name stringlengths 3 82 | kind stringclasses 7
values | released stringclasses 110
values | venue stringclasses 185
values | status stringclasses 5
values | scope stringclasses 1
value | year int64 2.01k 2.03k | url stringlengths 19 150 | paper stringlengths 26 109 ⌀ | code stringclasses 107
values | license stringclasses 32
values | scale stringlengths 9 350 | tasks stringlengths 7 132 | modalities stringlengths 7 118 ⌀ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
AgenticFocus | model | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.08857 | https://arxiv.org/abs/2607.08857 | null | null | Mixed-reality synthesis pipeline that restores occluded object geometry and full-hand motion from ordinary first-person human video, then retargets and composites the interaction into synchronized humanoid visual observations, actions, and states | human-demonstrations; robot-learning; cross-embodiment-transfer; retargeting; dexterous-manipulation | egocentric-video; hand-motion; object-meshes; robot-actions |
TactiDex | benchmark | 2026-07 | arXiv | watch | adjacent | 2,026 | https://tactidex.github.io/ | https://arxiv.org/abs/2607.09190 | null | null | Real-world human-demonstration benchmark synchronizing whole-hand tactile pressure, multi-granularity hand kinematics, object 6D states, language descriptions, and task phases, with TactiSkill transfer across single-hand and bimanual tasks | dexterous-manipulation; human-demonstrations; cross-embodiment-transfer; contact-understanding; bimanual-manipulation | tactile; hand-motion; object-pose; language-instructions |
DemoBridge | toolkit | 2026-07 | RSS 2026 RoboData Workshop | open | adjacent | 2,026 | https://gitlab.kuleuven.be/u0123974/demo-bridge/ | https://arxiv.org/abs/2607.09519 | https://gitlab.kuleuven.be/u0123974/demo-bridge/ | Apache-2.0 | Simulation-in-the-loop toolkit converting a single-view RGB stereo human-hand demonstration into a collision-aware, physics-validated robot-arm trajectory; evaluated on three real-demonstration tasks and a synthetic planner benchmark | human-demonstrations; retargeting; robot-learning; simulation; trajectory-synthesis | video; stereo-video; hand-motion; robot-actions; simulation |
PhysV2A | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.09365 | https://arxiv.org/abs/2607.09365 | null | null | Reachability-gated feasibility-completion framework converting video-derived 6D object motion into robot-executable trajectories through grasp-conditioned hypotheses, kinematic checks, and VLM-assisted semantic-mask relaxation; evaluated on four real tabletop tasks | human-demonstrations; robot-learning; cross-embodiment-transfer; manipulation; trajectory-synthesis | rgbd; object-pose; robot-actions; semantic-masks |
Hub Egocentric GoPro RGB+IMU | dataset | 2026-07 | Hugging Face | partial | null | 2,026 | https://huggingface.co/datasets/Hubdata/egocentric-gopro-rgb-imu | null | null | other | 19 GoPro HERO13 egocentric human-manipulation clips totaling 3.72 hours, with H.264 RGB, approximately 200 Hz GPMF IMU, CSV/Parquet/JSON sidecars, and Foxglove MCAP recordings | human-demonstrations; robot-learning; sensor-fusion; hand-object-interaction; imitation-learning | egocentric-video; imu; camera-intrinsics; sensor-sidecars |
Hub Egocentric iPhone RGB+IMU | dataset | 2026-07 | Hugging Face | partial | null | 2,026 | https://huggingface.co/datasets/Hubdata/egocentric-iphone-rgb-imu | null | null | other | 12 iPhone 13/15/15 Pro Max egocentric human-manipulation clips totaling 2.96 hours, with 1080p RGB, nominal 30 Hz CoreMotion and ARKit IMU/attitude, camera intrinsics, sidecars, and MCAP recordings | human-demonstrations; robot-learning; sensor-fusion; hand-object-interaction; imitation-learning | egocentric-video; imu; camera-attitude; camera-intrinsics; sensor-sidecars |
Hub Egocentric Stereo RGB-D | dataset | 2026-07 | Hugging Face | partial | null | 2,026 | https://huggingface.co/datasets/Hubdata/egocentric-stereo-rgbd | null | null | other | 9 ZED X Mini egocentric human-manipulation clips totaling 12.1 minutes, with stereo RGB, dense metric depth, IMU, 6DoF VIO pose, calibration, preview exports, MCAP, and per-clip LeRobot v3.0 packages | human-demonstrations; robot-learning; sensor-fusion; 3d-4d-understanding; imitation-learning | egocentric-video; stereo-video; depth; imu; camera-pose; camera-calibration |
Hub Egocentric iPhone RGB | dataset | 2026-07 | Hugging Face | partial | null | 2,026 | https://huggingface.co/datasets/Hubdata/egocentric-iphone-rgb | null | null | other | 40 iPhone egocentric household-manipulation clips totaling 1.60 hours, delivered as 1080p H.264 RGB video without IMU, depth, pose, or robot action streams | human-demonstrations; robot-learning; hand-object-interaction; action-recognition; imitation-learning | egocentric-video |
OpenGlass Visual Assistance | toolkit | 2026-07 | ACL 2026 System Demonstrations | partial | null | 2,026 | https://github.com/OpenSQZ/OpenGlass | https://arxiv.org/abs/2607.03213 | https://github.com/OpenSQZ/OpenGlass | null | Local-first visual-assistance reference system splitting ESP32 glasses sensing from nearby-device MLLM inference; reports 993 ms median user-to-audio latency for resized images and releases source, hardware instructions, prompts, evaluation data, and logs | visual-assistance; accessibility; edge-ai; on-device-perception; wearable-interaction | egocentric-video; smart-glasses; audio; speech; evaluation-logs |
Worldscape-MoE | model | 2026-07 | arXiv | watch | null | 2,026 | https://worldscape-moe.com/ | https://arxiv.org/abs/2607.03964 | https://github.com/EmbodiedCity/Worldscape-MoE.code | null | Mixture-of-experts diffusion world model unifying camera trajectories, robot actions, and hand-joint action maps across locomotion, manipulation, and egocentric hand-control experiments | world-modeling; world-action-models; video-generation; hand-object-interaction; robot-learning | egocentric-video; generated-video; hand-motion; robot-actions; camera-trajectories |
Trustworthy Visual Predicates | benchmark | 2026-06 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2606.08121 | https://arxiv.org/abs/2606.08121 | null | null | Predicate-level reliability framework evaluating contact, support, containment, motion coupling, grasp, release, and active-hand evidence under blur, occlusion, illumination change, low resolution, frame dropping, and detection noise | hand-object-interaction; manipulation-understanding; robustness; evaluation; neuro-symbolic-reasoning | egocentric-video; bimanual-video; hand-object-contact; predicate-graphs |
Immersive Social Interaction with VR and LLM-Assisted Humanoids | toolkit | 2024-11 | Humanoids 2024 Workshop | watch | null | 2,024 | https://arxiv.org/abs/2607.07430 | https://arxiv.org/abs/2607.07430 | null | null | Apple Vision Pro teleoperation framework coupling egocentric robot visual feedback, voice-controlled locomotion, wrist/finger retargeting, and recording of RGB, commands, joint states, hand motion, and eye gaze for a Unitree H1 | digital-teleoperation; humanoid-robotics; dexterous-manipulation; human-robot-interaction; multimodal-data-collection | egocentric-video; vr; gaze; speech; hand-motion; robot-actions |
ACE-Brain-0.5 | model | 2026-07 | arXiv | partial | adjacent | 2,026 | https://huggingface.co/ACE-Brain/ACE-Brain-0.5-8B | https://arxiv.org/abs/2607.04426 | https://github.com/DAXIAORobotics/ACE-Brain-0.5 | null | Unified 8B-class Qwen3-VL-derived embodied backbone spanning spatial perception, planning, navigation/manipulation action generation, progress monitoring, and self-improvement, evaluated across 15 benchmarks | embodied-ai; vla; spatial-reasoning; navigation; manipulation; progress-estimation | robot-camera-video; images; language-instructions; robot-actions; spatial-memory |
HeadRoom | model | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.08083 | https://arxiv.org/abs/2607.08083 | null | null | Edge-deployable smart-glasses pipeline that estimates visual and auditory channel availability from egocentric video and audio; a controlled 25-participant study reports faster responses when notifications are routed to the less-loaded channel | wearable-interaction; wearable-assistants; cognitive-state-aware-assistance; edge-ai; on-device-perception | egocentric-video; audio; smart-glasses |
VEGAS | benchmark | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.08489 | https://arxiv.org/abs/2607.08489 | null | null | Training-free Video caption Evaluation via GAze Score metric plus an evaluation dataset of egocentric activities and instructional slides paired with synchronized gaze and reference captions; the abstract does not report the dataset size | captioning; gaze; attention-modeling; video-text-retrieval; evaluation | egocentric-video; gaze; captions; annotations |
CI-HOI / DEHOI | benchmark | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.08514 | https://arxiv.org/abs/2607.08514 | null | null | Cue-Isolated HOI evaluation and DEHOI testbed separating hand- and object-related observations through inpainting to diagnose whether egocentric video-language models use manipulation and object-transformation cues rather than contextual shortcuts | hand-object-interaction; action-recognition; object-state-change; evaluation; robot-manipulation | egocentric-video; hand; objects; action-labels; synthetic-video |
Whareformer | model | 2026-07 | ECCV 2026 | open | null | 2,026 | https://jacobchalk.github.io/Whareformer/ | https://arxiv.org/abs/2607.08537 | https://github.com/JacobChalk/Whareformer | MIT | 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 | long-term-object-tracking; object-reidentification; memory; 3d-understanding; redetection | egocentric-video; object-tracks; 3d-scene; spatial-memory |
EgoWAM | model | 2026-07 | arXiv | watch | null | 2,026 | https://gatech-rl2.github.io/egowam.github.io/ | https://arxiv.org/abs/2607.08436 | null | null | Controlled human-robot co-training study comparing pixel, DINO, and 3D-flow world targets on three real-world bimanual tasks; DINO reports up to 4x gains in out-of-distribution object/scene generalization and 3D flow reports 20-30% in-domain gains | world-action-models; robot-learning; bimanual-manipulation; cross-embodiment-transfer; compositional-generalization | egocentric-video; robot-camera-video; robot-actions; 3d-flow; dense-features |
Datoric Industrial Egocentric Video | dataset | 2026-07 | Hugging Face | request | null | 2,026 | https://huggingface.co/datasets/Datoric/industrial-egocentric-video-50000h | null | null | other | Commercial card claims 50,000 hours of workplace first-person video spanning picking, packing, scanning, stocking, inventory, sorting, inspection, returns, assembly, and cleaning, with public specification and sample Parquet metadata | industrial-egocentric-video; action-recognition; hand-object-interaction; robot-learning; human-demonstrations | egocentric-video; annotations; action-labels |
Datoric Residential Egocentric Video | dataset | 2026-07 | Hugging Face | request | null | 2,026 | https://huggingface.co/datasets/Datoric/egocentric-residential-video-100000h | null | null | other | Commercial card claims 100,000 hours of 1080p+ household first-person video recorded in Brazilian homes, with task, action, object, hand-object interaction, and completion-state annotations plus public specification and sample Parquet metadata | real-world-routines; action-recognition; hand-object-interaction; robot-learning; imitation-learning | egocentric-video; annotations; action-labels; state-labels |
BiNoMaP | model | 2025-09 | arXiv | watch | null | 2,025 | https://hnuzhy.github.io/projects/BiNoMaP/ | https://arxiv.org/abs/2509.21256 | null | null | Category-level bimanual non-prehensile manipulation primitives learned from hand trajectories extracted from egocentric video demonstrations, geometrically refined and transferred across two dual-arm robot platforms | bimanual-manipulation; retargeting; robot-learning; human-demonstrations; cross-embodiment-transfer | egocentric-video; hand-motion; robot-actions; articulated-objects |
GIRAF | model | 2026-07 | CVPR 2026 HuMoGen Workshop | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.07880 | https://arxiv.org/abs/2607.07880 | null | null | Text-conditioned diffusion model for coordinated full-body interaction with articulated objects, using object-centric contact representations, mixed locomotion/interaction training, and contact-based augmentation | human-motion; motion-generation; articulated-objects; hand-object-interaction; simulation | human-motion; articulated-objects; hand-object-contact; motion-descriptions |
Video-Action Generalization Gap / Temporal Ratio | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://umishra.me/temporal-ratio/ | https://arxiv.org/abs/2607.08127 | null | null | Systematic video-action model study introducing Temporal Ratio to measure reliance on future latent rollouts, plus inference-time adaptive guidance that narrows compositional OOD gaps on LIBERO and real-world robot tasks | world-action-models; vla; compositional-generalization; action-evaluation; robot-learning | generated-video; latent-video; robot-actions; robot-camera-video |
LEEVLA | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.08182 | https://arxiv.org/abs/2607.08182 | null | null | VLA architecture combining drift-guided dynamic prioritization of task-relevant visual regions with structured latent feature-flow generation for manipulation benchmarks | vla; robot-learning; manipulation; spatial-reasoning | robot-camera-video; robot-actions; latent-actions |
LingBot-VA 2.0 | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.08639 | https://arxiv.org/abs/2607.08639 | null | null | Native video-action foundation model with a semantic visual-action tokenizer, causal pretraining, sparse mixture-of-experts backbone, and asynchronous closed-loop inference for few-shot robot manipulation | world-action-models; vla; robot-learning; manipulation; efficient-inference | robot-camera-video; robot-actions; latent-video; generated-video |
TFP | model | 2026-07 | RSS 2026 SemRob Workshop | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.08283 | https://arxiv.org/abs/2607.08283 | null | null | Temporally Conditioned Memory-Fusion Policy with a 3.3B VLA backbone and event-sensitive episode belief, reporting 98.75% on LIBERO, 93.77% on LIBERO-plus, and up to 75% on MIKASA ShellGameTouch | vla; long-horizon-memory; robot-learning; manipulation | robot-camera-video; robot-actions; memory |
AnyDexRT | toolkit | 2026-07 | arXiv | watch | adjacent | 2,026 | https://chenxi-wang.github.io/projects/anydexrt/ | https://arxiv.org/abs/2607.08341 | null | null | Calibration-free dexterous-hand retargeting that combines self-supervised fingertip correspondence, few-shot human guidance, and contact-aware pinch refinement across multiple hands and real teleoperation tasks | retargeting; digital-teleoperation; dexterous-manipulation; human-demonstrations | hand-pose; robot-actions; hand-object-contact |
Harness VLA | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.08448 | https://arxiv.org/abs/2607.08448 | null | null | Memory-guided agentic harness that composes a frozen VLA as a retryable contact-rich primitive with fixed analytic primitives, reporting gains on LIBERO-Pro, RoboCasa365, and RoboTwin C2R | vla; long-horizon-memory; manipulation; action-planning; robot-learning | robot-camera-video; robot-actions; memory |
FabriVLA | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.08575 | https://arxiv.org/abs/2607.08575 | null | null | Compact 1B-parameter VLA combining InternVL3.5 with a flow-matching action head, reporting a 90.0% tier-average success rate across Meta-World MT50 | vla; manipulation; robot-learning; efficient-inference | robot-camera-video; robot-actions; language-instructions |
DexVerse | benchmark | 2026-07 | arXiv | partial | adjacent | 2,026 | https://ycyao216.github.io/DexVerse.site/ | https://arxiv.org/abs/2607.08751 | https://github.com/ycyao216/DexVerse | BSD-3-Clause | Modular dexterous-manipulation benchmark with 100 tasks, three robot arms, six dexterous hands, configurable visual variation, a VR teleoperation interface, and 3,180 reported demonstrations | benchmark; dexterous-manipulation; bimanual-manipulation; vla; cross-embodiment-transfer | robot-camera-video; depth; point-clouds; robot-actions; demonstrations; simulation |
LingBot-Video | model | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.07675 | https://arxiv.org/abs/2607.07675 | null | null | Mixture-of-Experts DiT video pretraining paradigm for embodied intelligence, using a data-profiling engine that augments internet videos with robot-oriented manipulation, navigation, and egocentric-perspective footage plus physically grounded reward alignment | foundation-video; world-modeling; embodied-ai; robot-learning; video-generation; efficient-inference | video; generated-video; egocentric-video; robot-camera-video; robot-actions |
Ego3DLM / Ego-Human Motion Prediction | model | 2026-07 | ECCV 2026 | watch | null | 2,026 | https://arxiv.org/abs/2607.07001 | https://arxiv.org/abs/2607.07001 | null | null | 3D-aware LLM for egocentric human motion prediction that jointly decodes past pose, future pose, past narration, and future narration from three-point tracking, 3D scene features, and egocentric video; evaluated on Nymeria | human-motion; pose; trajectory-prediction; motion-language; future-prediction; ar-vr | egocentric-video; 3d-scene; 3d-human-motion; language; trajectories |
CoMind | dataset | 2026-07 | ECCV 2026 | watch | null | 2,026 | https://comind.ethz.ch/ | https://arxiv.org/abs/2607.06691 | null | null | Dual-view egocentric and exocentric collaborative-cooking dataset with two synchronized head-mounted cameras, two exocentric views, high-quality audio, gaze tracking, 3D scene scans, object scans, action/hand/object/gaze annotations, and benchmarks for joint attention, socially conditioned object-interaction anticipati... | social-reasoning; theory-of-mind; visual-grounding; action-anticipation; interaction-understanding | egocentric-video; exocentric-video; audio; gaze; scene-scans; object-meshes; 6dof; annotations |
LingBot-World 2.0 | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.07534 | https://arxiv.org/abs/2607.07534 | null | null | Interactive world simulator update with unbounded interaction horizon, a real-time distilled variant for 720p 60 FPS streams, richer action/event controls, agentic pilot/director harnesses, and paired 14B/1.3B models | world-modeling; simulation; video-generation; embodied-ai | generated-video; language-instructions; simulation |
RoboDojo | benchmark | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.04434 | https://arxiv.org/abs/2607.04434 | null | null | Unified sim-and-real benchmark for generalist robot manipulation with 42 simulation tasks, 18 real-world tasks, five capability dimensions, RoboDojo-RealEval remote evaluation, XPolicyLab integration, 30 evaluated policies, and a public leaderboard | benchmark; evaluation; robot-manipulation; vla; domain-generalization | robot-camera-video; robot-actions; simulation; evaluation |
LaMem-VLA | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.07608 | https://arxiv.org/abs/2607.07608 | null | null | Latent-memory-native VLA framework with curator, seeker, condenser, and weaver modules that reconstruct short- and long-term historical experience into latent memory tokens for long-horizon robotic manipulation | vla; robot-learning; manipulation; long-horizon-memory; memory-reasoning | robot-camera-video; robot-actions; latent-actions; memory |
TouchWorld | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.07287 | https://arxiv.org/abs/2607.07287 | null | null | Predictive-and-reactive tactile foundation model for dexterous manipulation that separates vision-language planning, tactile world-model prediction, visuo-tactile goal-conditioned action generation, and high-frequency tactile residual refinement | tactile-learning; dexterous-manipulation; world-modeling; robot-learning; contact-rich-manipulation | robot-camera-video; tactile; proprioception; robot-actions |
WAM-TTT | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.06988 | https://arxiv.org/abs/2607.06988 | null | null | 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 | world-action-models; test-time-scaling; robot-learning; human-demonstrations; manipulation | demonstration-video; robot-camera-video; robot-actions; memory |
EAGOR | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.06165 | https://arxiv.org/abs/2607.06165 | null | null | Training-free geometry-aware framework for 360-degree directional reasoning that maintains spherical-harmonic belief fields over target directions and reports gains on HOS, OSR-Bench, and legged-robot navigation | spatial-reasoning; navigation; embodied-reasoning; visual-grounding | 360-video; robot-observations; maps |
VLA Models Review | survey | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.06706 | https://arxiv.org/abs/2607.06706 | null | null | Survey of 183 VLA contributions from 2017-2026 across architectures, training recipes, action representations, bimanual coordination, UAV navigation/control, language grounding, memory, and world models | survey; vla; robot-learning; bimanual-manipulation; navigation | survey; robot-camera-video; robot-actions |
NativeMEM | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.06678 | https://arxiv.org/abs/2607.06678 | null | null | Memory-compression VLA policy that repurposes the VLA vision encoder to compress each historical frame from each camera view into a single memory token, improving long-horizon manipulation with low latency | vla; robot-learning; manipulation; long-horizon-memory; efficient-inference | robot-camera-video; robot-actions; memory |
Lift3D-VLA | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.06564 | https://arxiv.org/abs/2607.06564 | null | null | 3D geometry- and dynamics-aware VLA that lifts pretrained 2D models to point-cloud reasoning, uses geometry-centric masked autoencoding, and reports results across 22 simulated and 8 real-world manipulation tasks | vla; robot-learning; 3d-understanding; manipulation; spatial-reasoning | robot-camera-video; point-clouds; robot-actions |
Pelican-VLA 0.5 | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.06655 | https://arxiv.org/abs/2607.06655 | null | null | Unified VLA report integrating vision-language understanding, future-frame generation, and action prediction, with learnable reasoning slots that induce manipulation-centric attention across unseen scenes and embodiments | vla; robot-learning; manipulation; future-prediction; video-generation | robot-camera-video; generated-video; robot-actions; language-instructions |
ActionCache | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.06370 | https://arxiv.org/abs/2607.06370 | null | null | Training-free cache and refinement method for accelerating flow-matching VLA action heads, reusing past intermediate actions through compact multimodal keys and reporting up to 11.75x and 34.43x inference acceleration on representative VLA models | vla; robot-learning; efficient-inference; action-evaluation; manipulation | robot-camera-video; robot-actions; latent-actions |
Smooth Operator / SBR | toolkit | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.07491 | https://arxiv.org/abs/2607.07491 | null | null | Sampling-Based Retargeter for low-jitter real-time kinematic hand retargeting, evaluated in simulation and an 18-participant real-world user study over three complex manipulation tasks | retargeting; digital-teleoperation; human-demonstrations; dexterous-manipulation; robot-learning | hand-pose; trajectories; robot-actions |
EgoPolice | benchmark | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.06468 | https://arxiv.org/abs/2607.06468 | null | null | Benchmark of real egocentric police-civilian interactions from public body-worn camera footage, with second-by-second high-stakes action labels plus classification and multiple-choice QA tasks | benchmark; action-recognition; video-qa; safety; high-stakes-egocentric-understanding | egocentric-video; body-worn-camera; annotations; qa; action-labels |
RynnWorld-4D | model | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.06559 | https://arxiv.org/abs/2607.06559 | null | null | 4D embodied world model for robotic manipulation that co-generates future RGB, depth, and optical flow; reports Rynn4DDataset 1.0 with 254.4M+ frames across egocentric human and robotic manipulation videos | world-modeling; robot-learning; manipulation; vla; action-conditioned-generation | egocentric-video; robot-camera-video; rgb-d; optical-flow; generated-video; robot-actions |
RynnWorld-Teleop | model | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.06558 | https://arxiv.org/abs/2607.06558 | null | null | Digital-teleoperation system where an operator hand-pose stream drives a robot-centric generative world model to synthesize high-fidelity egocentric videos and action labels, with reported 40+ FPS single-pass generation | digital-teleoperation; robot-learning; imitation-learning; world-modeling; data-generation | hand-pose; generated-video; egocentric-video; robot-actions |
LingBot-VLA 2.0 | model | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.06403 | https://arxiv.org/abs/2607.06403 | null | null | VLA system update with a redesigned data pipeline and about 60K hours of pretraining data, including 50K hours of robot trajectories across 20 robot configurations and 10K hours of egocentric human videos | vla; robot-learning; manipulation; whole-body-control; world-modeling | egocentric-video; robot-camera-video; robot-actions; language-instructions; depth |
CGGS | model | 2026-07 | IEEE TIP 2026 | open | null | 2,026 | https://github.com/CGGS-26/CGGS | https://arxiv.org/abs/2607.03819 | https://github.com/CGGS-26/CGGS | mit | Text-to-3D framework for ego-centric 3D scene generation using a consistency-augmented ego-centric generator, depth/layout decoration from optical flow and point tracks, and geometric Gaussian refinement | text-to-3d; 3d-scene-generation; world-modeling; view-consistency; spatial-reasoning | text; generated-3d; generated-video; depth; optical-flow |
EgoInertia-MI | benchmark | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.03934 | https://arxiv.org/abs/2607.03934 | null | null | Multimodal benchmark dataset with synchronized egocentric video and wearable IMU for motor-impairment analysis over 19 upper- and lower-body activities and simulated no/mild/severe impairment levels | motor-impairment-assessment; activity-recognition; sensor-fusion; health-ai; wearable-sensing | egocentric-video; imu; wearable-sensing; action-labels |
SAGE Action-Gaze | model | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.04017 | https://arxiv.org/abs/2607.04017 | null | null | Synchronized Action-GazE framework for joint recognition and anticipation of human-object interaction and gaze patterns across egocentric close-up and exocentric contextual settings | gaze-prediction; action-anticipation; hand-object-interaction; human-behavior-understanding | egocentric-video; exocentric-video; gaze; action-labels |
R3D-Bench / R3D | benchmark | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.02921 | https://arxiv.org/abs/2607.02921 | null | null | Quantitative 3D spatial-reasoning benchmark for egocentric wearables with 3,033 questions over 57 Aria Digital Twin RGB-D video sequences, plus a model-agnostic 3D spatial tool-calling baseline | spatial-reasoning; 3d-understanding; video-qa; ar-wearables; benchmark | egocentric-video; rgb-d; 3d; qa; object-segmentation |
S-EMBER | benchmark | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.02689 | https://arxiv.org/abs/2607.02689 | null | null | Streaming Egocentric Memory Benchmark with 3,141 Ray-Ban Meta smart-glasses videos totaling 388 hours and 9,448 QA pairs requiring grounded temporal evidence under causal streaming retrieval | episodic-memory; streaming-video-understanding; retrieval; video-qa; wearable-assistants | egocentric-video; smart-glasses; qa; temporal-localization; streaming-video |
Humanola Egocentric Hand-Pose Sample | dataset | 2026-07 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/humanola-inc/egocentric-sample | null | null | cc-by-4.0 | LeRobot v2.1 sample delivery with 9 head-mounted manipulation episodes, 48,272 frames at 30 fps, about 27 minutes, 188-D state vectors, four synchronized 1920x1200 camera streams, 200 Hz head/wrist IMU, 3D hand keypoints, wrist/head poses, overlays, and Rerun logs | hand-tracking; robot-learning; manipulation; imitation-learning; sensor-fusion; loader-testing | egocentric-video; wrist-camera; hand-pose; imu; camera-pose; trajectories |
UniDataPro Egocentric Video Preview | dataset | 2026-05 | Hugging Face | partial | null | 2,026 | https://huggingface.co/datasets/UniDataPro/egocentric-video | null | null | cc-by-nd-4.0 | Commercial egocentric dataset preview card claiming a 4,050-hour first-person robotics/physical-AI collection with VR-headset/Zed-camera capture, IMU/quaternion tracking, 3D pose signals, and household hand-object scenarios; public repository exposes sample video, CSV, and tracking text only | robot-learning; action-recognition; hand-object-interaction; pose-estimation; data-collection | egocentric-video; imu; pose; tracking; rgb-d |
Voxel51 Egocentric-10K Subset | dataset | 2025-11 | Hugging Face | open | null | 2,025 | https://huggingface.co/datasets/Voxel51/Egocentric_10K_subset | null | null | apache-2.0 | FiftyOne-packaged open subset of Egocentric-10K focused on Factory 51, with the card reporting 416 samples / first 51 videos per worker, 1080p H.265 head-mounted monocular clips, worker/factory metadata, and app-loading instructions | action-recognition; hand-object-interaction; industrial-egocentric-video; prototyping; data-visualization | egocentric-video; metadata |
Embodied.cpp | toolkit | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.02501 | https://arxiv.org/abs/2607.02501 | null | null | Portable C++ inference runtime for embodied VLA and WAM deployment, with layered adapters, sequence builders, backbone execution, head plugins, and robot-interface deployment support | vla; world-action-models; inference-runtime; robot-deployment; edge-ai | robot-camera-video; robot-actions; embodied-models |
EVA-Client | toolkit | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.02646 | https://arxiv.org/abs/2607.02646 | null | null | Unified real-robot client for embodied policy deployment, data collection, and evaluation, with decoupled robot backends, inference strategies, transport middleware, and training-ready rollout logging | robot-deployment; data-collection; evaluation; vla; manipulation | robot-camera-video; robot-actions; logs |
InternVLA-A1.5 | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.04988 | https://arxiv.org/abs/2607.04988 | null | null | Robot manipulation VLA combining a native VLM backbone, continuous action expert, subtask/VQA supervision, and latent foresight tokens distilled from a frozen video-generation model | vla; robot-learning; manipulation; future-prediction; compositional-generalization | robot-camera-video; robot-actions; language-instructions; latent-video |
SIEVE | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.06442 | https://arxiv.org/abs/2607.06442 | null | null | Structure-aware data-selection method for VLA imitation learning that discovers visuo-motor primitives, allocates selection budgets to composition patterns, and selects medoid trajectories | vla; imitation-learning; data-selection; robot-learning | robot-camera-video; robot-actions; trajectories |
SVA / Look Before You Leap | model | 2026-07 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.03751 | https://arxiv.org/abs/2607.03751 | null | null | Search, Value, and Act framework that distills Monte-Carlo tree-search rollouts into a lightweight action evaluator for frozen VLA policies, improving test-time action selection without changing the backbone | vla; robot-learning; action-evaluation; test-time-scaling; planning | robot-camera-video; robot-actions; simulation |
USA Egocentric | dataset | 2026-07 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/Grably/usa-egocentric | null | null | other | Three iPhone kitchen-manipulation episodes with normalized RGB videos, raw videos, camera calibration and trajectory files, MediaPipe and HaMeR hand outputs, action segments, contact/object tracks on the strongest run, review manifests, and quality reports | robot-learning; imitation-learning; hand-object-interaction; manipulation; data-conversion | egocentric-video; hand-pose; object-tracks; camera-trajectory; annotations; calibration |
TimeScribe Egocentric Annotations | dataset | 2026-06 | Hugging Face | request | null | 2,026 | https://huggingface.co/datasets/play0718/timescribe-egocentric-annotations | null | null | cc-by-nc-4.0 | Manual human dense-caption annotations for eight egocentric videos, with frame-aligned micro-events, Chinese captions, English mirrors where available, per-video JSON, manifest, and source clips | captioning; hand-action-recognition; procedure-understanding; annotation-tooling | egocentric-video; captions; annotations; text |
Lo6yu Egocentric RGB-D + EMG/IMU | dataset | 2026-05 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/Lo6yu/egocentric_dataset | null | null | cc-by-nc-4.0 | Eight first-person household-activity packages, about 30.1 GB total, with RGB-D video, wrist EMG/IMU, hand keypoints, object masks, hand-object contact, per-finger force annotations, semantic action segments, and training-ready exports | robot-learning; imitation-learning; manipulation; hand-object-interaction; action-segmentation; sensor-fusion; contact-understanding | egocentric-video; rgb-d; emg; imu; hand-pose; object-masks; contact; force; annotations |
RoboX Egocentric Collection | dataset | 2026-04 | Hugging Face | request | null | 2,026 | https://huggingface.co/datasets/RoboXTechnologies/RoboX-Egocentric-Collection-v0.2 | null | null | cc-by-nc-4.0 | Auto-gated crowdsourced first-person robotics collection with 7,342 annotated clips across grasping, daily activities, scene captures, and navigation; includes videos, metadata, hand keypoints, object tracks, action segments, IMU/camera-pose signals, and campaign reports | robot-learning; imitation-learning; manipulation; navigation; scene-understanding; action-segmentation; hand-object-interaction | egocentric-video; text; annotations; hand-pose; object-tracks; imu; camera-pose |
WhissleAI Egocentric Activity Sample | dataset | 2026-04 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/WhissleAI/egocentric-activity-sample | null | null | cc-by-4.0 | Small Ego4D-style prototyping sample with 19 first-person 540p clips, about 9.5 minutes total, 99 narrations, 57 NLQ queries, 19 moment annotations, 57 FHO action records, and bundled taxonomy/metadata | action-recognition; natural-language-query; moment-query; captioning; video-qa; annotation-tooling | egocentric-video; narrations; annotations; text |
OBayData Egocentric Dexterous Manipulation Demo | dataset | 2026-03 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/obaydata/egocentric-dexterous-manipulation-demo | null | null | cc-by-nc-4.0 | Public demo with 500 train and 100 test manipulation sessions, synchronized head-mounted and dual wrist videos, 3D hand joints, camera extrinsics, action labels, language annotations, and metadata for dexterous manipulation quality review | robot-learning; dexterous-manipulation; imitation-learning; hand-object-interaction; action-segmentation; language-grounding | egocentric-video; wrist-camera; hand-pose; camera-pose; action-labels; language |
Kriya-Egocentric-100K | dataset | 2026-03 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/ankk98/kriya-egocentric-100k | null | null | mit | Preview annotation subset for Egocentric-100K with Action100M-style hierarchical temporal action trees, one JSON file per video clip, video metadata, segment nodes, and LLM-generated action descriptions | action-segmentation; procedure-understanding; captioning; annotation-tooling; human-demonstrations | egocentric-video; annotations; text |
HoyerChou Egocentric Bimanual Videos | dataset | 2026-01 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/HoyerChou/EgocentricVideos | https://arxiv.org/abs/2501.14208 | null | apache-2.0 | YOTO/YOTO++ bimanual manipulation release with three raw first-person task videos, 119 segmented short demonstrations, 241 rows, about 973 MB total, and extracted MANO hand meshes, 3D keypoints, camera calibration, and camera/world transforms | bimanual-manipulation; robot-learning; imitation-learning; hand-object-interaction; hand-tracking | egocentric-video; hand-meshes; hand-pose; camera-pose; annotations |
Egocentric Specialties | dataset | 2026-01 | Hugging Face | request | null | 2,026 | https://huggingface.co/datasets/ThisIsPepper/EgocentricSpecialties | null | null | apache-2.0 | Auto-gated skilled-jewelry manufacturing dataset card describing synchronized egocentric and top-down videos for diamond setting and specialty manufacturing workflows | skilled-activity; industrial-assembly; robot-learning; imitation-learning; hand-object-interaction | egocentric-video; exocentric-video; text; annotations |
Egocentric Adjust Bottle LeRobot | dataset | 2026-05 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/suz22/egocentric_adjust_bottle | null | null | apache-2.0 | LeRobot-format human_mano adjust-bottle task dataset with 500 episodes, 36,897 frames at 15 fps, image, text, and timeseries fields, chunked Parquet files, and video exports | robot-learning; manipulation; imitation-learning; data-conversion; sample-data | egocentric-video; egocentric-images; text; trajectories; robot-actions |
SIGNIQ Egocentric Hand-Pose Sample | dataset | 2026-06 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/jackji/signiq-egocentric-handpose-sample | null | null | cc-by-4.0 | Five short head-mounted GoPro manipulation clips at 1280x720 and 30 fps, packaged with Label Studio import JSON and metadata for hand-pose and action-segment annotation workflows | hand-tracking; manipulation; imitation-learning; annotation-tooling; action-segmentation | egocentric-video; annotations; text |
Grid Egocentric Residential Samples | dataset | 2026-05 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/aryan56789gridai/grid-egocentric-residential-samples | null | null | cc-by-4.0 | Public preview pack of first-person residential manipulation recordings, including full-length household samples at 1920x1080, 30 fps, H.265/HEVC video | robot-learning; imitation-learning; manipulation; human-demonstrations | egocentric-video |
Industrial Workplace Egocentric FHD Samples | dataset | 2026-05 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/TrainThemAI/Industrial-Workplace-Egocentric-FHD-Samples | null | null | mit | 21 rights-cleared first-person industrial and workplace clips captured with a head-mounted smartphone, with paired JSON metadata across factory, construction, services, retail, transit, and food-service domains | human-demonstrations; vla; robot-learning; world-modeling; data-collection | first-person-video; text; annotations |
Egocentric-10K | dataset | 2025-11 | Hugging Face | request | null | 2,025 | https://huggingface.co/datasets/builddotai/Egocentric-10K | null | null | apache-2.0 | Factory-only egocentric video corpus reporting 10,000 hours and 1.08B frames, WebDataset-style tar shards, intrinsics files, and companion hand-visibility / active-manipulation evaluation data | human-demonstrations; action-recognition; hand-object-interaction; robot-learning; data-collection | egocentric-video; text; calibration |
Egocentric-10K Evaluation | benchmark | 2025-11 | Hugging Face | open | null | 2,025 | https://huggingface.co/datasets/builddotai/Egocentric-10K-Evaluation | null | null | apache-2.0 | 30K-frame public comparison set with 10K sampled frames each from Egocentric-10K, Ego4D, and EPIC-KITCHENS-100, labeled for hand visibility and active manipulation with published prompts | benchmark; evaluation; hand-object-interaction; action-understanding | egocentric-video; egocentric-images; annotations; text |
Future-Privileged Causal Ego Gaze | model | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.01437 | https://arxiv.org/abs/2607.01437 | null | null | Controlled causal egocentric gaze-estimation study using a future-aware training branch that is discarded at inference; evaluates EGTEA Gaze+ and Ego4D and finds peak gains around 1.7-3.3 seconds of future context | gaze-prediction; streaming-video-understanding; ar-wearables; attention-modeling | egocentric-video; gaze; streaming-video |
NormAct | benchmark | 2026-06 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2606.27826 | https://arxiv.org/abs/2606.27826 | null | null | Embodied planning benchmark for hidden social-norm compliance in egocentric environments, reporting Goal Achievement, Norm Compliance, and Task Success and showing a large norm-grounding gap in frontier MLLMs | normative-action-reasoning; action-planning; embodied-reasoning; safety; benchmark | egocentric-observations; language-instructions; annotations |
Astra | model | 2026-06 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2606.06476 | https://arxiv.org/abs/2606.06476 | null | null | Agentic spatial-reasoning framework coupling an RL-trained VLM policy with a view-consistent world simulator to request imagined novel-view evidence from limited egocentric observations | spatial-reasoning; agentic-reasoning; world-modeling; view-synthesis; multimodal-reasoning | egocentric-observations; generated-video; text; camera-motion |
Egocentric-100K | dataset | 2025-12 | Hugging Face | request | null | 2,025 | https://huggingface.co/datasets/builddotai/Egocentric-100K | null | null | apache-2.0 | Manual-labor egocentric video corpus reporting 100,405 hours, 10.8B frames, 2,010,759 clips, WebDataset packaging, intrinsics, and 238 factory directories | human-demonstrations; action-recognition; hand-object-interaction; robot-learning; data-collection | egocentric-video; text; calibration |
Egocentric-100K Evaluation | benchmark | 2025-12 | Hugging Face | open | null | 2,025 | https://huggingface.co/datasets/builddotai/Egocentric-100K-Evaluation | null | null | apache-2.0 | 30K-frame public evaluation set comparing Egocentric-100K, Ego4D, and EPIC-KITCHENS-100 on hand visibility and active manipulation density | benchmark; evaluation; hand-object-interaction; action-understanding | egocentric-video; egocentric-images; annotations; text |
Manus Egocentric Sample | dataset | 2026-01 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/OpenGraphLabs-Research/manus-egocentric-sample | null | null | apache-2.0 | 20 LeRobot-format egocentric manipulation episodes with Manus glove finger tracking, OAK-D RGB/depth, IMU, per-episode parquet files, depth binaries, and egocentric videos | robot-learning; imitation-learning; hand-tracking; manipulation; sensor-fusion | egocentric-video; depth; hand-mocap; imu; robot-actions |
Motion Focus Recognition | model | 2026-01 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2601.07154 | https://arxiv.org/abs/2601.07154 | null | null | Real-time motion-focus recognition for fast-moving egocentric video, estimating wearer locomotion intention from arbitrary first-person video with camera-pose foundation-model features and sliding-batch inference | activity-recognition; motion-estimation; edge-ai; efficient-inference | egocentric-video; camera-motion; motion |
Egocentric Affordance Dataset | dataset | 2026-02 | Hugging Face | open | null | 2,026 | https://huggingface.co/datasets/Kavin60606/egocentric-affordance-dataset | null | null | cc-by-4.0 | 48,400 annotated egocentric video frames across cleaning, workshop/DIY, factory work, gardening/outdoor, and medical/healthcare activity categories with Qwen3-VL-2B affordance labels | affordance; hand-object-interaction; object-detection; action-recognition; robot-learning | egocentric-images; annotations; text |
CIVIL Lifelog Retrieval | model | 2025-10 | arXiv | watch | null | 2,025 | https://arxiv.org/abs/2510.04010 | https://arxiv.org/abs/2510.04010 | null | null | Captioning-Integrated Visual Lifelog retrieval system that converts wearable-camera lifelog images into first-person experience captions before text-query retrieval | lifelogging; personalized-retrieval; memory-assistance; captioning | lifelogging-images; captions; text |
THOR | model | 2025-07 | arXiv | watch | null | 2,025 | https://arxiv.org/abs/2507.06442 | https://arxiv.org/abs/2507.06442 | null | null | Thermal-guided adaptive RGB sampling for wearable hand-object activity monitoring, validated with 14 in-the-wild participants and Ego4D, using about 3% of RGB data while preserving activity segments | hand-object-interaction; activity-recognition; wearable-sensing; efficient-inference; healthcare-assistance | egocentric-video; thermal; object-crops |
Physical Activity VLM Annotation | model | 2025-05 | arXiv | watch | null | 2,025 | https://arxiv.org/abs/2505.03374 | https://arxiv.org/abs/2505.03374 | null | null | Compares VLMs and discriminative models for labeling sedentary, light, and moderate-to-vigorous activity from Autographer wearable-camera images across Oxfordshire and Sichuan free-living studies | activity-recognition; lifelogging; wearable-sensing; domain-generalization | egocentric-images; action-labels |
Head+Cane Camera Navigation | model | 2025-04 | arXiv | watch | null | 2,025 | https://arxiv.org/abs/2504.19345 | https://arxiv.org/abs/2504.19345 | null | null | Systematic comparison of synchronized head- and cane-mounted cameras for blind last-mile navigation, evaluating SLAM localization and NeRF-style 3D reconstruction across five real environments | blind-assistance; navigation; scene-reconstruction; sensor-fusion | egocentric-video; wearable-sensing; scene-reconstruction |
Audio-Narrated FPV Domain Generalization | model | 2024-09 | arXiv | watch | null | 2,024 | https://arxiv.org/abs/2409.09611 | https://arxiv.org/abs/2409.09611 | null | null | Multimodal first-person action-recognition framework using motion, audio, appearance, audio narrations, and audio-visual consistency ratings to improve domain generalization on ARGO1M | action-recognition; audio-visual; domain-generalization; multimodal-generalization | first-person-video; audio; narrations; motion |
Smart-Glasses Engagement Prediction | dataset | 2024-09 | arXiv | watch | null | 2,024 | https://arxiv.org/abs/2409.09135 | https://arxiv.org/abs/2409.09135 | null | null | Smart-glasses dyadic-conversation dataset from 34 participants with self-reported engagement ratings and a multimodal-transcript LLM fusion strategy for behavioral reasoning | social; engagement-detection; multimodal-conversation; behavior-understanding | smart-glasses; audio; transcripts; annotations |
Egocentric 360 VideoQA for Visual Impairment | dataset | 2024-05 | arXiv | watch | null | 2,024 | https://arxiv.org/abs/2405.19794 | https://arxiv.org/abs/2405.19794 | null | null | Visual-question-answering dataset captured with a 360-degree egocentric wearable camera for multiple real-life obstacles faced by people with visual impairments | assistive-videoqa; 360-video; blind-assistance; video-qa | first-person-360-video; qa; annotations |
AIM-2 Food Ingestion Environment | model | 2024-05 | arXiv | watch | null | 2,024 | https://arxiv.org/abs/2405.07827 | https://arxiv.org/abs/2405.07827 | null | null | Two-stage fine-tuning and transfer-learning method for recognizing ingestion environment from AIM-2 egocentric wearable-camera images in the UA Free Living Study | nutrition-estimation; activity-recognition; lifelogging; wearable-sensing | egocentric-images; nutrition-labels |
LIME | model | 2026-07 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2607.02417 | https://arxiv.org/abs/2607.02417 | null | null | Language-conditioned camera-motion generator mined from egocentric video, pairing free-form intents and observation-gain descriptions with relative SE(3) target poses for next-view prediction | active-perception; spatial-reasoning; robot-learning; vla; navigation | egocentric-video; camera-motion; language-instructions; robot-camera-video |
LongEgoRefer | benchmark | 2026-07 | ECCV 2026 | open | null | 2,026 | https://github.com/shunya-kato/LongEgoRefer | https://arxiv.org/abs/2607.02096 | https://github.com/shunya-kato/LongEgoRefer | null | Long-form Ego4D referring-expression benchmark with 1,498 expressions over videos averaging 45 minutes, requiring temporal and spatial localization of sparse referred objects | referring-expression-comprehension; temporal-grounding; object-grounding; visual-grounding; benchmark | egocentric-video; text; object-boxes; annotations |
HandsOnWorld | model | 2026-07 | arXiv | watch | null | 2,026 | https://shad0wta9.github.io/handsonworld-page/ | https://arxiv.org/abs/2607.02075 | null | null | Unconstrained egocentric video generation with camera-disentangled hand control; introduces EgoVid-Pro with 103K clips and roughly 12M frames of protagonist-only 3D hand trajectories | video-generation; hand-object-interaction; world-modeling; 3d-hand-pose | egocentric-video; hand-trajectories; camera-motion; generated-video |
SG-Ego / GLEN | dataset | 2026-07 | arXiv | watch | null | 2,026 | https://francescapistilli.github.io/GLEN | https://arxiv.org/abs/2607.02425 | null | null | Large-scale Ego4D spatio-temporal scene-graph annotation set plus GLEN graph model for scene-evolution reasoning, retrieval, and activity-driven graph-edit forecasting | structured-reasoning; action-understanding; scene-graph-qa; temporal-reasoning | egocentric-video; scene-graphs; action-labels; text |
Wearable Gait MoCap with Shank-Mounted Ego Cameras | dataset | 2026-07 | Scientific Data 2026 | watch | null | 2,026 | https://doi.org/10.1038/s41597-026-07657-7 | https://doi.org/10.1038/s41597-026-07657-7 | null | null | Wearable motion-capture dataset for gait analysis using IMUs and shank-mounted egocentric cameras | activity-recognition; human-motion; sensor-fusion; wearable-sensing | egocentric-video; imu; motion; wearable-sensing |
TAP VLA | model | 2026-07 | ICML 2026 | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.02466 | https://arxiv.org/abs/2607.02466 | null | null | Task-Agnostic Pretraining framework that learns transferable motor priors from unlabeled interaction data before grounding them with limited language-instruction data | vla; robot-learning; representation-learning; imitation-learning | robot-camera-video; robot-actions; language-instructions |
The Moving Eye | model | 2026-07 | IROS 2026 | watch | adjacent | 2,026 | https://arxiv.org/abs/2607.02322 | https://arxiv.org/abs/2607.02322 | null | null | Hybrid dynamic data-collection strategy for VLA spatial generalization, using a dual-arm setup where one arm manipulates while the other acts as a moving environmental camera | vla; robot-learning; active-perception; data-collection; spatial-reasoning | robot-camera-video; robot-actions; multiview-video |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.