externally_to_rlds	92
cmu_stretch	10
asu_table_top_converted_externally_to_rlds	109
dlr_sara_pour_converted_externally_to_rlds	51
utokyo_xarm_bimanual_converted_externally_to_rlds	27
robo_set	18164
dobbe	5200
berkeley_autolab_ur5	882
qut_dexterous_manpulation	192
aloha_mobile	264
dlr_sara_grid_clamp_converted_externally_to_rlds	40
ucsd_pick_and_place_dataset_converted_externally_to_rlds	569
ucsd_kitchen_dataset_converted_externally_to_rlds	39
jaco_play	956
utokyo_pr2_opening_fridge_converted_externally_to_rlds	64
conq_hose_manipulation	56
fmb	7836
plex_robosuite	398
utokyo_pr2_tabletop_manipulation_converted_externally_to_rlds	189
viola	44

Raw Dataset for Affordance

Raw Dataset	Number of Raws
utokyo_pr2_tabletop_manipulation_converted_externally_to_rlds	24
utokyo_xarm_pick_and_place_converted_externally_to_rlds	23
ucsd_kitchen_dataset_converted_externally_to_rlds	10
ucsd_pick_and_place_dataset_converted_externally_to_rlds	112
nyu_door_opening_surprising_effectiveness	85
jaco_play	171
bridge	2610
utokyo_pr2_opening_fridge_converted_externally_to_rlds	12
asu_table_top_converted_externally_to_rlds	24
viola	1
berkeley_autolab_ur5	122
aloha_mobile	23
conq_hose_manipulation	1
dobbe	717
fmb	561
plex_robosuite	13
qut_dexterous_manpulation	16
robo_set	1979
dlr_edan_shared_control_converted_externally_to_rlds	18
Summary	6522

Raw Dataset for Trajectory

Raw Dataset	Number of Raws
utokyo_pr2_tabletop_manipulation_converted_externally_to_rlds	35
utokyo_xarm_pick_and_place_converted_externally_to_rlds	36
ucsd_kitchen_dataset_converted_externally_to_rlds	19
dlr_sara_grid_clamp_converted_externally_to_rlds	1
ucsd_pick_and_place_dataset_converted_externally_to_rlds	109
nyu_door_opening_surprising_effectiveness	74
jaco_play	175
utokyo_xarm_bimanual_converted_externally_to_rlds	7
bridge	2986
utokyo_pr2_opening_fridge_converted_externally_to_rlds	12
asu_table_top_converted_externally_to_rlds	22
berkeley_autolab_ur5	164
dobbe	759
fmb	48
qut_dexterous_manpulation	29
robo_set	2374
dlr_sara_pour_converted_externally_to_rlds	3
dlr_edan_shared_control_converted_externally_to_rlds	17
Summary	6870

Data Format

Planning

{
 "id"{
        "id": 0,
        "task": "Future_Prediction_Task",
        "selected_step": 3,
        "conversations": [
            {
                "from": "human",
                "value": "<image 0-25> After <move the grasped banana towards the mug>, what's the most probable next event?"
            },
            {
                "from": "gpt",
                "value": "<place the banana into the mug>"
            }
        ],
        "image": [
            "/path/to/image_0-25"
        ]
    }        
}

     

Affordance

{

        "id": 2486,
        "meta_data": {
            "original_dataset": "bridge",
            "original_width": 640,
            "original_height": 480
        },
        "instruction": "place the red fork to the left of the left burner",
        "affordance": {
            "x": 352.87425387858815,
            "y": 186.47871614766484,
            "width": 19.296008229513156,
            "height": 14.472006172134865
    }

Visualize Code

import json
import os
import cv2
import numpy as np

img_dir = '/path/to/your/original/images/dir'
affordance_json = '/path/to/your/affordances/json'
output_img_dir = '/path/to/your/visualized/images/dir'

with open(affordance_json, 'r') as f:
    data = json.load(f)
    for item in data:
        filepath = os.path.join(img_dir, item['id'])

        image = cv2.imread(filepath)
        color = (255, 0, 0)
        thickness = 2

        x_min,y_min = item['affordance']['x'], item['affordance']['y']
        x_max,y_max = item['affordance']['x']+item['affordance']['width'], item['affordance']['y']+item['affordance']['height']

        # 定义矩形的四个顶点坐标
        pts = np.array([
            [x_min, y_min],  # 左上角
            [x_max, y_min],  # 右上角
            [x_max, y_max],  # 右下角
            [x_min, y_max]   # 左下角
        ], dtype=np.float32)

        # 绘制矩形框
        cv2.polylines(image, [pts.astype(int)], isClosed=True, color=color, thickness=thickness)

        # 获取相对路径并拼接目标路径
        relative_path = os.path.relpath(filepath, img_dir)  # 获取相对于 img_dir 的相对路径
        output_img_path = os.path.join(output_img_dir, relative_path)  # 拼接目标路径

        # 创建目标文件夹
        output_directory = os.path.dirname(output_img_path)
        if not os.path.exists(output_directory):
            os.makedirs(output_directory)

        # 打印调试信息
        print(f"Input filepath: {filepath}")
        print(f"Output image path: {output_img_path}")
        print(f"Output directory: {output_directory}")

        # 保存图像
        cv2.imwrite(output_img_path, image)

Trajectory

{
        "id": 456,
        "meta_data": {
            "original_dataset": "bridge",
            "original_width": 640,
            "original_height": 480
        },
        "instruction": "reach for the carrot",
        "points": [
            [
                265.45454545454544,
                120.0
            ],
            [
                275.1515151515152,
                162.42424242424244
            ],
            [
                280.0,
                213.33333333333331
            ],
            [
                280.0,
                259.3939393939394
            ]
        ]
    },

Visualize Code

import json
import os
from PIL import Image, ImageDraw

trajectory_final = '/path/to/your/trajectory_json'
img_dir = '/path/to/your/original/images/dir'
output_img_dir = '/path/to/your/visualzed/images/dir'

with open(trajectory_final, 'r') as f:
    data = json.load(f)
    for item in data:
        filepath = os.path.join(img_dir, item['id'])
        points = item['points']

        image = Image.open(filepath).convert("RGB")  # 确保图像是 RGB 模式
        draw = ImageDraw.Draw(image)  # 创建绘图对象
        # 定义颜色和线宽
        color = (255, 0, 0)  # 红色 (RGB 格式)
        thickness = 2


        scaled_points = [
                (point[0], point[1])
                for point in points
            ]
        # 按照顺序连接相邻的点
        for i in range(len(scaled_points) - 1):
            draw.line([scaled_points[i], scaled_points[i + 1]], fill=color, width=thickness)

        # 获取相对路径并拼接目标路径
        relative_path = os.path.relpath(filepath, img_dir)
        output_img_path = os.path.join(output_img_dir, relative_path)

        # 创建目标文件夹
        output_directory = os.path.dirname(output_img_path)
        if not os.path.exists(output_directory):
            os.makedirs(output_directory)

        # 打印调试信息
        print(f"Input filepath: {filepath}")
        print(f"Output image path: {output_img_path}")
        print(f"Output directory: {output_directory}")

        # 保存图像
        image.save(output_img_path)

Evaluation

Reference

[1] Pierre Sermanet, Tianli Ding, Jeffrey Zhao, Fei Xia, Debidatta Dwibedi, Keerthana Gopalakrishnan, Christine Chan,Gabriel Dulac-Arnold, Sharath Maddineni, Nikhil J Joshi,et al. Robovqa: Multimodal long-horizon reasoning forrobotics. In ICRA, pages 645–652, 2024.

[2] Abby O’Neill, Abdul Rehman, Abhinav Gupta, AbhiramMaddukuri, Abhishek Gupta, Abhishek Padalkar, AbrahamLee, Acorn Pooley, Agrim Gupta, Ajay Mandlekar, et al.Open x-embodiment: Robotic learning datasets and rt-xmodels. arXiv preprint arXiv:2310.08864, 2023.

Citation

@article{ji2025robobrain,
  title={RoboBrain: A Unified Brain Model for Robotic Manipulation from Abstract to Concrete},
  author={Ji, Yuheng and Tan, Huajie and Shi, Jiayu and Hao, Xiaoshuai and Zhang, Yuan and Zhang, Hengyuan and Wang, Pengwei and Zhao, Mengdi and Mu, Yao and An, Pengju and others},
  journal={arXiv preprint arXiv:2502.21257},
  year={2025}
}