JD EgoLive Egocentric Human Video Corpus
JoyAI-RA 0.5论文描述的超过20,000小时、60FPS第一视角人类视频语料，用于潜动作世界模型和跨数据VLA预训练。官方项目页只链接京东云通用具身数据交易平台，未解析到EgoLive专属文件页、清单、直接下载或数据许可证；论文另述约53,000小时更广泛的人类视频语料，不能与EgoLive规模混写。
- 开放状态
- 承诺/部分开放
- 许可证
- No EgoLive dataset license published
- 商业使用
- not established
- 人工评审结论
- verified_described_large_scale_human_egocentric_corpus_not_directly_released
- 机器人
- human egocentric video; AgiBot G1 used for downstream physical evaluation
- 任务
- large-scale daily human manipulation activities; exact task taxonomy not released
- 模态
- egocentric RGB video、implicitly inferred latent action
- 格式
- not directly released; general marketplace entry only
- 采集/生成方式
- large-scale human first-person video collection; capture hardware, consent and privacy procedures not publicly specified
- 已知问题
- No dedicated dataset page, file manifest, direct download, license, privacy ledger, wearer/task counts, action calibration or independent use report; do not conflate the 20K-hour EgoLive claim with the broader 53K-hour human-video corpus.