name stringlengths 3 82 | kind stringclasses 7
values | released stringclasses 112
values | venue stringclasses 190
values | status stringclasses 5
values | scope stringclasses 2
values | year int64 2.01k 2.03k | url stringlengths 19 150 | paper stringlengths 26 109 ⌀ | code stringclasses 125
values | license stringclasses 44
values | scale stringlengths 9 350 | tasks stringlengths 7 132 | modalities stringlengths 7 118 ⌀ | created_by stringlengths 4 353 ⌀ | released_by stringclasses 98
values | verified_at stringdate 2026-06-15 00:00:00 2026-10-29 00:00:00 ⌀ | project stringclasses 64
values | data stringclasses 158
values | weights stringclasses 2
values | benchmark stringclasses 1
value | leaderboard stringclasses 2
values | mirror stringclasses 1
value | license_url stringclasses 195
values | access stringclasses 41
values | access_url stringclasses 8
values | release_note stringlengths 44 296 ⌀ | scope_note stringclasses 163
values | derived_from stringclasses 37
values |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
SOMA | benchmark | 2026-10 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2610.36384 | https://arxiv.org/abs/2610.36384 | null | null | 1,632 human-verified questions on body configuration, motion dynamics, and temporal reasoning from first-person video | body-motion-understanding; temporal-reasoning; egocentric-vqa | egocentric-video; body-motion; video-qa | Matea Tashkovska, Nael Ouerghemi, Haozhe Qi, and Alexander Mathis | SOMA author team (paper) | 2026-10-29 | https://cnai.epfl.ch/soma | null | null | null | null | null | null | null | null | The paper announces benchmark, training data, and code; downloadable artifacts were not verified. | null | null |
ProactEgoMem | benchmark | 2026-10 | arXiv | benchmark | egocentric | 2,026 | https://proactegomem.github.io/ | https://arxiv.org/abs/2610.30089 | https://github.com/mayhugotong/ProactEgoMem | CC-BY-NC-4.0 annotations; source-video terms apply | 300 daily-life and 100 scientific clips; 1,500 QA items testing proactive information capture, state tracking, and location tracking | proactive-assistance; episodic-memory; streaming-video; benchmark | egocentric-video; video-qa; tool-calls | Ruotong Liao, Meikel Kokowski, and collaborators at LMU Munich, Stanford, Google, and TUM | mayhugotong on GitHub; Meikel on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/Meikel/ProactEgoMem | null | null | https://www.codabench.org/competitions/18340/ | null | https://huggingface.co/datasets/Meikel/ProactEgoMem | null | null | Annotations, evaluation code, and leaderboard are available. Obtain source video separately. Benchmark data, including dev items, must not be used for training. | null | Ego4D; FineBio; BioVL-QR |
STOU | benchmark | 2026-10 | arXiv | partial | egocentric | 2,026 | https://onedayoneyeah.github.io/safety-for-whom/ | https://arxiv.org/abs/2610.17728 | null | CC-BY-NC-4.0 annotations; source-video terms apply | Paper: 2,902 video-audio/text safety queries; gated release: 1,952 instances over 713 source clips, separating harm understanding from refusal | safety-evaluation; multimodal-reasoning; wearable-assistants | egocentric-video; audio; text; video-qa | Yewon Han, Jisu Jeon, Suhyun Choi, Juan Yeo, Minsoo Jo, Gihoon Kim, Hyunggyu Lim, and Taesup Kim | Chloeing on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/Chloeing/STOU | null | null | null | null | https://huggingface.co/datasets/Chloeing/STOU | Hugging Face agreement; source videos acquired separately | null | The public card documents a smaller gated release than the paper. Clip reconstruction instructions and annotations are listed; evaluation code is still marked coming soon. | null | Ego4D; Aria Everyday Activities; Nymeria+ |
EgoVoice | model | 2026-10 | EMNLP 2026 | watch | egocentric | 2,026 | https://egocentricvoice.github.io/ | https://arxiv.org/abs/2610.12248 | null | null | Proactive spoken assistance from HoloAssist-derived video/audio streams, jointly evaluating speech timing and guidance content | proactive-assistance; speech-generation; wearable-assistants | egocentric-video; audio; speech | Heeseung Kim | EgoVoice author (paper and project) | 2026-10-29 | null | null | null | null | null | null | null | null | null | The project schedules code, models, and dataset for November 2026; these assets are not yet verified as released. | null | HoloAssist |
Ropedia W0 | model | 2026-10 | arXiv | partial | egocentric | 2,026 | https://github.com/Ropedia/ropedia-w0 | https://arxiv.org/abs/2610.26855 | https://github.com/Ropedia/ropedia-w0 | Apache-2.0 code; CC-BY-NC-4.0 weights; upstream model terms apply | Camera-pose and skeleton-conditioned loco-manipulation world model trained with 10K human and 3K robot hours; 5B pre-trained and mid-trained checkpoints released | world-modeling; video-generation; cross-embodiment-transfer; robot-learning | egocentric-video; robot-camera-video; camera-pose; skeletons | Tianqi Liu and collaborators; Ropedia | Ropedia on GitHub; ropedia-ai on Hugging Face | 2026-10-29 | https://ropedia.github.io/ropedia-w0/ | null | https://huggingface.co/ropedia-ai/Ropedia-W0 | null | null | null | https://github.com/Ropedia/ropedia-w0#license | null | null | 5B inference and two checkpoints are available. A14B, benchmark-tuned checkpoints, evaluation package, and post-training code remain pending. The 13K-hour training mixture is not a separately verified full data release. | null | null |
Xp-Teleop | dataset | 2026-10 | Hugging Face | open | adjacent | 2,026 | https://huggingface.co/datasets/ropedia-ai/Xp-Teleop | https://arxiv.org/abs/2610.26855 | null | cc-by-4.0 | 27 Unitree G1 humanoid episodes, 543 action segments, and 96,064 frames with clean head RGB, keypoints, masks, calibration, and IMU | world-modeling; humanoid-robotics; loco-manipulation | robot-camera-video; keypoints; imu; segmentation; calibration | Ropedia | ropedia-ai on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/ropedia-ai/Xp-Teleop | null | null | null | null | https://huggingface.co/datasets/ropedia-ai/Xp-Teleop | null | null | Teleoperation data are downloadable; the W0 evaluation-code package is a separate, pending release. | Robot head-camera teleoperation, not human-worn capture; complements human-to-robot world-model evaluation. | null |
TouchScale | dataset | 2026-10 | arXiv | partial | egocentric | 2,026 | https://huggingface.co/datasets/2077AIDataFoundation/TouchScale | https://arxiv.org/abs/2610.10288 | null | cc-by-nc-4.0 | Current release: 100 hours, 15,324 episodes, 929 tasks, and 193 object meshes; synchronized head RGB-D, wrist RGB, and 880 tactile taxels per hand | hand-object-interaction; tactile-learning; bimanual-manipulation; robot-learning | egocentric-video; wrist-camera-video; depth; tactile; object-meshes | Dayou Li and the TouchScale collaboration | 2077AIDataFoundation on Hugging Face | 2026-10-29 | https://touch-scale.github.io/ | https://huggingface.co/datasets/2077AIDataFoundation/TouchScale | null | null | null | null | https://huggingface.co/datasets/2077AIDataFoundation/TouchScale | Hugging Face agreement required | null | The full 500-hour corpus is planned for release before November 2026; only the documented 100-hour portion is counted as available. | null | null |
EMBER-Bench | benchmark | 2026-10 | arXiv | open | egocentric | 2,026 | https://zhaoalexgoat.github.io/EMBER-Bench/index.html | https://arxiv.org/abs/2610.05013 | https://github.com/ZHAOAlexGoat/EMBER-Bench | cc-by-nc-4.0 | 189 recorded household tasks and 699 history-dependent questions: 548 next-action predictions and 151 causal traceback questions | episodic-memory; causal-reasoning; action-anticipation; benchmark | egocentric-video; video-qa; action-labels | Aoyang Cai, Boning Zhao, and collaborators at Tsinghua, HKU, and BAAI | ZHAOAlexGoat on GitHub; Alexzhao0605gt on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/Alexzhao0605gt/EMBER-Bench | null | null | null | null | https://huggingface.co/datasets/Alexzhao0605gt/EMBER-Bench | null | null | Benchmark media, portable catalog, and evaluation CLI are available. The maintained evaluator is not an exact reproduction of the paper's frozen leaderboard. | null | null |
Ego-SpaCR / SpaC-MEM | benchmark | 2026-10 | arXiv | benchmark | egocentric | 2,026 | https://github.com/joeliang0520/SAPCR-MEM | https://arxiv.org/abs/2610.05526 | https://github.com/joeliang0520/SAPCR-MEM | null | 620 ScanNet video sessions, 95 task-oriented conversations, and 3,100 evaluation queries for spatially grounded conversational memory | spatial-reasoning; conversational-memory; object-grounding | rgb-d; dialogue; scene-reconstruction; video-qa | Jiazhou Liang, Liam Gallagher, Kiko Chen, David Guo, Armin Toroghi, Yifan Simon Liu, and Scott Sanner | joeliang0520 on GitHub | 2026-10-29 | null | null | null | null | null | null | null | null | null | Code and a 3,218-question annotation ZIP are available; use the 3,100 official_eval questions for the reported protocol. Obtain ScanNet separately. Free-text judging and other paper baselines are outside the released scorer. | ScanNet-derived first-person spatial conversations, not a newly captured wearable corpus. | ScanNet |
PWM / PWM-Bench | benchmark | 2026-10 | arXiv | open | egocentric | 2,026 | https://github.com/AIGeeksGroup/PWM | https://arxiv.org/abs/2610.04920 | https://github.com/AIGeeksGroup/PWM | null | 150 first-person world-model customization tasks across indoor, outdoor, and Minecraft scenes; 300 paired support/test videos and 33,750 frames | world-modeling; video-generation; personalization; benchmark | egocentric-video; synthetic-video; camera-motion; captions | Zhexin Lou, Guancheng Lu, Zeyu Zhang, Yi Zhang, Yang Zhao, and Hao Tang | AIGeeksGroup on GitHub and Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/AIGeeksGroup/PWM-Bench | null | null | null | null | null | Public benchmark files and code; consult upstream video and base-model terms | null | Support and test segments are disjoint within each scene, not a cross-scene generalization split. Reuse terms are not summarized by a single dataset license. | null | EgoVid-5M; InCrowd-VI; MineDojo |
EgoExo-Next | benchmark | 2026-10 | arXiv | benchmark | egocentric | 2,026 | https://huggingface.co/datasets/yutongli2024/EgoExo-Next | https://arxiv.org/abs/2610.04506 | null | Non-commercial research annotations; inherited source-image terms | 2,503 human-curated four-choice questions from six ego/ego-exo sources, testing visual next-state prediction and cross-view correspondence | action-anticipation; cross-view-reasoning; spatial-reasoning; benchmark | egocentric-images; exocentric-images; video-qa | Yutong Li, Molin Wang, Xiaotong Li, Yanyan Fang, Daoguo Dong, and Ziyi Ye | yutongli2024 on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/yutongli2024/EgoExo-Next | null | null | null | null | https://huggingface.co/datasets/yutongli2024/EgoExo-Next | null | null | Image options, input metadata, and answers are released. Dataset-viewer image-row counts are not benchmark question counts. | null | null |
NextMe-800 / NextAct | dataset | 2026-10 | arXiv | request | egocentric | 2,026 | https://kkkkawayi.github.io/nextme-800/ | https://arxiv.org/abs/2610.01461 | null | CC-BY-NC-4.0; additional research/privacy conditions | 795 hours from one Aria wearer across a 126-day span, with 527 recordings on 114 recording days; gaze, audio, five-level captions, and a 1,500-point NextAct benchmark | lifelogging; personalized-assistance; action-anticipation; episodic-memory | egocentric-images; gaze; audio; captions | Zhaoxu Meng, Yiming Sun, Mingyuan Gao, Jiachang Zhang, Zhuhan Dai, Yipeng Du, Zheng Lian, and Jian-Qiao Zhu | mmm8383 on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/mmm8383/NextMe-800 | null | null | null | null | https://huggingface.co/datasets/mmm8383/NextMe-800 | Hugging Face research-use and privacy agreement | null | The card lists 8.81 TB of recordings plus captions and benchmark scripts. NextAct also uses EgoLife; one wearer's long history does not establish population diversity. | null | null |
Argus | toolkit | 2026-10 | GitHub | open | egocentric | 2,026 | https://github.com/Pantheon-Industries-Inc/argus | null | https://github.com/Pantheon-Industries-Inc/argus | null | Dense video annotation and data-quality checks across LeRobot, MCAP, and video archives; publisher reports 3,546 audited human/robot episodes across nine datasets | data-annotation; quality-control; human-demonstrations; robot-learning | egocentric-video; robot-camera-video; robot-actions; language | Pantheon Industries | Pantheon-Industries-Inc on GitHub | 2026-10-29 | https://pantheon.inc/blog/argus | null | null | null | null | null | null | null | null | Public annotation harness and inspection outputs; generated quality judgments are not independently validated ground truth. | Mixed human-ego and robot data tooling, with explicit support for OpenAoE, Gen-HumanEgo, and Egocentric-100K. | null |
APM-Bench | benchmark | 2026-09 | arXiv | partial | egocentric | 2,026 | https://huggingface.co/datasets/Jianguo-Huang11/APM-Bench | https://arxiv.org/abs/2609.37559 | https://github.com/Jianguo-Huang11/APM-Bench | CC-BY-NC-4.0; source-dataset terms apply | 549 sessions in 104 trajectories, with 2,719 candidate interactions for cross-session memory, current-scene perception, and adaptive response | episodic-memory; streaming-video; proactive-assistance; benchmark | egocentric-video; video-qa | Jianguo Huang, Jinming Liu, and collaborators | Jianguo-Huang11 on GitHub and Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/Jianguo-Huang11/APM-Bench | null | null | null | null | https://huggingface.co/datasets/Jianguo-Huang11/APM-Bench | null | null | The 549 session videos are hosted; annotations and evaluation code are still being prepared. Do not treat the video upload as a complete runnable benchmark. | null | EgoLife; HD-EPIC |
HumanVerse-500 / lambda_0 | dataset | 2026-09 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2610.00438 | https://arxiv.org/abs/2610.00438 | null | null | Reported 500-hour wearable corpus synchronizing egocentric video with body and hand motion, used by the lambda_0 humanoid loco-manipulation policy | human-demonstrations; full-body-motion; loco-manipulation; vla | egocentric-video; body-motion; hand-motion | Chongyang Xu, Zhao Wu, Jin Chen, and collaborators | HumanVerse author team (paper) | 2026-10-29 | null | null | null | null | null | null | null | null | null | The paper promises code, models, and data; a downloadable release was not verified. First submission was September 30 despite the October arXiv identifier. | null | null |
Ego4WAM | model | 2026-09 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2609.40341 | https://arxiv.org/abs/2609.40341 | null | null | Controlled world-action-model study separating human-robot alignment, duration, task diversity, action labels, and training-stage effects | world-modeling; robot-learning; cross-embodiment-transfer | egocentric-video; robot-actions | Zhihao Sun, Liu Liu, Xinjiang Wang, and collaborators | Ego4WAM author team (paper) | 2026-10-29 | null | null | null | null | null | null | null | null | null | Paper verified; executable code and model weights were not verified. | null | null |
IronMind | model | 2026-09 | arXiv | watch | egocentric | 2,026 | https://xpeng-robotics.github.io/ironmind/ | https://arxiv.org/abs/2609.39403 | null | null | Humanoid VLA pretraining on human ego video and robot data using camera-space actions; studies budgets from 250 to 10,000 hours | vla; dexterous-manipulation; cross-embodiment-transfer | egocentric-video; robot-camera-video; hand-pose; robot-actions | XPENG Robotics | XPENG Robotics (report and project) | 2026-10-29 | null | null | null | null | null | null | null | null | null | The project provides the report and demonstrations; code, checkpoints, and the training corpus were not verified as downloadable. | null | null |
EgoTSR++ / SpatialLogic-Bench | model | 2026-09 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2609.31751 | https://arxiv.org/abs/2609.31751 | null | null | Goal-conditioned progress reasoning from bidirectional state pairs, with order-robust comparisons and long-horizon failure/recovery evaluation | spatial-reasoning; temporal-reasoning; task-progress; failure-detection | egocentric-video; language; state-pairs | Xiaoda Yang, Can Wang, Yuxiang Liu, Pengfei Zhou, Jianwen Lou, and Shuicheng Yan | EgoTSR++ author team (paper) | 2026-10-29 | null | null | null | null | null | null | null | null | null | A distinct follow-up to EgoTSR; paper verified, runnable model and benchmark releases not verified. | null | EgoTSR |
EgoWild / EgoWild2Dex | dataset | 2026-09 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2609.23755 | https://arxiv.org/abs/2609.23755 | null | null | Reported 538.9 hours, 179,049 episodes, 125,961 task descriptions, and 1,282 object categories from in-the-wild head-camera manipulation | human-demonstrations; dexterous-manipulation; cross-embodiment-transfer | egocentric-video; hand-motion; task-descriptions | Kunyang Lin, Xutao Wen, Jingxi Lin, and collaborators | EgoWild2Dex author team (paper) | 2026-10-29 | null | null | null | null | null | null | null | null | null | The paper announces data, models, and code. The reported corpus size is not a verified downloadable release size. | null | null |
SlugTrails | benchmark | 2026-09 | arXiv | open | egocentric | 2,026 | https://github.com/Head-inthe-Cloud/SlugTrails | https://arxiv.org/abs/2609.19876 | https://github.com/Head-inthe-Cloud/SlugTrails | null | Aria recordings across three campus buildings and six floors, with CAD floor plans, survey-aligned trajectories, and single-frame, sweep, and walking-stream protocols | visual-localization; indoor-navigation; spatial-reasoning; benchmark | egocentric-video; imu; floor-plans; camera-pose; point-clouds | Yunqian Cheng and Roberto Manduchi | Head-inthe-Cloud on GitHub and Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/Head-inthe-Cloud/SlugTrails | null | null | null | null | null | null | null | A downloadable package, checksums, and viewers are provided. Public video is anonymized; original unblurred VRS is not distributed. | null | null |
XPACE | model | 2026-09 | arXiv | watch | egocentric | 2,026 | https://xpeng-robotics.github.io/xpace/ | https://arxiv.org/abs/2609.17372 | null | null | Joint future-video and action modeling from heterogeneous human and robot experience, including simulated recoveries for policy improvement | world-modeling; vla; cross-embodiment-transfer; robot-learning | egocentric-video; robot-camera-video; robot-actions | Jiacheng Wei, Jerry Bai, and collaborators at XPENG Robotics | XPENG Robotics (report and project) | 2026-10-29 | null | null | null | null | null | null | null | null | null | Technical report and demonstrations verified; downloadable model assets were not verified. | null | null |
EEH-R / EventEgoHands++ | dataset | 2026-09 | IEEE Access 2026 | request | egocentric | 2,026 | https://ryhara.github.io/EventEgoHandsV2/ | https://arxiv.org/abs/2609.17189 | https://github.com/ryhara/EventEgoHandsV2 | CC-BY-NC-4.0 code; dataset request terms apply | Real event-camera hand dataset with about 1M annotated frames, 85 sequences, eight participants, and 2.36 hours; accompanies an expanded synthetic N-HOT3D | 3d-hand-reconstruction; hand-pose-estimation; event-based-vision | event-camera; hand-mesh; hand-masks; hand-boxes | Ryosei Hara, Wataru Ikeda, Masashi Hatano, and Mariko Isogawa; Keio University and JST PRESTO | ryhara on GitHub | 2026-10-29 | null | null | null | null | null | null | https://github.com/ryhara/EventEgoHandsV2 | EEH-R and N-HOT3D request forms linked from the official repository | null | Training, evaluation, and viewer code is public; dataset access is form-based. | null | null |
EgoPathBench | benchmark | 2026-09 | arXiv | open | egocentric | 2,026 | https://huggingface.co/datasets/runder1/EgoPathBench | https://arxiv.org/abs/2609.16610 | null | cc-by-nc-sa-4.0 | 31,852 training, 1,345 validation, and 1,111 benchmark questions for traversability, waypoint routes, and intent-conditioned paths from first-person RGB | navigation; spatial-reasoning; path-planning; benchmark | egocentric-images; waypoints; language | Yang Zhao, Zhuo Chen, and Xubo Yang; Shanghai Jiao Tong University | runder1 on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/runder1/EgoPathBench | null | null | null | null | https://huggingface.co/datasets/runder1/EgoPathBench | null | null | File-based release with 36 archives and checksums; use the corrected benchmark_evalfix labels. A dataset-viewer format error does not mean the archives are unavailable. | null | null |
HuRo | dataset | 2026-09 | CoRL 2026 | partial | egocentric | 2,026 | https://3587jjh.github.io/HuRo/ | https://arxiv.org/abs/2609.10706 | https://github.com/3587jjh/HuRo | null | Reported 630K robotized episodes and 142M frames from five human egocentric sources; aligns rendered robot observations with retargeted actions | human-to-robot-transfer; vla; data-annotation; motion-retargeting | egocentric-video; rendered-video; robot-actions; language | Jinho Jeong and collaborators at RLWRLD and Yonsei University | 3587jjh on GitHub | 2026-10-29 | null | null | null | null | null | null | null | null | null | Robotization pipeline code is available; the project does not yet provide a verified full-dataset download. Source-video rights still apply. | null | EgoDex; EgoVerse; Egocentric-10K; Ego4D; EPIC-KITCHENS |
MINT / EgoPipeline | model | 2026-09 | arXiv | open | egocentric | 2,026 | https://github.com/wuji-technology/wuji-ego-mint | https://arxiv.org/abs/2609.04958 | https://github.com/wuji-technology/wuji-ego-mint | MIT original code; upstream models and MANO have separate terms | Joint world-space camera and two-hand trajectory prediction; public checkpoint, training/inference code, EgoPipeline, and non-video annotations covering 1,021 hours | hand-pose-estimation; camera-pose-estimation; motion-retargeting; data-annotation | egocentric-video; camera-pose; hand-mesh; hand-trajectories | Zijie Zhu and collaborators at Wuji Technology, ShanghaiTech, Tsinghua, HKU, and Zhejiang University | wuji-technology on GitHub; ZZJAsher on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/ZZJAsher/wuji_ego_mint | https://huggingface.co/ZZJAsher/mint_v1 | null | null | null | https://github.com/wuji-technology/wuji-ego-mint/blob/main/THIRD_PARTY_NOTICES.md | null | null | Source videos are not redistributed; separately licensed MANO and optional pipeline assets must be obtained from their providers. Pseudo-labels are not motion-capture ground truth. | null | null |
TempCloze | benchmark | 2026-09 | EMNLP 2026 Findings | benchmark | egocentric | 2,026 | https://huggingface.co/datasets/CedPei/TempCloze | https://arxiv.org/abs/2609.01515 | https://github.com/CedricPei/Temporal-Cloze | null | 1,521 videos from seven sources for selecting a missing middle clip; separates semantic plausibility, temporal alignment, and progression | temporal-reasoning; video-understanding; benchmark | egocentric-video; general-video; temporal-annotations | Wenqi Pei, Henry Hengyuan Zhao, Yilai Liu, Jiahao Meng, Han Chen, Ziyu Wang, and Hongyang Du | CedricPei on GitHub; CedPei on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/CedPei/TempCloze | null | null | null | null | null | null | null | The Hugging Face package provides source identifiers and gap boundaries, not a self-contained video corpus. Obtain source footage under its own terms. | Mixed-source benchmark centered on long-take and egocentric video; not all source footage is wearable capture. | null |
Eidon Tracker POV | dataset | 2026-09 | Hugging Face | open | egocentric | 2,026 | https://huggingface.co/datasets/eidon-ai/tracker-pov | null | null | cc-by-4.0 | 1,273.8 hours and 13,451 head-camera recordings from 27 contributors, paired with seven-slot upper-body IMU streams at 24 Hz | human-demonstrations; sensor-fusion; robot-learning; activities-of-daily-living | egocentric-video; imu; orientation; quality-metadata | Eidon AI / Solidic Labs Inc | eidon-ai on Hugging Face | 2026-10-29 | https://eidonai.online/ | https://huggingface.co/datasets/eidon-ai/tracker-pov | null | null | null | null | https://huggingface.co/datasets/eidon-ai/tracker-pov | null | null | Join video with the separate tracker-pov-imu repository on recording_id; they are one corpus, not additive hours. Some rigs lack slots. Video is not face-blurred and may contain incidental private information. | null | null |
Eidon Egocentric POV | dataset | 2026-09 | Hugging Face | open | egocentric | 2,026 | https://huggingface.co/buckets/eidon-ai/egocentric-pov | null | null | cc-by-4.0 | 305.7 hours and 1,370 head-mounted video-only recordings from 37 contributors, without the Tracker POV IMU harness | human-demonstrations; robot-learning; activities-of-daily-living | egocentric-video; metadata | Eidon AI / Solidic Labs Inc | eidon-ai on Hugging Face | 2026-10-29 | https://eidonai.online/ | https://huggingface.co/buckets/eidon-ai/egocentric-pov | null | null | null | null | https://huggingface.co/buckets/eidon-ai/egocentric-pov | null | null | A Storage Bucket, not a datasets-library repository. Contributor identities overlap with Tracker POV, so cross-corpus subject splits require care. | null | null |
Eidon Capture and Playback Tools | toolkit | 2026-09 | GitHub | open | egocentric | 2,026 | https://eidonai.online/ | null | https://github.com/Eidon-AI/eidon-tracker | null | Public wearable IMU hardware/firmware, glove designs, and Eidon Sim for arm kinematics and video-synchronized trajectory playback | data-collection; sensor-synchronization; visualization; motion-retargeting | imu; hand-motion; body-motion; egocentric-video | Eidon AI / Solidic Labs Inc | Eidon-AI on GitHub | 2026-10-29 | https://github.com/Eidon-AI/eidon-sim | null | null | null | null | null | null | null | null | Archived capture ecosystem; Eidon has ceased operations and the old mobile-app backend is shut down. Sim is MIT; check each hardware/code repository's terms separately. | null | null |
EgoViz-120 | dataset | 2026-08 | Hugging Face | open | egocentric | 2,026 | https://huggingface.co/datasets/humaidtech/EgoViz-120 | null | null | cc-by-4.0 | 120.4 hours, 5,392 clips, and 12 workplace categories; stereo RGB-D, wrist cameras, three IMUs, hand pose, and action labels in MCAP | human-demonstrations; hand-object-interaction; sensor-fusion; robot-learning | stereo-video; depth; wrist-camera-video; imu; hand-pose; action-labels | Humaid Tech Inc | humaidtech on Hugging Face | 2026-10-29 | null | https://huggingface.co/buckets/humaidtech/EgoViz-120 | null | null | null | null | https://huggingface.co/datasets/humaidtech/EgoViz-120 | null | null | The dataset repository holds documentation; 7.25 TB of media and metadata live in the linked Storage Bucket. The empty dataset-viewer warning does not describe the bucket. | null | null |
AnyWorld | model | 2026-08 | arXiv | open | egocentric | 2,026 | https://xpeng-robotics.github.io/anyworld/ | https://arxiv.org/abs/2608.29242 | https://github.com/xpeng-robotics/AnyWorld | Apache-2.0 code; external model terms apply | Factorizes action, camera, and embodiment to recompose unpaired human interactions into robot-native video-action experience | world-modeling; video-generation; cross-embodiment-transfer | egocentric-video; robot-camera-video; camera-pose; actions | Cheng Chen, Jerry Bai, Jiacheng Wei, and collaborators | xpeng-robotics on GitHub | 2026-10-29 | null | null | null | null | null | null | https://github.com/xpeng-robotics/AnyWorld/blob/main/LICENSE | null | null | Image-editor training/inference and world-model inference code are public. External base components and adapted checkpoint setup follow the repository's model-weight guide. | null | null |
EgoArgus | benchmark | 2026-08 | arXiv | partial | egocentric | 2,026 | https://github.com/NYCU-NLP-Lab/EgoArgus | https://arxiv.org/abs/2608.25561 | https://github.com/NYCU-NLP-Lab/EgoArgus | null | 6,978 understanding examples over 3,719 source-derived clips and 789 intervention decisions over 311 hosted videos | proactive-assistance; multimodal-reasoning; intervention-prediction; benchmark | egocentric-video; dialogue; video-qa | NYCU NLP Lab / EgoArgus author team | NYCU-NLP-Lab on GitHub; tommytyc on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/tommytyc/EgoArgus | null | null | null | null | null | null | null | Annotations and evaluation code are public; understanding videos require separate source access. The repository calls itself a release draft and explicitly says reuse rights are not yet granted. Fifteen rows have out-of-option labels. | null | QaEgo4D; EgoPlan-Bench2 |
RealTimeVideo-Instruct-112K | dataset | 2026-08 | arXiv | open | egocentric | 2,026 | https://huggingface.co/datasets/zeyun-zhong/RealTimeVideo-Instruct-112K | https://arxiv.org/abs/2608.13416 | https://github.com/zeyun-zhong/StreamTTT | CC-BY-4.0 annotations except ADT subset: CC-BY-NC-SA-4.0 | 112,102 real-time QA/caption instruction samples, including 22,493 Ego4D and 6,861 Aria Digital Twin samples; queries occur when evidence first becomes visible | streaming-video; instruction-tuning; video-qa; spatial-reasoning | annotations; captions; temporal-annotations | Joya Chen, Zeyun Zhong, and Mike Zheng Shou | zeyun-zhong on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/zeyun-zhong/RealTimeVideo-Instruct-112K | null | null | null | null | https://huggingface.co/datasets/zeyun-zhong/RealTimeVideo-Instruct-112K/blob/main/README.md | null | null | Annotations only; fetch each source-video collection separately. The ADT subset has stricter non-commercial/share-alike terms. | Mixed ego and general-video annotation corpus; only the named egocentric subsets are first-person. | Ego4D; Aria Digital Twin; E.T. Instruct 164K; LLaVA-Video-178K; ActivityNet; QVHighlights |
EgoDatasets | survey | 2026-10 | Applied Sciences 2026 | open | egocentric | 2,026 | https://www.mdpi.com/2076-3417/16/20/10268 | https://doi.org/10.3390/app162010268 | null | null | Dataset-centric taxonomy of 96 resources from 2008-2026: 79 core egocentric/multiview datasets and 17 contextual exocentric datasets | survey; action-anticipation; dataset-selection; evaluation | literature | Alexandros Mitsou and Evaggelos Spyrou; University of Thessaly | Applied Sciences / MDPI | 2026-10-29 | null | null | null | null | null | null | null | null | null | Open-access survey published October 16; distinguishes native anticipation support from adapted and post-hoc protocols. | null | null |
Xperience-10M | dataset | 2026-03 | Hugging Face | request | egocentric | 2,026 | https://huggingface.co/datasets/ropedia-ai/xperience-10m | null | null | other | 10M experiences, 10K hours, six video streams, audio, stereo depth, camera pose, hand/body mocap, IMU, hierarchical language, about 1 PB total | embodied-ai; world-modeling; robot-learning; sensor-fusion; 3d-4d-understanding; imitation-learning | fisheye-video; stereo-video; audio; stereo-depth; camera-pose; hand-mocap; full-body-mocap; imu; hierarchical-language | Ropedia | ropedia-ai on Hugging Face | 2026-06-20 | null | null | null | null | null | null | null | controlled non-commercial access; external agreement-signing step may be required | null | null | null | null |
Xperience-10M Sample | dataset | 2026-03 | Hugging Face | open | egocentric | 2,026 | https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample | null | null | cc-by-nc-4.0 | Public sample episode; Hugging Face viewer reports 6 rows | sample-data; loader-testing; visualization; task-suite-prototyping | video; hdf5-annotations | Ropedia | ropedia-ai on Hugging Face | 2026-06-20 | null | null | null | null | null | null | https://huggingface.co/datasets/ropedia-ai/xperience-10m-sample | null | null | null | null | null |
Egocentric Manufacturing | dataset | 2026-08 | Hugging Face | request | egocentric | 2,026 | https://huggingface.co/datasets/Worlddatalabs/egocentric-manufacturing | null | null | World Data Labs custom commercial-use terms | 259 first-person manufacturing clips spanning 14 process domains, about 100 hours, 256 workers, 3,911 scenes, and 2,081 entities; the current gated repository contains 533 files and 60.6 GiB | industrial-egocentric-video; procedure-understanding; hand-object-interaction; vla; robot-learning | egocentric-video; action-labels; hand-pose; scene-segmentation; metadata | World Data Labs | World Data Labs on Hugging Face | 2026-08-23 | null | https://huggingface.co/datasets/Worlddatalabs/egocentric-manufacturing | null | null | null | null | https://huggingface.co/datasets/Worlddatalabs/egocentric-manufacturing | Manual Hugging Face approval under the publisher's dataset agreement | https://huggingface.co/datasets/Worlddatalabs/egocentric-manufacturing | The dataset card, 259 media/annotation records, and 60.6 GiB repository are live, but file access is manually gated and reuse follows custom terms rather than an OSI or Creative Commons license. | null | null |
Axis Ego Samples | dataset | 2026-08 | Hugging Face | open | egocentric | 2,026 | https://huggingface.co/datasets/axisrobotics/axis-ego-samples | null | null | other | Public 100-hour LeRobot egocentric sample across education, hospitality, retail, warehousing, handicraft, catering, office, production, and logistics, plus ten episodes sampled from a separate 150-hour delivery corpus | human-demonstrations; robot-learning; imitation-learning; hand-object-interaction | egocentric-video; depth; hand-object-interaction; lerobot | Axis Robotics | axisrobotics on Hugging Face | 2026-08-23 | null | https://huggingface.co/datasets/axisrobotics/axis-ego-samples | null | null | null | null | https://huggingface.co/datasets/axisrobotics/axis-ego-samples/blob/main/README.md | null | null | The media is downloadable without an access form, but the card declares license: other and does not state concrete reuse terms; users should confirm permission before redistribution or commercial use. | null | null |
RealFactory-Ego-Data | dataset | 2026-08 | Hugging Face | open | egocentric | 2,026 | https://huggingface.co/datasets/FlyBirdData/RealFactory-Ego-Data | null | null | apache-2.0 | Five head-mounted episodes totaling about 14 minutes on a live touchscreen production line, with RGB, 21-landmark hand pose, and frame-level action boundaries in LeRobot-compatible packages | industrial-egocentric-video; hand-object-interaction; action-segmentation; imitation-learning | egocentric-video; hand-pose; action-labels; lerobot | FlyBirdData / Shenzhen Mingzhi DataTech | FlyBirdData on Hugging Face | 2026-08-23 | null | https://huggingface.co/datasets/FlyBirdData/RealFactory-Ego-Data | null | null | null | null | https://huggingface.co/datasets/FlyBirdData/RealFactory-Ego-Data/blob/main/README.md | null | null | null | null | null |
BaseMatrix EGO Desktop Demo | dataset | 2026-08 | Hugging Face | open | egocentric | 2,026 | https://huggingface.co/datasets/basematrix/ego-desktop-demo | null | null | cc-by-nc-4.0 | One public 70-second episode with 2,114 frames, binocular fisheye egocentric video, an exocentric view, 6-DoF camera pose, metric 3D hand keypoints, contact state, and derived wrist/grasp actions | hand-object-interaction; 3d-hand-reconstruction; human-demonstrations; robot-learning | stereo-video; exocentric-video; camera-pose; hand-pose; contact-labels; lerobot | BaseMatrix | basematrix on Hugging Face | 2026-08-23 | null | https://huggingface.co/datasets/basematrix/ego-desktop-demo | null | null | null | null | https://huggingface.co/datasets/basematrix/ego-desktop-demo/blob/main/README.md | null | null | null | null | null |
ActTrace Ego-Home | dataset | 2026-08 | Hugging Face | open | egocentric | 2,026 | https://huggingface.co/datasets/ActTrace/acttrace-ego-home | null | null | cc-by-nc-4.0 | Version 0.2 provides 13 head-mounted episodes over 11 household tasks and 18 minutes, with RGB, metric LiDAR depth, 100 Hz IMU, metric 3D hands, action segments, and IMU-verified 6-DoF head trajectories (r=0.971-0.995) | human-demonstrations; hand-object-interaction; sensor-fusion; imitation-learning; vla | egocentric-video; depth; imu; camera-pose; hand-pose; action-labels; lerobot | ActTrace | ActTrace on Hugging Face | 2026-08-23 | null | https://huggingface.co/datasets/ActTrace/acttrace-ego-home | null | null | null | null | https://huggingface.co/datasets/ActTrace/acttrace-ego-home/blob/main/README.md | null | null | null | null | null |
Everyday Manipulation 3D | dataset | 2026-08 | Hugging Face | open | egocentric | 2,026 | https://caryx.ai/data | null | null | CC BY-NC-SA 4.0 with MANO-assets exception; raw companion CC BY 4.0 | Annotated release: 116 episodes, 28,019 frames at 15 fps, ten tasks, and five participants with metric LiDAR depth, 6-DoF camera pose, MANO hands, contact, masks, and language; companion raw release: 1,513 clips, 10.28 hours, and 279 GiB | human-demonstrations; hand-object-interaction; 3d-hand-reconstruction; segmentation; robot-learning | egocentric-video; rgb-d; depth; camera-pose; hand-mesh; contact-labels; segmentation; natural-language; lerobot | CaryX AI | CaryxAI on Hugging Face | 2026-08-23 | https://caryx.ai/data | https://huggingface.co/datasets/CaryxAI/everyday-manipulation-3d | null | null | null | null | https://huggingface.co/datasets/CaryxAI/everyday-manipulation-3d/blob/main/README.md | null | null | The catalog groups the dense annotated sample and its larger raw RGB-D companion as one resource. Their licenses differ, so users should follow the terms of the specific package they download. | null | null |
EgoTac | model | 2026-08 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.15060 | https://arxiv.org/abs/2608.15060 | null | null | Predicts continuous force and binary contact directly from egocentric human video; trained on more than 5.7M image-tactile pairs and reports below 0.06 N mean in-domain force error | tactile-learning; force-estimation; contact-estimation; hand-object-interaction; robot-learning | egocentric-video; tactile; force; contact-labels | Wenkang Zhang, Chengbo Yuan, Zicheng Zhang, Zhengxue Cheng, and Yang Gao | null | 2026-08-23 | null | null | null | null | null | null | null | null | null | The paper is public, but no official code, checkpoint, training corpus, or reuse license was linked in the arXiv record. | null | null |
EgoGazeLite | model | 2026-08 | ECCV 2026 WearableAI Workshop | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.15614 | https://arxiv.org/abs/2608.15614 | null | null | Software-only gaze-and-crop model for token-efficient wearable MLLM input with 15.7M parameters, 6.71 GFLOPs, and 21.6 ms/frame end-to-end latency on consumer accelerator hardware | gaze-prediction; video-language; wearable-ai; on-device-perception | egocentric-video; gaze; natural-language | Matteo Stoiber and Niels Buus Lassen | null | 2026-08-23 | null | null | null | null | null | null | null | null | null | The archival WearableAI workshop paper is public, but no code, checkpoint, evaluation package, or license was linked. | null | null |
Long-Tail Embodied Urban Navigation from In-the-Wild Videos | model | 2026-08 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.16476 | https://arxiv.org/abs/2608.16476 | null | null | Automatically annotates web-scale in-the-wild egocentric videos with metric trajectories and structured navigation semantics, then trains a VLA policy while exposing rare perception-motion failure modes | navigation; vla; data-curation; long-tail-analysis; embodied-ai | egocentric-video; camera-pose; natural-language; trajectories | Bingyi Xia, Han Bao, Zhewei Chen, Hanjing Ye, Jingwen Yu, Yuhan Pang, Wenjun Xu, and Jiankun Wang | null | 2026-08-23 | null | null | null | null | null | null | null | null | null | The arXiv paper is public, but no annotation pipeline, training data, implementation, checkpoint, or license was linked. | null | null |
EgoHRV | model | 2026-08 | ECCV 2026 | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.18711 | https://arxiv.org/abs/2608.18711 | null | null | Estimates heart rate and heart-rate variability from headset gaze cameras; adding its uncertainty-aware HRV features improves the reported EgoExo4D proficiency result by 17.8% | physiological-signals; health-ai; skill-assessment; wearable-ai | egocentric-video; gaze; physiological-signals | Berken Utku Demirel and Christian Holz | null | 2026-08-23 | null | null | null | null | null | null | null | null | null | The ECCV paper links an official GitHub repository, but that URL returned 404 during this audit; code, checkpoint, and license therefore remain unverified. | null | null |
G3Ego | model | 2026-08 | ECCV 2026 CONTEXTUS Workshop | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.20157 | https://arxiv.org/abs/2608.20157 | null | null | Uses wearer gaze to prune action-scene graphs built from sparse frames, improving class-balanced action recognition and anticipation on EGTEA Gaze+ and MECCANO without large video-model pretraining | action-recognition; action-anticipation; gaze; hand-object-interaction | egocentric-video; gaze; scene-graphs; action-labels | Marko Haralovic, Akash Ramakrishnan, and Estefania Talavera Martinez | null | 2026-08-23 | null | null | null | null | null | null | null | null | null | The workshop paper is public, but no implementation, graph annotations, checkpoints, or license was linked. | null | null |
DreamHand | model | 2026-08 | arXiv | watch | egocentric | 2,026 | https://ggxxii.github.io/dreamhand/ | https://arxiv.org/abs/2608.20308 | null | null | Repurposes a video diffusion model as a deterministic geometry encoder for continuous metric bimanual trajectories, reducing MPJPE-p by 30% on ARCTIC and 40% on HOT3D, with 46-61% gains when out-of-sight hands are counted | 3d-hand-reconstruction; hand-tracking; hand-object-interaction; robot-learning | egocentric-video; hand-pose; camera-pose; video-features | Yufei Liu, Xixi Wang, Hao Li, Ganlong Zhao, Kaitong Cai, Chengkai Jin, Chunxiao Liu, Jianbo Liu, Siyuan Huang, Xingang Pan, and Hongsheng Li | null | 2026-08-23 | https://ggxxii.github.io/dreamhand/ | null | null | null | null | null | null | null | null | The paper and project page are public, but no code, checkpoint, or license was available during this audit. | null | null |
Helios 2.0 | model | 2025-03 | ECCV 2026 Event-Based Multimodal Vision Workshop | watch | egocentric | 2,026 | https://arxiv.org/abs/2503.07825 | https://arxiv.org/abs/2503.07825 | null | null | Event-camera smart-glasses gesture system trained only on synthetic data, reporting more than 80% F1 for the six-channel model at 6-8 mW on a Qualcomm Hexagon DSP | gesture-recognition; wearable-interaction; energy-efficient-smart-glasses; on-device-perception | event-camera; smart-glasses; hand-gesture | Prarthana Bhattacharyya, Joshua Mitton, Ryan Page, Owen Morgan, Oliver Powell, Benjamin Menzies, Gabriel Homewood, Kemi Jacobs, Paolo Baesso, Taru Muhonen, Richard Vigars, and Louis Berridge | null | 2026-08-23 | null | null | null | null | null | null | null | null | null | The revised arXiv record confirms presentation at the ECCV 2026 Event-Based Multimodal Vision Workshop; no implementation, trained model, dataset package, or license was linked. | null | null |
Vision-Language Models for Egocentric Video | survey | 2026-08 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.18671 | https://arxiv.org/abs/2608.18671 | null | null | Critical survey connecting egocentric VLMs across hand-object interaction, temporal reasoning, efficient clip selection, graph/object-centric reasoning, wearable assistance, and embodied AI | survey; video-language; hand-object-interaction; embodied-ai | egocentric-video; natural-language | Mohammad Zamani and Fatemeh Ziaeetabar | null | 2026-08-23 | null | null | null | null | null | null | null | null | null | The survey preprint is public; no companion bibliography, code, or structured data release was linked. | null | null |
AlloEgo-VLM | model | 2026-08 | arXiv | partial | adjacent | 2,026 | https://github.com/CKL9001/AlloEgo-VLM | https://arxiv.org/abs/2608.15605 | https://github.com/CKL9001/AlloEgo-VLM | null | AlloEgo-View triplets and an SFT framework for resolving allocentric versus egocentric spatial-reference ambiguity, with an NVIDIA Isaac Sim object-search demonstration | spatial-reasoning; referential-grounding; embodied-ai; question-answering | images; natural-language; spatial-annotations | Kuan-Lin Chen, Tzu-Ti Wei, Chao-Chi Liao, Yu-Chee Tseng, and Jen-Jee Chen | null | 2026-08-23 | null | https://github.com/CKL9001/AlloEgo-VLM | null | null | null | null | null | null | null | Implementation and processed AlloEgo-View images are public, but the repository does not declare a top-level software license and the processed images inherit restrictions from several source datasets. | Reference-frame reasoning over general images and simulation rather than wearable first-person capture. | null |
WorldRover-10M | dataset | 2026-08 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.15659 | https://arxiv.org/abs/2608.15659 | null | null | Synthetic minute-scale world explorations rendered from first-person, third-person, and 360-degree views with RGB, metric depth, camera trajectories, action signals, optical flow, and long-range 2D/3D tracks | world-modeling; data-generation; navigation; 3d-4d-understanding | synthetic-video; egocentric-video; exocentric-video; 360-video; depth; camera-pose; optical-flow; trajectories | Xiaojie Xu, Zhengyuan Lin, Runyi Li, Yihao Liu, Kaipeng Zhang, Yongtao Ge | null | 2026-08-23 | null | null | null | null | null | null | null | null | null | The arXiv comments say code and data links will be updated, but no live engine, dataset, repository, or license was linked during this audit. | Artist-built synthetic environments with multiple viewpoints, not real wearable capture. | null |
UniWM | model | 2025-10 | ECCV 2026 | partial | adjacent | 2,026 | https://github.com/UWMILab/UniWM | https://arxiv.org/abs/2510.08713 | https://github.com/UWMILab/UniWM | null | Memory-augmented unified world model for robot navigation foresight and planning across Go Stanford, ReCon, SACSoN, SCAND, and unseen TartanDrive environments | world-modeling; navigation; planning; memory | robot-camera-video; natural-language; trajectories; memory-index | University of Wisconsin-Madison Intelligent Systems Lab and collaborators | null | 2026-08-23 | null | https://huggingface.co/datasets/fly1113/UniWM_Dataset | null | null | null | null | null | null | null | The ECCV paper, implementation, and Hugging Face dataset are public, but a repository-wide software/data license was not verified. | Robot-navigation camera data rather than human wearable first-person capture. | null |
EATR-Stereo | model | 2026-08 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.17453 | https://arxiv.org/abs/2608.17453 | null | null | Routes paired head-mounted stereo evidence with embodiment-conditioned tokens for a 33-DoF humanoid, reporting 60% full-task, 100% grasp, and 80% stage success on long-horizon manipulation | vla; humanoid-manipulation; long-horizon-manipulation; sensor-fusion | robot-camera-video; stereo-video; proprioception; robot-actions | Songwei Wu, Rui Zhao, Fan Yang, Zhongqiang Nie, Zhiduo Jiang, Wandong Sun, Yuwei Li, Jian Hu, Yang Liu, and Hong Liu | null | 2026-08-23 | null | null | null | null | null | null | null | null | null | The paper is public, but no code, checkpoint, robot dataset, or license was linked. | Robot head-camera stereo rather than human wearable video. | null |
Hydra-0 | model | 2026-08 | arXiv | watch | adjacent | 2,026 | https://nvidia-isaac.github.io/video_to_data/hydra-0/ | https://arxiv.org/abs/2608.18077 | null | null | Generalist action-flow world model trained after filtering 2,202 hours of multi-embodiment video; reports 90.4% lower robot-motion error, 60.2% lower object-motion error, and r=0.96 replay correlation on RoboLab | world-modeling; robot-control; human-to-robot-transfer; video-generation | robot-camera-video; human-video; optical-flow; robot-actions | NVIDIA, MIT, Brown University, and collaborators | null | 2026-08-23 | https://nvidia-isaac.github.io/video_to_data/hydra-0/ | null | null | null | null | null | null | null | null | The paper and project page are public, but the project marks code and model artifacts as coming soon. | General multi-embodiment robot/human video rather than a specifically egocentric corpus. | null |
DECOWAM | model | 2026-08 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.20114 | https://arxiv.org/abs/2608.20114 | null | null | Decouples camera ego-motion, mobile-base motion, and arm actions in a whole-body world-action model; introduces ARMDOG and reports a 21.7% action-MSE reduction with 25.95M trainable adaptation parameters | world-action-models; mobile-manipulation; whole-body-control; video-generation | robot-camera-video; robot-actions; proprioception; natural-language | Siyuan Ma, Boshi Zhang, Yutian Zhang, Qinglian Wu, Jiaqi Zhai, Dong Wei, and Qiaojun Yu | null | 2026-08-23 | null | null | null | null | null | null | null | null | null | The paper is public, but the ARMDOG dataset, implementation, checkpoints, and license were not linked. | Moving robot-camera perception rather than human egocentric capture. | null |
RoboEdit | model | 2026-08 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.18948 | https://arxiv.org/abs/2608.18948 | null | null | Edits human manipulation video into action-consistent robot video with aligned 3D states; the reported RoboEdit-14M corpus contains 174K paired videos, 14M frames, and seven robot embodiments | human-to-robot-editing; cross-embodiment-transfer; video-generation; robot-learning | human-video; generated-video; hand-pose; robot-actions | Yaowei Guo, Zeng Tao, Yuxin Jiang, Yunuo Chen, Zhiyang Dou, Yuxiang Ma, Yin Yang, Demetri Terzopoulos, Ying Jiang, and Chenfanfu Jiang | null | 2026-08-23 | null | null | null | null | null | null | null | null | null | The paper is public, but no code, model, RoboEdit-14M package, or license was linked. | General human manipulation video; a wearable first-person viewpoint is not required. | null |
HOST | model | 2026-07 | arXiv | partial | adjacent | 2,026 | https://host-site.host-robotics.workers.dev/ | https://arxiv.org/abs/2607.20033 | https://github.com/CGuangyan-BIT/HOST | null | Inference-time acquisition from one human video in 29 seconds on average, reporting 62% success across novel manipulation tasks; public code covers preprocessing, alignment, coupling, and policy training, with a 15.9 GB checkpoint | human-to-robot-transfer; imitation-learning; robot-learning; world-action-models | human-video; robot-camera-video; robot-actions; natural-language | Guangyan Chen, Meiling Wang, Te Cui, Zichen Zhou, Qi Shao, Shalfun Li, Hang Su, Roy Gan, Hao Wang, Mengyin Fu, Yi Yang, and Yufeng Yue | null | 2026-08-23 | https://host-site.host-robotics.workers.dev/ | https://huggingface.co/Guangyan/HOST | null | null | null | null | null | null | null | Training/preprocessing code and an MIT-licensed policy checkpoint are public. The checkpoint is trained robot-to-robot, the root code repository has no repository-wide license, and user normalization metadata is still required for evaluation. | Human demonstration video is central, but the method is not restricted to wearable egocentric capture. | null |
RoboMirror | model | 2025-12 | arXiv | watch | adjacent | 2,025 | https://arxiv.org/abs/2512.23649 | https://arxiv.org/abs/2512.23649 | null | null | Retargeting-free video-to-humanoid locomotion from raw egocentric or third-person video, using VLM-derived motion intent to condition a diffusion policy; reports 80% lower control latency and 3.7% higher task success | humanoid-control; video-language; cross-embodiment-transfer; motion-generation | egocentric-video; exocentric-video; natural-language; robot-actions | Zhe Li, Boan Zhu, Yangyang Wei, Shuanghao Bai, Yuheng Ji, Yibo Peng, Tao Huang, Pengwei Wang, Zhongyuan Wang, S.-H. Gary Chan, Chang Xu, Cheng Chi, Jianfei Yang, and Shanghang Zhang | null | 2026-08-23 | null | null | null | null | null | null | null | null | null | The revised paper is public, but no official implementation, checkpoint, evaluation data, or license was linked. | Supports egocentric input, but targets humanoid locomotion rather than wearable-video understanding. | null |
HiPHI | dataset | 2026-08 | Hugging Face | request | adjacent | 2,026 | https://noitom-robotics.github.io/hiphi/ | null | null | ModalityNet Open Research License | 617.5 hours and 200.1M optical-motion-capture frames from 132 performers, with 22 Frames, 214 Frame-LUs, and 245.7 hours of hand-object interaction; the gated repository is about 516 GB | human-motion; hand-object-interaction; humanoid-control; robot-learning | mocap; human-motion; hand-object-interaction; object-pose | Noitom Robotics | noitomrobotics on Hugging Face | 2026-08-23 | https://noitom-robotics.github.io/hiphi/ | https://huggingface.co/datasets/noitomrobotics/HiPHI | null | null | null | null | https://huggingface.co/datasets/noitomrobotics/HiPHI | Automatic Hugging Face gate under the ModalityNet Open Research License | https://huggingface.co/datasets/noitomrobotics/HiPHI | The complete 32-archive, roughly 516 GB repository is gated for research use; it is motion capture rather than egocentric video. | Optical full-body and hand-object motion capture without a central first-person camera stream. | null |
NVIDIA Video-to-Data Robot Dexterity Dataset | dataset | 2026-08 | Hugging Face | open | adjacent | 2,026 | https://huggingface.co/datasets/nvidia/video-to-data-robot-dexterity-task-library-and-dataset | null | https://github.com/nvidia-isaac/video_to_data | cc-by-4.0 | 1,215 TACO-derived human motion sequences retargeted into 121,500 bimanual Sharpa robot episodes and 1,215 motion parquets, totaling 924.8 GB in Parquet and LeRobot v3 formats | retargeting; bimanual-manipulation; robot-learning; imitation-learning | human-video; robot-camera-video; trajectories; robot-actions; lerobot | NVIDIA | NVIDIA on Hugging Face | 2026-08-23 | null | https://huggingface.co/datasets/nvidia/video-to-data-robot-dexterity-task-library-and-dataset | null | null | null | null | https://huggingface.co/datasets/nvidia/video-to-data-robot-dexterity-task-library-and-dataset/blob/main/README.md | null | null | null | Synthetic robot executions derived from TACO human demonstrations rather than a new wearable-video corpus. | null |
RoboLab-EgoX | dataset | 2026-08 | Hugging Face | open | adjacent | 2,026 | https://huggingface.co/datasets/Voxel51/RoboLab-EgoX | https://arxiv.org/abs/2604.09860 | null | apache-2.0 | 4,000 simulation episodes and six hours across 28 RoboLab tasks and 99 scenes, including 632 successes and 3,368 failures with wrist-view RGB-D, two exterior RGB-D views, calibration, arm telemetry, actions, and language | robot-learning; manipulation; failure-understanding; benchmark | robot-camera-video; exocentric-video; depth; camera-calibration; robot-actions; natural-language | DAVIAN Robotics, Voxel51, and NVIDIA RoboLab | Voxel51 on Hugging Face | 2026-08-23 | null | https://huggingface.co/datasets/Voxel51/RoboLab-EgoX | null | null | null | null | https://huggingface.co/datasets/Voxel51/RoboLab-EgoX/blob/main/README.md | null | null | null | Robot wrist-view simulation benchmark rather than human wearable capture. | null |
MERIT | model | 2026-08 | ECCV 2026 Oral | watch | egocentric | 2,026 | https://choi-yeeun.github.io/MERIT/ | https://arxiv.org/abs/2608.07663 | null | null | Simple episodic memory with four retrieval keys per 30-second clip and query-time neighbor filtering; reports 71.2 accuracy on EgoLifeQA, 71.8 on LVBench, and 77.7 on Video-MME Long with a GPT-5 backbone | long-video-qa; episodic-memory; temporal-retrieval; video-language | egocentric-video; audio; natural-language; memory-index | Yeeun Choi, Youngbeom Yoo, Joon-Young Lee, Hyolim Kang, and Seon Joo Kim | null | 2026-08-18 | https://choi-yeeun.github.io/MERIT/ | null | null | null | null | null | null | null | null | The ECCV 2026 Oral paper and project page are public. The page's code link currently returns 404, so implementation and license remain unverified. | null | null |
SCOUT | model | 2026-08 | ACM MM 2026 | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.07959 | https://arxiv.org/abs/2608.07959 | null | null | Recovery-aware Chain-of-Tool-Thought agent for hours-to-days egocentric video, with adaptive region switching, uncertainty-prioritized UPS-GRPO, and turn-level temporal-alignment rewards | long-video-qa; temporal-retrieval; agentic-reasoning; video-language | egocentric-video; natural-language; tool-trajectories | Keyang Zhong, Kuo Wang, Peng Liu, Quanlong Zheng, Junlin Xie, Zhijia Liang, Yanhao Zhang, and Guanbin Li | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The paper confirms acceptance at ACM Multimedia 2026, but no code, model, training trajectories, or license is linked. | null | null |
EgoTrack3D | model | 2026-08 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.08016 | https://arxiv.org/abs/2608.08016 | null | null | Modular dynamic 3D object tracker that lifts 2D masks into a global frame and reports an 11% PCL improvement over the strongest baseline on Aria Digital Twin | 3d-object-tracking; dynamic-scene-understanding; scene-memory; spatial-reasoning | egocentric-rgb; segmentation; depth; 3d-tracks | Jan Kulik, Bjarni Dagur Thor Karason, Yung-Hsu Yang, Boyang Sun, Marc Pollefeys, and Xi Wang | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The paper is public, but no code, checkpoint, or standalone benchmark release is linked. | null | null |
Ego-OSCAR | toolkit | 2026-08 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.08285 | https://arxiv.org/abs/2608.08285 | null | null | Sub-$200 open-hardware stereo-inertial head-mounted capture design; the paper reports about 550 hours per camera, 1,462 sessions, synchronized 6-axis IMU, dense captions, calibration, and per-frame 3D hands | data-collection; sensor-synchronization; hand-object-interaction; robot-learning | stereo-video; imu; camera-calibration; action-captions; 3d-hand-pose | Gunjan Paul, Senthil Palanisamy, Satpal Singh Rathore, Pratyush Kumar Patnaik, Shubhanshu Khatana, and Abhishek Anand | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The paper states that hardware, software, and data are open-sourced, but it provides no verifiable artifact URL; availability and reuse terms therefore remain on watch. | null | null |
VideoNIG | benchmark | 2026-08 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.08596 | https://arxiv.org/abs/2608.08596 | null | null | Goal-oriented navigation-instruction benchmark with 60K simulated ego-view tour videos and 37K text- or image-goal prompts, evaluated by language, spatial-consistency, and navigation-execution tests | navigation-instruction-generation; spatial-reasoning; vision-language-navigation | simulated-first-person-video; natural-language; navigation-trajectories | Fangdi Li, Juncheng Liao, Changxu Cheng, Jiazhi Wang, Senda Chen, Tao Wang, and Wuyue Zhao | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The work is adjacent because its first-person tours are simulator-generated navigation-agent views rather than human wearable recordings. No data or code release is linked. | null | null |
EgoHieraLoc | model | 2026-08 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.09656 | https://arxiv.org/abs/2608.09656 | null | null | Unified VQL-2D/VQL-3D framework combining segmentation-guided query parsing, deformable correlation filtering, regional feedback, and geometry-semantic confidence for credible multi-view 3D estimates | visual-query-localization; 2d-localization; 3d-localization; object-retrieval | egocentric-video; segmentation; depth; object-queries | Yifei Cao, Guolong Wang, Mingliang Hou, Xiya Bu, Daming Liu, and Yu Liu | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The manuscript is under review and links no code, checkpoint, or evaluation package. | null | null |
Fast-Slow ReAct ObjectNav | model | 2026-08 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.09816 | https://arxiv.org/abs/2608.09816 | null | null | Hierarchical zero-shot ObjectNav agent with a reactive value-map controller, coordinate-anchored semantic memory, pose-tagged first-person keyframes, and bounded VLM deliberation; reports 68.75% HM3D and 47.29% MP3D success | object-goal-navigation; spatial-memory; agentic-reasoning | robot-camera-video; semantic-map; pose; natural-language | Zhaochen Lan, Zhi Yang, Yuxiang Fu, and Mengxiang Lin | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The work is adjacent because its recalled first-person views come from a navigation robot, not a human wearer. No implementation is linked. | null | null |
R4DSG | model | 2026-08 | ACM MM 2026 | watch | egocentric | 2,026 | https://dualtransparency.github.io/R4DSG/ | https://arxiv.org/abs/2608.11017 | null | null | Relative 4D scene-graph memory evaluated on 255 object questions from EgoLifeQA; reports a 6.7-point overall gain and a 12.5-point gain on when questions over EgoRAG-Text | long-video-qa; object-memory; temporal-retrieval; spatial-reasoning | egocentric-video; scene-graphs; object-tracks; natural-language | Ke Ma, Yamin Mao, Weiming Li, Shuai Tan, Yijie Zhong, Hao Chen, Haofen Wang, and Meng Wang | null | 2026-08-18 | https://dualtransparency.github.io/R4DSG/ | null | null | null | null | null | null | null | null | The ACM Multimedia 2026 paper and project page are public. The linked GitHub repository contains the project site, not the method implementation or evaluation release. | null | null |
CrossTracer | model | 2026-08 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.06688 | https://arxiv.org/abs/2608.06688 | null | null | Cross-embodiment navigation framework that proposes normalized image-plane traces and adapts them with embodiment-conditioned residuals; reports 45.68 on NaviTrace and real wheeled/legged deployment | vision-language-navigation; cross-embodiment-navigation; traversability | robot-camera-video; image-waypoints; segmentation; robot-actions | Yao Wang, Siyuan Wang, Zhirui Sun, Wenzheng Chi, Liang Lin, Jiankun Wang, and Wenjun Xu | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The work is adjacent because its egocentric observations are robot-camera views. No NaviTrace data, code, or checkpoint is linked. | null | null |
AtlasVLA | model | 2026-08 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.06729 | https://arxiv.org/abs/2608.06729 | null | null | Wrist-camera VLA with persistent voxel-hashed 4D world memory and ego working-state memory; reports 9.4-point LIBERO-Long and 17.5-point real-world long-horizon gains over multi-view baselines | vla; long-horizon-manipulation; spatial-memory; task-progress | wrist-camera-video; 4d-memory; robot-actions; natural-language | Guiyu Zhao, Longteng Guo, Yanghong Mei, Zilin Zhu, Yu Zhang, Bin Cao, Mingming Yu, Xingjian He, Jie Jiang, and Jing Liu | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The work is adjacent because it uses a robot wrist camera rather than human wearable capture. No code or checkpoint is linked. | null | null |
WNM-3D | model | 2026-08 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.07267 | https://arxiv.org/abs/2608.07267 | null | null | Continuous-VLN world navigation model conditioning future-view and action generation on 3D scene tokens reconstructed from monocular ego-view history, evaluated on GN-Bench | vision-language-navigation; world-modeling; future-view-generation; closed-loop-control | robot-camera-video; monocular-rgb; 3d-scene-tokens; robot-actions | Yuehao Huang, Yunzi Wu, Xiaotao Zhang, Xinhai Li, Jiankun Dong, Jiajun Lv, Chi Zhang, Chenjia Bai, Yong Liu, and Xuelong Li | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The work is adjacent because the egocentric history is captured by a navigation agent. No code, checkpoint, or training-data release is linked. | null | null |
SLT 2026 SmartGlasses Challenge | challenge | 2026-08 | IEEE SLT 2026 Challenge | request | egocentric | 2,026 | https://aslp-lab.github.io/SmartGlasses/ | https://arxiv.org/abs/2608.12034 | https://github.com/ASLP-lab/Smart-Glass-Challenge | null | 106.98 hours across 714 four-channel smart-glasses audio sessions: 518 dyadic sessions totaling 44.95 hours and 196 multi-party meetings totaling 62.03 hours | speaker-attributed-asr; spoken-language-understanding; dialogue-understanding; meeting-understanding | egocentric-audio; multichannel-audio; speech; speaker-labels; natural-language | Dehui Gao, Zhixian Zhao, Zhennan Lin, Yujie Liao, Yuhang Dai, Yike Zhu, Longshuai Xiao, Hui Bu, Xin Xu, Xie Chen, Shuai Wang, Liumeng Xue, Zhonghua Fu, Jun Du, Eng-Siong Chng, Jun Zhou, and Lei Xie | null | 2026-08-18 | https://aslp-lab.github.io/SmartGlasses/ | null | null | null | null | null | null | Training, development, and test data are distributed to registered challenge participants | https://aslp-lab.github.io/SmartGlasses/ | The challenge site and evaluation toolkit are public, but audio access requires registration and the public repository does not declare a license. | null | null |
HandEdit | benchmark | 2026-08 | arXiv | open | egocentric | 2,026 | https://handedit.github.io/ | https://arxiv.org/abs/2608.12122 | https://github.com/HandEdit/HandEdit | cc-by-nc-nd-4.0 | 300K+ egocentric clips and 200M+ human-to-robot editing instances from five source datasets, covering 26 embodiments split into 13 hand-only and 13 hand-arm configurations | human-to-robot-editing; dexterous-manipulation; embodiment-transfer; benchmark | egocentric-images; hand-masks; robot-renders; urdf; editing-pairs | Zhenjie Yang, Xingyu Jiao, Guopeng Zhong, Shuzhe Yang, Shi Che, Chao Wu, Chenyu Jiang, Dongjie Zhang, Yideng Zhang, Zheng Zhang, Muyun Jiang, Haisheng Su, Shuang Jin, Donghang Zhang, Chao Yang, Li Chen, Hongyang Li, Zuxuan Wu, Yu-Gang Jiang, Xiaosong Jia, and Junchi Yan | null | 2026-08-18 | https://handedit.github.io/ | https://huggingface.co/datasets/HandEdit/HandEdit | null | null | null | null | https://huggingface.co/datasets/HandEdit/HandEdit/blob/main/README.md | null | null | The full Hugging Face dataset and evaluation toolkit are public. Dataset files use CC-BY-NC-ND-4.0; the GitHub repository does not currently expose a separate code license. | null | null |
Proxemic Risk from Robot Ego Images | model | 2026-08 | ECCV 2026 EMR Workshop | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.12515 | https://arxiv.org/abs/2608.12515 | null | null | Four-level proxemic-danger classification study comparing InternVL, Qwen-VL, and SmolVLM under three prompts and two QLoRA rounds, with a separate person-localization analysis | proxemic-risk; robot-safety; spatial-grounding; classification | robot-camera-images; natural-language; safety-labels | Vladyslava Rudas and Dmytro Kuzmenko | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The work is adjacent because the images come from a robot's viewpoint rather than a human wearable camera. No image set, labels, or code is linked. | null | null |
EgoCITE | model | 2026-08 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.12627 | https://arxiv.org/abs/2608.12627 | null | null | Context-augmented multi-view memory indexing and temporal-intent-aware evidence retrieval evaluated on EgoLifeQA, EgoMem, and EgoR1-Bench; reports 4.4-14.2-point gains and 36x lower cost than long-context agents | long-video-qa; episodic-memory; temporal-retrieval; agentic-reasoning | egocentric-video; audio; captions; speech-transcripts; memory-index | Le Zhang and Ke Sun | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The paper is public, but no implementation, index release, checkpoint, or license is linked. | null | null |
HumanoidVLN | benchmark | 2026-08 | arXiv | watch | adjacent | 2,026 | https://humanoid-vln.github.io/ | https://arxiv.org/abs/2608.12860 | null | null | NVIDIA Isaac Sim benchmark spanning four humanoid embodiments and 933 collision-aware reference episodes, each with one fine-grained and three stylistic coarse instructions, plus a 20-episode Unitree G1 sim-to-real pilot | vision-language-navigation; humanoid-control; sim-to-real; cross-embodiment-evaluation | simulated-robot-video; natural-language; navigation-trajectories; humanoid-state | Quan-Dung Pham, Anh Dao, The-Anh Nguyen, Minh Nguyen-Dinh, Phuong Nam Dang, Tri Pham, Hung Tran, Bach Dao, Tuyen P. Le, Truong Nguyen, and Quan Nguyen | null | 2026-08-18 | https://humanoid-vln.github.io/ | null | null | null | null | null | null | null | null | The work is adjacent because its first-person views are simulated humanoid observations. The authors state that code, benchmark, and data will be released upon acceptance, so artifacts are not yet public. | null | null |
EgoPHI | model | 2026-08 | ECCV 2026 | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.13014 | https://arxiv.org/abs/2608.13014 | null | null | Joint dense contact-map and 3D force estimation from one egocentric RGB image and object geometry, trained with simulation supervision and tested on a real set from eight participants and two instrumented objects | contact-estimation; force-estimation; hand-object-interaction; physical-reasoning | egocentric-rgb; hand-mesh; object-mesh; contact-maps; force-fields | Andela Ilic, Rachel Schuchert, Yijing Jiang, and Christian Holz | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The ECCV 2026 paper is public, but the simulation pipeline, model, and eight-participant real-world evaluation set are not linked. | null | null |
H2R-Bench | benchmark | 2026-08 | arXiv | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.13049 | https://arxiv.org/abs/2608.13049 | null | null | Human-to-robot video-generation benchmark covering 11 models, six manipulation families, two robot embodiments, and five dimensions spanning goals, actions, contact, embodiment, and visual quality | human-to-robot-transfer; video-generation; world-modeling; benchmark | egocentric-human-video; generated-robot-video; action-labels; contact-annotations | Dingyi Rong, Yue Shi, Chaofan Ma, Jiezhang Cao, Zongrui Wang, Zeyu Zhang, Yao Mu, Guangtao Zhai, and Ning Liu | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The benchmark paper is public, but evaluation instances, annotations, scripts, and license are not linked. | null | null |
Clinical Freezing-of-Gait with Egocentric Vision | model | 2026-08 | ECCV 2026 Workshop | watch | egocentric | 2,026 | https://arxiv.org/abs/2608.13283 | https://arxiv.org/abs/2608.13283 | null | null | At-home study of 13 participants with Parkinson's disease using synchronized egocentric video, wearable IMUs, and expert freezing-of-gait labels under leave-one-subject-out evaluation | freezing-of-gait-detection; clinical-motion-understanding; wearable-sensing; sensor-fusion | egocentric-video; imu; clinical-labels | Vayalet Stefanova, Diwas Lamsal, Margot Genbrugge, Maxim Yudayev, Christian Schlenstedt, Moran Gilat, Bart Vanrumste, and Benjamin Filtjens | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The ECCV 2026 workshop paper is public, but the sensitive at-home recordings, labels, and code are not released. | null | null |
ContactGuard | model | 2026-08 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.13438 | https://arxiv.org/abs/2608.13438 | null | null | Pre-contact execution monitor that predicts short-horizon visual latents under a robot policy's planned action chunk and aborts likely failures before wrist-camera contact | failure-prediction; execution-monitoring; contact-rich-manipulation; robot-safety | wrist-camera-video; robot-actions; latent-visual-features | Gehan Zheng, Matthew Johnson-Roberson, and Weiming Zhi | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The work is adjacent because it uses robot wrist and multi-view cameras rather than human wearable capture. No code or data is linked. | null | null |
AdvDex | model | 2026-08 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.14028 | https://arxiv.org/abs/2608.14028 | null | null | Human-and-robot VLA framework introducing OmniShare multimodal demonstrations, a canonical SE(3) wrist plus 15-joint hand action space, and domain-adversarial embodiment alignment | dexterous-manipulation; human-to-robot-transfer; vla; cross-embodiment-learning | human-demonstrations; robot-demonstrations; hand-kinematics; tactile; robot-actions | Zhiyue Zhao, Jingyi Wu, Hairuo Liu, Mingyu Liu, Liyang Li, Hengdi Zhang, Tong He, and Zhengxue Cheng | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The work is adjacent because the paper does not establish OmniShare as a human wearable first-person corpus. Dataset scale, files, code, and license are not publicly linked. | null | null |
OccPlanner | model | 2026-08 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2608.14160 | https://arxiv.org/abs/2608.14160 | null | null | Pixel-goal diffusion planner conditioned on local 3D occupancy, with L3ROcc supervision reconstructed from monocular robot videos; reports 71.55% average success at 5-8 m versus 20.81% for NavDP | pixel-goal-navigation; occupancy-prediction; diffusion-planning; sim-to-real | robot-camera-video; local-3d-occupancy; pixel-goals; robot-actions | Binling Huang, Nianjin Ye, Xi Yang, Liang Hu, Zhou Huang, Shuang Wei, Longrui Yang, Yanchi Chen, and Lanpeng Jia | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The work is adjacent because L3ROcc and OccPlanner use robot-view navigation video. No code, model, or L3ROcc release is linked. | null | null |
HomeSafeBench | benchmark | 2025-09 | arXiv | open | adjacent | 2,025 | https://github.com/BITHLP/HomeSafeBench | https://arxiv.org/abs/2509.23690 | https://github.com/BITHLP/HomeSafeBench | cc-by-4.0 | VirtualHome free-exploration safety benchmark with 1,000 human-validated test tasks, 3,400 training tasks, five hazard categories, and 3,158 CueBack supervision trajectories | home-safety-inspection; embodied-reasoning; navigation; risk-recognition | simulated-first-person-video; natural-language; navigation-trajectories; hazard-labels | Jiashu Yao, Haoyu Wen, Siyuan Gao, Yuhang Guo, Zeming Liu, and Heyan Huang | null | 2026-08-18 | null | https://huggingface.co/datasets/Navinvue/HomeSafeBench | null | null | null | null | https://github.com/BITHLP/HomeSafeBench#license | null | null | The benchmark data and CueBack supervision are public under CC-BY-4.0, and the evaluation code is MIT-licensed. It is adjacent because observations are rendered agent views, not wearable recordings. | null | null |
Point of View Robot Sociability Study | model | 2026-03 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2603.28272 | https://arxiv.org/abs/2603.28272 | null | null | Immersive-VR user study comparing allocentric, egocentric-proximal, and egocentric-distal judgments of identical robot trajectories and a communicative head-nod gesture | human-robot-interaction; social-navigation; perspective-evaluation; proxemics | virtual-reality; simulated-first-person-view; robot-trajectories; human-ratings | Subham Agrawal, Aftab Akhtar, Nils Dengler, and Maren Bennewitz | null | 2026-08-18 | null | null | null | null | null | null | null | null | null | The work is adjacent because it studies simulated pedestrian viewpoints rather than providing human wearable-camera data. No study package or code is linked. | null | null |
VITRA Egocentric Instruction Sidecars | dataset | 2026-08 | Hugging Face | partial | egocentric | 2,026 | https://huggingface.co/datasets/MIT-Media-Lab/vitra-instructions-qwen35-122b-egocentric-v1 | null | null | null | 175,327 Qwen3.5-generated candidate hand-action instructions with 95,831 accepted rows across GigaHands, HOT3D, and OakInk2; 317 repository files retain exact source revisions and paths | vla; instruction-generation; hand-object-interaction; dataset-curation | natural-language; action-instructions; metadata; source-pointers | MIT Media Lab VLA-HAND contributors | MIT-Media-Lab on Hugging Face | 2026-08-18 | null | https://huggingface.co/datasets/MIT-Media-Lab/vitra-instructions-qwen35-122b-egocentric-v1 | null | null | null | null | null | null | null | All annotation sidecars are public and source videos remain in their original repositories. The card declares no standalone license, so source-dataset terms and annotation reuse rights must be checked separately. | null | null |
InHandPlus Wrist-View Camera and IMU Sample | dataset | 2026-08 | Hugging Face | request | egocentric | 2,026 | https://huggingface.co/datasets/inhandplus/egocentric-wrist-view-camera-IMU-data | null | null | cc-by-nc-4.0 | Auto-gated 30-episode, 119.7 MiB early sample for cloth folding and organization, with wrist video, synchronized 6-axis raw IMU, and LeRobot v3 metadata but no action labels | human-demonstrations; sensor-fusion; visual-pretraining; manipulation | wrist-camera-video; imu; lerobot; metadata | InHandPlus | null | 2026-08-18 | null | https://huggingface.co/datasets/inhandplus/egocentric-wrist-view-camera-IMU-data | null | null | null | null | https://huggingface.co/datasets/inhandplus/egocentric-wrist-view-camera-IMU-data/blob/main/README.md | Free automatic Hugging Face access after a six-question form | https://huggingface.co/datasets/inhandplus/egocentric-wrist-view-camera-IMU-data | The complete 30-episode sample is hosted on Hugging Face under CC-BY-NC-4.0, but download requires the automatic access form and this early release has no action labels. | null | null |
InfoBayAI Egocentric Video Sample | dataset | 2026-08 | Hugging Face | request | egocentric | 2,026 | https://huggingface.co/datasets/InfoBayAI/egocentric_video | null | null | cc-by-4.0 | Auto-gated 31.87 GB repository containing 10 MP4 sample videos and one metadata Parquet file; the card separately advertises a 229,189-hour collection that is not present in the public manifest | video-understanding; action-recognition; vision-language; temporal-reasoning | egocentric-video; metadata | InfoBayAI | null | 2026-08-18 | null | https://huggingface.co/datasets/InfoBayAI/egocentric_video | null | null | null | null | https://huggingface.co/datasets/InfoBayAI/egocentric_video/blob/main/README.md | Automatic Hugging Face access agreement | https://huggingface.co/datasets/InfoBayAI/egocentric_video | The status and scale describe the files actually visible through the Hugging Face API. The card's 229,189-hour headline and 1K-10K size tag are not substantiated by the 10-video repository manifest. | null | null |
UniDataPro Egocentric Video Sample | dataset | 2026-08 | Hugging Face | request | egocentric | 2,026 | https://huggingface.co/datasets/UniDataPro/egocentric-video | null | null | cc-by-nd-4.0 | Commercial card claims 4,050 hours across Pico VR, motion trackers, IMU, quaternion orientation, and ZED stereo capture; the public 2.33 GB repository contains one video, one metadata row, and one tracking file | hand-object-interaction; human-motion; 3d-pose-estimation; robot-learning | egocentric-video; stereo-video; imu; motion-tracking; metadata | UniDataPro | null | 2026-08-18 | null | https://huggingface.co/datasets/UniDataPro/egocentric-video | null | null | null | null | https://huggingface.co/datasets/UniDataPro/egocentric-video/blob/main/README.md | One preview is public; the full collection requires a commercial purchase request | https://unidata.pro | CC-BY-ND-4.0 is declared on the public Hugging Face card. Full-corpus availability, file counts, and commercial terms cannot be verified from the one-video preview. | null | null |
RekaDaily-10k (raw) | dataset | 2026-08 | Hugging Face | open | egocentric | 2,026 | https://huggingface.co/datasets/RekaAI/RekaDaily-10k-raw | null | null | apache-2.0 | Publisher reports 10,865 hours, 412,050 videos, 11,127 WebDataset shards, and 80 TB of first-person footage from head-mounted and handheld phones; some recordings include synchronized IMU | egocentric-pretraining; everyday-activity; action-understanding; robot-learning | egocentric-video; video; metadata | Reka AI and Claru | null | 2026-10-29 | null | https://huggingface.co/datasets/RekaAI/RekaDaily-10k-raw | null | null | null | null | https://huggingface.co/datasets/RekaAI/RekaDaily-10k-raw/blob/main/README.md | null | null | The publisher calls this the full release and notes additional recordings may be appended. Counts are from the dataset card and may change. | null | null |
RekaDaily-10k (processed) | dataset | 2026-09 | Hugging Face | open | egocentric | 2,026 | https://huggingface.co/datasets/RekaAI/RekaDaily-10k-processed | null | null | apache-2.0 | Dataset card reports 6,373,064 clips and 10,200 hours; captions and question-answer exchanges are model-generated | egocentric-pretraining; everyday-activity; action-understanding; robot-learning | egocentric-video; video; captions; metadata | Reka AI and Claru | Reka AI on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/RekaAI/RekaDaily-10k-processed | null | null | null | null | https://huggingface.co/datasets/RekaAI/RekaDaily-10k-processed | null | null | Publisher reports that captions and Q&A are model-generated and not human-verified. Clips link to their source recordings; counts can change as processing continues. | null | RekaDaily-10k (raw) |
Gen-HumanEgo | dataset | 2026-09 | Hugging Face | request | egocentric | 2,026 | https://huggingface.co/datasets/genrobot2025/Gen-HumanEgo | null | null | cc-by-sa-4.0 | Publisher reports 1,847.7 hours, 44,632 episodes, 10,257 tasks, six synchronized RGB cameras, reconstructed 3D hands, depth, and hierarchical annotations | hand-object-interaction; robot-learning; human-demonstrations | egocentric-video; hand-pose; depth; action-labels | GenRobot | GenRobot on Hugging Face | 2026-10-29 | https://www.gen-robot.org/blog/gen-human-ego | https://huggingface.co/datasets/genrobot2025/Gen-HumanEgo | null | null | null | null | https://huggingface.co/datasets/genrobot2025/Gen-HumanEgo | Hugging Face form; access automatically granted on submission | null | The dataset card describes model-derived hand, depth, and semantic signals; access requires submitting a form. | null | null |
EgoBrain-Mini | dataset | 2026-08 | Hugging Face | request | egocentric | 2,026 | https://huggingface.co/datasets/ut-vision/EgoBrain-Mini | https://arxiv.org/abs/2506.01353 | null | cc-by-nc-4.0 | 40 participants and 1,562 action segments with synchronized first-person video, EEG, IMU, and audio; released subset excludes original raw video | action-recognition; sensor-fusion; everyday-activity | egocentric-video; audio; imu; eeg | University of Tokyo and Microsoft Research Asia EgoBrain team | UT-Vision on Hugging Face | 2026-10-29 | null | https://huggingface.co/datasets/ut-vision/EgoBrain-Mini | null | null | null | null | https://huggingface.co/datasets/ut-vision/EgoBrain-Mini | Hugging Face information-sharing agreement | null | All 40 subjects were announced on 2026-10-13. The release contains processed phase-2 video, not original raw video; IMU rates vary by subject. | null | EgoBrain |
CapMem | benchmark | 2026-09 | EMNLP 2026 | watch | egocentric | 2,026 | https://arxiv.org/abs/2609.17688 | https://arxiv.org/abs/2609.17688 | null | null | Paper reports 75 videos totaling 33.7 hours and 1,000 human-annotated multiple-choice questions across 16 scenarios | memory; qa; video-language | egocentric-video; captions | Dingli Liang and collaborators | null | 2026-10-29 | null | null | null | null | null | null | null | null | null | Paper reports the benchmark; public data and evaluation-code availability remain unverified. | null | null |
Stereo-550 | dataset | 2026-07 | Hugging Face | open | egocentric | 2,026 | https://huggingface.co/datasets/fpvlabs/stereo-550 | null | null | FPV Labs research-use license | 550 hours per camera (about 1,100 stereo camera-hours), 1,462 calibrated stereo sessions, 209,315 dense action segments, 460 verbs, 32,630 object phrases, and synchronized 6-axis IMU for 1,271 sessions | action-segmentation; video-language; hand-object-interaction; robot-learning; sensor-fusion | stereo-video; camera-calibration; imu; action-labels; captions | FPV Labs | null | 2026-08-09 | null | https://huggingface.co/datasets/fpvlabs/stereo-550 | null | null | null | null | https://fpvlabs.ai/license | null | null | The dataset repository is ungated and the file structure is documented. Reuse is governed by the custom FPV Labs research-use license rather than a standard open-source data license. | null | null |
Ego500 | dataset | 2026-08 | Hugging Face | request | egocentric | 2,026 | https://huggingface.co/datasets/humanarchive/ego500 | null | null | cc-by-nc-4.0 | Card reports 500 hours of first-person work video, more than 310,000 structured action annotations, 390 task-environment combinations, and 60+ environments | action-segmentation; procedure-understanding; hand-object-interaction; robot-learning | egocentric-video; action-labels; hand-object-interaction; metadata | Human Archive | null | 2026-08-09 | https://ego500.humanarchive.ai | https://huggingface.co/datasets/humanarchive/ego500 | null | null | null | null | https://huggingface.co/datasets/humanarchive/ego500/blob/main/README.md | Manual Hugging Face access request; non-commercial use | null | The repository is manually gated and its card still says data upload pending, so the headline figures describe the announced release rather than a fully verified public file manifest. | null | null |
EgoSuite-Open100K | dataset | 2026-08 | Hugging Face | request | egocentric | 2,026 | https://huggingface.co/collections/LightwheelAI/egosuite-open100k | null | null | commercial-training-no-resale-v1.0 | Controlled-access collection advertising 100,000 hours across EgoStandard (90K head-view hours) and EgoPro (10K head-and-wrist hours), with hand pose, optional full-body pose, semantic annotations, and 15K+ tasks/scenes; current Hugging Face storage totals about 13.2 TB across its three repositories | egocentric-pretraining; human-motion; hand-tracking; vla; world-modeling | egocentric-video; wrist-camera-video; hand-pose; body-pose; annotations | Lightwheel AI | LightwheelAI on Hugging Face | 2026-08-23 | https://egosuite100k.lightwheel.ai/ | https://huggingface.co/datasets/LightwheelAI/EgoDemo | null | null | null | null | https://huggingface.co/datasets/LightwheelAI/EgoDemo/blob/main/LICENSE | Manual Hugging Face approval; files are distributed through mutable EgoDemo, EgoStandard, and EgoPro buckets | https://huggingface.co/collections/LightwheelAI/egosuite-open100k | The official collection and all three gated bucket access points are live. The cards describe planned line-level durations and state that each current bucket manifest is the source of truth, so 100K hours is retained as the advertised collection target rather than a fully enumerated local count. | null | null |
ACE-Data-0 | dataset | 2026-07 | arXiv | partial | egocentric | 2,026 | https://ace-data-engine.github.io/ACE-Data-0/ | https://arxiv.org/abs/2607.28625 | null | ACE-Data-0 Research License | Current public snapshot contains 30 room-scale takes, 1,617 videos, 92,095 images, calibration/motion/object/audio/tactile sidecars, 95,238 files, and about 3.61 TB; 150 hours, 17M+ frames, and 75K+ interactions remain the full-release target | human-demonstrations; hand-object-interaction; human-motion; tactile-learning; world-modeling; vla | fisheye-video; exocentric-video; full-body-mocap; hand-mocap; object-pose; audio; tactile; imu | Yukang Cao, Haozhe Xie, Beichen Wen, Runmao Yao, Yinghao Liu, and collaborators at NTU S-Lab and ACE Robotics | null | 2026-08-23 | https://ace-data-engine.github.io/ACE-Data-0/ | https://huggingface.co/datasets/ACERobotics/ACE-Data-0 | null | null | null | null | https://huggingface.co/datasets/ACERobotics/ACE-Data-0/blob/main/LICENSE | Repository files currently download without a Hugging Face gate; custom non-commercial research terms still apply | null | The repository now hosts a substantial 3.61 TB snapshot even though parts of the card still describe the data as unpublished. Keep partial because the current 30-take manifest is smaller than the announced 150-hour release and the card/file-state mismatch remains unresolved. | null | null |
EgoMed-IEMIS | dataset | 2026-06 | arXiv | partial | egocentric | 2,026 | https://huggingface.co/datasets/daizywang/EgoMed-IEMIS | https://arxiv.org/abs/2607.17341 | https://github.com/wdyyyyyy/EgoMed-Agent | null | 523 smart-glasses videos and 173,657 frames with referred-target masks across five medical imaging modalities, 12 targets, and five real-world viewing scenes | segmentation; visual-grounding; medical-assistance; wearable-ai | smart-glasses; egocentric-video; segmentation; masks | Rongjun Ge, Dongyang Wang, Heng Zhu, Zhirui Li, Yang Chen, and Yuting He | null | 2026-08-09 | null | https://huggingface.co/datasets/daizywang/EgoMed-IEMIS | null | null | null | null | null | null | null | All five modality subsets and the method code are public. The card declares license: other and inherits terms from five source medical datasets without stating one unified redistribution license. | null | null |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.