genrobot2025/Gen-EgoData

--- license: cc-by-sa-4.0 task_categories: - robotics - reinforcement-learning language: - en - zh tags: - video - vla - dual-arm - embodied intelligence size_categories: - 100B<n<1T --- # Summary 1. This is a sample of our Gen EgoData. 2. The dataset contains 500 ego-centric samples with a total duration of 4.23h, including ego-SLAM pose information generated by our algorithm. The dataset covers household scenarios and includes 10 tasks, with 50 samples for each task, totaling 500 data samples. 3. More details about our ego data collection device can be found here: https://www.genrobot.ai/products/ego # Data Structure The dataset follows a hierarchical structure to organize robotic manipulation behaviors. Each sample is described using four semantic levels: Domain, Scenario, Task, and Skill. | Domain | Scenario | Task | Skill | |--------|----------|------|-------| | domestic_services | bedroom | clothing_organization | fold_clothes | | domestic_services | bedroom | clothing_organization | fold_pants | | domestic_services | kitchen | organization | organize_utensils | | domestic_services | kitchen | organization | organize_plates_and_bowls | | domestic_services | living_room | organization | organize_tabletop | | domestic_services | living_room | organization | organize_medications | | domestic_services | living_room | shoes_handling | organize_scattered_shoes | | domestic_services | living_room | shoes_handling | lace_up_shoes | | domestic_services | study | organization | organize_books | | domestic_services | study | organization | organize_desk | # Data Format and How to Load Data The dataset is stored in .mcap files. Each .mcap file corresponds to a single skill instance and contains all associated multimodal data streams, including observations, actions, and metadata. To load and parse the data, please use the official toolkit: [GitHub - genrobot-ai/das-datakit](https://github.com/genrobot-ai/das-datakit) # How to Vis Data https://monitor.genrobot.click/#/index # Contact Us Any questions, suggestions or desired data collection scenarios/skills are welcome during usage. Let’s co-build this project to digitize all human skills. X：https://x.com/GenrobotAI Linkin：https://www.linkedin.com/company/108767412/admin/dashboard/ Email：[e-mail]

搜集汇总

背景与挑战

背景概述

该数据集是一个面向家用机器人具身智能的自我中心多模态数据集，包含500个样本（总时长4.23小时），覆盖卧室、厨房、客厅、书房等家庭场景中的10个任务（如衣物整理、餐具摆放、鞋子整理等），每个任务提供50个带ego-SLAM位姿信息的演示数据。数据集采用四层语义结构（域、场景、任务、技能）组织，数据以.mcap格式存储，便于加载和用于训练视觉-语言-动作（VLA）模型。