feed.xml

<?xml version="1.0" encoding="UTF-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Awesome Egocentric Atlas updates</title>
  <id>https://chaoyue0307.github.io/awesome-egocentric-atlas/feed.xml</id>
  <link href="https://chaoyue0307.github.io/awesome-egocentric-atlas/feed.xml" rel="self"/>
  <link href="https://chaoyue0307.github.io/awesome-egocentric-atlas/"/>
  <updated>2026-08-09T00:00:00Z</updated>
  <subtitle>Recently added egocentric AI resources, newest first.</subtitle>
<entry>
  <title>VLAff / EgoAffordance</title>
  <id>https://ojh6404.github.io/vlaff/</id>
  <link href="https://ojh6404.github.io/vlaff/"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>EgoAffordance contains 204K episodes with 5.6M visual affordances and 11.6M grasp and trajectory affordances; VLAff predicts heatmaps, grasp poses, and executable trajectories</summary>
</entry>
<entry>
  <title>SoniSpeech</title>
  <id>https://doi.org/10.7298/xjjr-9m85</id>
  <link href="https://doi.org/10.7298/xjjr-9m85"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="dataset"/>
  <summary>34 hours and 18,000 voiced and silent utterances with synchronized ultrasound echoes, audio, and frontal video; 5,356 unique words and full phoneme coverage</summary>
</entry>
<entry>
  <title>SmartRes</title>
  <id>https://arxiv.org/abs/2608.01638</id>
  <link href="https://arxiv.org/abs/2608.01638"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>Dynamic pixel-space resolution routing for Ego4D and EgoIntention grounding, reducing visual tokens by up to 67%, retaining 86.4% of full-resolution performance, and reporting up to 1.66x faster inference</summary>
</entry>
<entry>
  <title>SiMDex</title>
  <id>https://lin-nie.github.io/SiMDex/</id>
  <link href="https://lin-nie.github.io/SiMDex/"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>Similarity-based mining selects about 1.49M task-relevant samples from a pool of roughly 32M egocentric human samples and raises reported dexterous-manipulation success from 47.7% to 61.1% over equal-sized random sampling</summary>
</entry>
<entry>
  <title>RoboReact</title>
  <id>https://arxiv.org/abs/2608.03387</id>
  <link href="https://arxiv.org/abs/2608.03387"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>Agentic pipeline that generates egocentric RGB-D manipulation video from a single observation, extracts geometry-preserving keyframes, retargets them to humanoids, and refines execution through closed-loop VLM guidance</summary>
</entry>
<entry>
  <title>RekaDaily-10k (raw)</title>
  <id>https://huggingface.co/datasets/RekaAI/RekaDaily-10k-raw</id>
  <link href="https://huggingface.co/datasets/RekaAI/RekaDaily-10k-raw"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="dataset"/>
  <summary>Incremental raw release currently containing 7,834 hours, 397,171 videos, 9,836 WebDataset shards, and about 70 TB of unscripted first-person daily-life footage from head-mounted and handheld phones</summary>
</entry>
<entry>
  <title>ROCO IROS 2026 UMI Dataset</title>
  <id>https://huggingface.co/datasets/rocochallenge2025/roco_iros2026_umi_dataset</id>
  <link href="https://huggingface.co/datasets/rocochallenge2025/roco_iros2026_umi_dataset"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="dataset"/>
  <summary>4,928 LeRobot v3 episodes and 7,459,440 frames across five tasks at 30 fps, with stereo first-person video, four hand-camera streams, and 16D bimanual pose and gripper state/action</summary>
</entry>
<entry>
  <title>Omega-0</title>
  <id>https://arxiv.org/abs/2608.06375</id>
  <link href="https://arxiv.org/abs/2608.06375"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>Latent predictive whole-body world-action model supporting ego RGB plus exocentric RGB/depth; reports Omega-HOME with 40+ hours of synchronized household humanoid observations, motion, states, and action latents</summary>
</entry>
<entry>
  <title>Nexdata 10,000-Hour Egocentric Video Dataset</title>
  <id>https://huggingface.co/datasets/Nexdata-AI/10000-Hour-Egocentric-Video-Dataset</id>
  <link href="https://huggingface.co/datasets/Nexdata-AI/10000-Hour-Egocentric-Video-Dataset"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="dataset"/>
  <summary>Commercial card describes 10,000 hours across residential, retail, and office scenes with PICO 4 Ultra 4K stereo video, wrist and ankle IMU, 76-point full-body pose, and dense action annotations</summary>
</entry>
<entry>
  <title>LAWM-3D</title>
  <id>https://arxiv.org/abs/2608.05706</id>
  <link href="https://arxiv.org/abs/2608.05706"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>3D-aware latent-action learning from multiview human video using invariant action tokenization, geometric foundation-model alignment, and RGB-D reconstruction for robot world-model pretraining</summary>
</entry>
<entry>
  <title>JoyAI-RA 0.5</title>
  <id>https://joyai-ra-05.github.io/</id>
  <link href="https://joyai-ra-05.github.io/"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>Vision-Language-World-Action framework aligning action-free human egocentric video, simulation, and robot trajectories through latent and canonical action spaces for generalist manipulation</summary>
</entry>
<entry>
  <title>HumynLabs Egocentric Sample Collection</title>
  <id>https://huggingface.co/humyn-labs</id>
  <link href="https://huggingface.co/humyn-labs"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="collection"/>
  <summary>Five CC-BY-4.0 sample packs spanning synchronized head and dual-wrist video with IMU, APAC stereo and monocular labels, residential voice-over captions, and LATAM residential IMU</summary>
</entry>
<entry>
  <title>HiResNets</title>
  <id>https://arxiv.org/abs/2608.02140</id>
  <link href="https://arxiv.org/abs/2608.02140"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>Native Full-HD recognition architecture using foveal residual streams and log-polar read/write operations to preserve fine egocentric objects without quadratic feature-grid growth</summary>
</entry>
<entry>
  <title>HOPE</title>
  <id>https://subin6.github.io/page-hope</id>
  <link href="https://subin6.github.io/page-hope"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>Hand-centric video transformer that predicts temporally evolving per-vertex hand pressure and contact by unifying tactile-glove, planar-pressure, and contact supervision</summary>
</entry>
<entry>
  <title>GST-Bench</title>
  <id>https://arxiv.org/abs/2608.05747</id>
  <link href="https://arxiv.org/abs/2608.05747"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="benchmark"/>
  <summary>Human-verified global spatial VQA derived from 6,790 minutes of synthetic video, mapping egocentric streams to unseen viewpoints and top-down scenes; the best reported zero-shot VLM scores 42.68 versus 79.08 for humans</summary>
</entry>
<entry>
  <title>EventKitchen</title>
  <id>https://arxiv.org/abs/2608.04865</id>
  <link href="https://arxiv.org/abs/2608.04865"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="dataset"/>
  <summary>5.5 hours from 10 participants in 13 kitchens with egocentric stereo events, synchronized RGB, depth, and IMU, plus 10,762 action segments and 13,482 object boxes</summary>
</entry>
<entry>
  <title>EgoSuite-Open10K</title>
  <id>https://huggingface.co/datasets/xuejf/EgoSuite-Open10K</id>
  <link href="https://huggingface.co/datasets/xuejf/EgoSuite-Open10K"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="dataset"/>
  <summary>Announcement card describes 10,000 selected egocentric and wrist-view clips across six SKUs and seven scene groups, with hand pose, body pose, and semantic annotations</summary>
</entry>
<entry>
  <title>EgoAfford</title>
  <id>https://egoafford.github.io/</id>
  <link href="https://egoafford.github.io/"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="benchmark"/>
  <summary>About 15.5K human-verified images from 2,000 generated multi-step scenes plus 102 real egocentric images across 26 tasks, with plans and role-specific affordance masks</summary>
</entry>
<entry>
  <title>Ego500</title>
  <id>https://huggingface.co/datasets/humanarchive/ego500</id>
  <link href="https://huggingface.co/datasets/humanarchive/ego500"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="dataset"/>
  <summary>Card reports 500 hours of first-person work video, more than 310,000 structured action annotations, 390 task-environment combinations, and 60+ environments</summary>
</entry>
<entry>
  <title>Ego2Robot</title>
  <id>https://www-ye.github.io/ego2robot_blog/</id>
  <link href="https://www-ye.github.io/ego2robot_blog/"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>Pipeline converting curated and in-the-wild egocentric human manipulation video into 18,561 hours of robot-format data across 15 robot morphologies through retargeting, visual synthesis, and quality curation</summary>
</entry>
<entry>
  <title>DreamTraj / MOVE</title>
  <id>https://whathappen0.github.io/DreamTraj/</id>
  <link href="https://whathappen0.github.io/DreamTraj/"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>MOVE provides 5,038 object-centric egocentric 6-DoF trajectories with fine-grained language; DreamTraj decodes trajectories from unrendered video-diffusion latents and reports 4.6x faster inference than generate-then-extract pipelines</summary>
</entry>
<entry>
  <title>Assistant Placement Aria</title>
  <id>https://arxiv.org/abs/2608.00652</id>
  <link href="https://arxiv.org/abs/2608.00652"/>
  <updated>2026-08-01T00:00:00Z</updated>
  <category term="benchmark"/>
  <summary>Synthetic and real indoor scenes with 2D images, 3D point clouds, text descriptions, and annotations for panel placement, sitting suggestion, and TV placement</summary>
</entry>
<entry>
  <title>yyyyywv/egocentric</title>
  <id>https://huggingface.co/datasets/yyyyywv/egocentric</id>
  <link href="https://huggingface.co/datasets/yyyyywv/egocentric"/>
  <updated>2026-07-01T00:00:00Z</updated>
  <category term="dataset"/>
  <summary>About 32.8 GB and 1,019 robot episodes totaling 994,459 frames across seven LeRobot task groups: 800 gripper episodes plus plate, shoe, tea, wash, gift-in-hand, and flower-picking sets, with eye or head and wrist cameras, state, and action streams at 20-25 fps</summary>
</entry>
<entry>
  <title>Xiaomi-Robotics-U0</title>
  <id>https://arxiv.org/abs/2607.11643</id>
  <link href="https://arxiv.org/abs/2607.11643"/>
  <updated>2026-07-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>38B unified embodied-synthesis model spanning multi-view scene generation, embodiment transfer, editing, and video generation; reports improving pi0.5 out-of-distribution manipulation success from 36.9% to 63.2%</summary>
</entry>
<entry>
  <title>Worldscape-MoE</title>
  <id>https://worldscape-moe.com/</id>
  <link href="https://worldscape-moe.com/"/>
  <updated>2026-07-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>Mixture-of-experts diffusion world model unifying camera trajectories, robot actions, and hand-joint action maps across locomotion, manipulation, and egocentric hand-control experiments</summary>
</entry>
<entry>
  <title>World Action Models to Embodied Brains Roadmap</title>
  <id>https://arxiv.org/abs/2607.11689</id>
  <link href="https://arxiv.org/abs/2607.11689"/>
  <updated>2026-07-01T00:00:00Z</updated>
  <category term="survey"/>
  <summary>Roadmap organizing WAM gaps across model roles and representations, objectives and standardization, and system composition, then proposing embodied brains, physical harnesses, shared contracts, and closed-loop post-training</summary>
</entry>
<entry>
  <title>Whareformer</title>
  <id>https://jacobchalk.github.io/Whareformer/</id>
  <link href="https://jacobchalk.github.io/Whareformer/"/>
  <updated>2026-07-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>First learning-based solution for online Out of Sight, Not out of Mind tracking, trained on 56 videos and evaluated on 260 long videos from EPIC-KITCHENS-100, IT3DEgo, and HD-EPIC with updatable appearance and 3D-location memory</summary>
</entry>
<entry>
  <title>Wearable Gait MoCap with Shank-Mounted Ego Cameras</title>
  <id>https://doi.org/10.1038/s41597-026-07657-7</id>
  <link href="https://doi.org/10.1038/s41597-026-07657-7"/>
  <updated>2026-07-01T00:00:00Z</updated>
  <category term="dataset"/>
  <summary>Wearable motion-capture dataset for gait analysis using IMUs and shank-mounted egocentric cameras</summary>
</entry>
<entry>
  <title>WANDA / Worlds in One Demo</title>
  <id>https://wanda.lecar-lab.org/</id>
  <link href="https://wanda.lecar-lab.org/"/>
  <updated>2026-07-01T00:00:00Z</updated>
  <category term="dataset"/>
  <summary>Synthetic-data engine expanding one real demonstration per task into 16,922 LeRobot episodes and about 26.8M timesteps (roughly 248 hours at 30 fps) across five long-horizon mobile-manipulation tasks and diverse generated worlds</summary>
</entry>
<entry>
  <title>WAM-TTT</title>
  <id>https://arxiv.org/abs/2607.06988</id>
  <link href="https://arxiv.org/abs/2607.06988"/>
  <updated>2026-07-01T00:00:00Z</updated>
  <category term="model"/>
  <summary>Test-time training framework that steers frozen world-action models from unlabeled raw human videos by adapting a lightweight memory through self-supervised video prediction and paired human-robot meta-training</summary>
</entry>
</feed>