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 ⌀ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
FARM | model | 2025-10 | arXiv | watch | null | 2,025 | https://tactile-farm.github.io | https://arxiv.org/abs/2510.13324 | https://tactile-farm.github.io | null | Force-aware diffusion-policy framework using a modified UMI gripper with GelSight Mini tactile sensing and an actuated deployment gripper for high-, low-, and dynamic-force tasks | robot-learning; tactile-learning; manipulation; imitation-learning; dexterous-manipulation | wrist-camera-video; tactile; force; robot-actions; human-demonstrations |
UMI on Legs | model | 2024-07 | arXiv | watch | null | 2,024 | https://umi-on-legs.github.io | https://arxiv.org/abs/2407.10353 | https://umi-on-legs.github.io | null | Combines handheld UMI task demonstrations with simulation-trained whole-body controllers to transfer manipulation policies onto quadruped embodiments | robot-learning; cross-embodiment-transfer; mobile-manipulation; imitation-learning; whole-body-manipulation | wrist-camera-video; robot-actions; human-demonstrations; trajectories |
UMI-on-Air | model | 2025-10 | arXiv | watch | null | 2,025 | https://arxiv.org/abs/2510.02614 | https://arxiv.org/abs/2510.02614 | null | null | Embodiment-aware diffusion policy that adapts handheld UMI demonstrations to aerial manipulators through controller-feedback-guided sampling | robot-learning; cross-embodiment-transfer; manipulation; mobile-manipulation; policy-optimization | wrist-camera-video; robot-actions; human-demonstrations; trajectories |
EmbodiSwap | model | 2025-10 | arXiv | watch | null | 2,025 | https://arxiv.org/abs/2510.03706 | https://arxiv.org/abs/2510.03706 | null | null | Photorealistic robot-overlay generation over in-the-wild ego-centric human videos, with synthetic robot datasets over EPIC-KITCHENS, HOI4D, and Ego4D for zero-shot imitation learning | robot-learning; cross-embodiment-transfer; video-generation; imitation-learning; manipulation | egocentric-video; synthetic-video; robot-actions; human-demonstrations |
Mitty | model | 2025-12 | arXiv | watch | null | 2,025 | https://arxiv.org/abs/2512.17253 | https://arxiv.org/abs/2512.17253 | null | null | Diffusion Transformer for human-to-robot video generation that synthesizes paired human-robot examples from large egocentric datasets and evaluates on EPIC-KITCHENS | video-generation; robot-learning; cross-embodiment-transfer; world-modeling | egocentric-video; synthetic-video; robot-actions; human-demonstrations |
tau0-WM | model | 2026-05 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2606.01027 | https://arxiv.org/abs/2606.01027 | null | null | Unified video-action world model trained on about 27,300 hours of real-robot teleoperation, UMI-style interaction, egocentric human videos, and rollout/failure trajectories | world-modeling; robot-learning; vla; manipulation; action-prediction | wrist-camera-video; egocentric-video; multiview-video; robot-actions; language-instructions |
DR-MV3D | model | 2026-06 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2606.23557 | https://arxiv.org/abs/2606.23557 | null | null | Dense-reward framework for multi-view 3D VQA that decomposes reasoning into global-map construction, question-conditioned view planning, and egocentric answer grounding | spatial-reasoning; video-qa; 3d-scene-understanding; embodied-reasoning; reward-modeling | egocentric-video; multiview-video; 3d; qa; text |
EventPrune | model | 2026-05 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2605.19506 | https://arxiv.org/abs/2605.19506 | null | null | Training-free event-camera-guided visual-token pruning for first-person dynamic spatial reasoning, introducing the ESR-Real RGB-event benchmark | efficient-inference; spatial-reasoning; video-qa; ar-sensing-navigation | first-person-video; event-camera; qa |
SpatioRoute | model | 2026-05 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2605.18209 | https://arxiv.org/abs/2605.18209 | null | null | Question-aware prompt routing for zero-shot egocentric spatial QA, specializing prompts by question type without fine-tuning or 3D sensor input | spatial-reasoning; video-qa; zero-shot; prompt-learning | egocentric-video; qa; text |
MAGIC-Video | model | 2026-05 | arXiv | watch | null | 2,026 | https://github.com/lijiazheng0917/MAGIC-video | https://arxiv.org/abs/2605.08271 | https://github.com/lijiazheng0917/MAGIC-video | null | Training-free multimodal memory graph and narrative-chain framework for ultra-long agentic video reasoning over egocentric recordings, live streams, and surveillance footage | memory; long-video-reasoning; video-qa; agentic-reasoning; retrieval | egocentric-video; text; qa; scene-graphs |
World2VLM | model | 2026-04 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2604.26934 | https://arxiv.org/abs/2604.26934 | null | null | Training-time distillation from view-consistent world-model imagination into VLMs for dynamic spatial reasoning under egocentric motion | world-modeling; spatial-reasoning; video-language; representation-learning | egocentric-video; synthetic-video; camera-trajectory; text |
Video ReCap | model | 2024-02 | CVPR 2024 | watch | null | 2,024 | https://sites.google.com/view/vidrecap | https://arxiv.org/abs/2402.13250 | https://sites.google.com/view/vidrecap | null | Recursive captioning model for videos from one second to two hours, plus Ego4D-HCap with 8,267 long-range summaries for hour-long egocentric videos | captioning; long-video-understanding; video-qa; memory | egocentric-video; captions; text |
GPT4Ego | model | 2024-01 | arXiv | watch | null | 2,024 | https://arxiv.org/abs/2401.10039 | https://arxiv.org/abs/2401.10039 | null | null | Zero-shot egocentric action-recognition framework that aligns fine-grained concept descriptions with VLM features over EPIC-KITCHENS-100, EGTEA, and Charades-Ego | zero-shot; action-recognition; video-language; representation-learning | egocentric-video; text |
EVA02-AT | model | 2025-06 | arXiv | watch | null | 2,025 | https://github.com/xqwang14/EVA02-AT | https://arxiv.org/abs/2506.14356 | https://github.com/xqwang14/EVA02-AT | null | EVA02-based egocentric video-language foundation models with spatial-temporal rotary positional embeddings and symmetric multi-similarity optimization | video-language-pretraining; video-text-retrieval; representation-learning; efficient-inference | egocentric-video; text |
X-MIC | model | 2024-03 | ECCV 2024 | watch | null | 2,024 | https://github.com/annusha/xmic | https://arxiv.org/abs/2403.19811 | https://github.com/annusha/xmic | null | Cross-modal instance conditioning framework for adapting VLMs to egocentric action generalization across EPIC-KITCHENS, Ego4D, and EGTEA | action-recognition; domain-generalization; video-language; representation-learning | egocentric-video; text |
EgoAdapt Efficient Perception | model | 2025-06 | arXiv | watch | null | 2,025 | https://arxiv.org/abs/2506.21080 | https://arxiv.org/abs/2506.21080 | null | null | Adaptive multisensory distillation and policy learning for efficient egocentric action recognition, active-speaker localization, and behavior anticipation on EPIC-KITCHENS, EasyCom, and AEA | efficient-inference; action-recognition; audio-visual; anticipation; ar-sensing-navigation | egocentric-video; audio; gaze; imu |
Spatial Cognition / LMK | model | 2024-04 | arXiv | watch | null | 2,024 | https://arxiv.org/abs/2404.05072 | https://arxiv.org/abs/2404.05072 | null | null | Out-of-sight active-object tracking from egocentric video using Lift, Match, and Keep over 100 long EPIC-KITCHENS videos | spatial-reasoning; object-tracking; memory; 3d-scene-understanding | egocentric-video; 3d; object-tracks |
TimeSearch-R | model | 2025-11 | arXiv | watch | null | 2,025 | https://github.com/Time-Search/TimeSearch-R | https://arxiv.org/abs/2511.05489 | https://github.com/Time-Search/TimeSearch-R | null | Reinforcement-learned temporal search for long-form video understanding with completeness self-verification, evaluated on Haystack-Ego4D and broader long-video benchmarks | long-video-understanding; temporal-retrieval; video-qa; reinforcement-learning | egocentric-video; text; qa |
Know-Show | benchmark | 2025-12 | arXiv | watch | null | 2,025 | https://github.com/LUNAProject22/Know-Show | https://arxiv.org/abs/2512.05513 | https://github.com/LUNAProject22/Know-Show | null | 2.5K human-authored questions over Charades, Action Genome, and Ego4D that jointly evaluate spatio-temporal reasoning and visual/temporal grounding | video-qa; spatial-reasoning; temporal-grounding; grounded-reasoning; benchmark | egocentric-video; video; qa; annotations |
PCSR-Bench | benchmark | 2026-05 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2605.12413 | https://arxiv.org/abs/2605.12413 | null | null | 84,373 QA pairs from 2,600 omnidirectional indoor images over eight perspective-conditioned spatial reasoning tasks including egocentric rotation and limited-FOV visibility | spatial-reasoning; question-answering; embodied-reasoning; benchmark | first-person-360-video; egocentric-image; qa; annotations |
SpaMEM | benchmark | 2026-04 | arXiv | watch | null | 2,026 | https://huggingface.co/datasets/mill-ct-liao/SpaMEM | https://arxiv.org/abs/2604.22409 | null | null | Dynamic spatial-memory benchmark with 10.6M high-fidelity images across RGB/depth/instance/semantic modalities from 25K+ action-conditioned interaction sequences in 1,000 houses | spatial-reasoning; memory; benchmark; embodied-reasoning | egocentric-video; depth; segmentation; annotations; qa |
Ego4D-OSCA | benchmark | 2024-05 | arXiv | watch | null | 2,024 | https://arxiv.org/abs/2405.12789 | https://arxiv.org/abs/2405.12789 | null | null | Object-state-change anticipation annotations over Ego4D procedural video, paired with a baseline that integrates visual and language history | anticipation; object-state-change; procedural-reasoning; benchmark | egocentric-video; annotations; text |
OR-Action | benchmark | 2026-06 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2606.13332 | https://arxiv.org/abs/2606.13332 | null | null | Fine-grained multi-role action benchmark built on a public ego-exocentric operating-room dataset, deriving dense actions from scene-graph state changes | surgical-understanding; action-recognition; temporal-segmentation; scene-graph-qa; benchmark | egocentric-video; exocentric-video; scene-graphs; annotations |
IMPACT-HOI | toolkit | 2026-05 | arXiv | watch | null | 2,026 | https://github.com/541741106/IMPACT_HOI | https://arxiv.org/abs/2605.01666 | https://github.com/541741106/IMPACT_HOI | null | Mixed-initiative supervisory-control framework for constructing onset-anchored partial HOI event graphs in egocentric procedural video | annotation-tooling; hand-object-interaction; procedural-reasoning; robot-learning | egocentric-video; annotations; scene-graphs |
EgoScreen-Emotion / ESE | benchmark | 2026-04 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2604.15823 | https://arxiv.org/abs/2604.15823 | null | null | 224 movie trailers captured under egocentric screen-view conditions with 28,667 temporally aligned keyframes and confidence-aware multi-label emotion annotations | affect-recognition; long-context; audio-visual; benchmark | egocentric-video; audio; annotations |
EyeCue | model | 2026-05 | arXiv | watch | null | 2,026 | https://github.com/langzhang2000/EyeCue | https://arxiv.org/abs/2605.07859 | https://github.com/langzhang2000/EyeCue | null | Gaze-empowered egocentric video model and CogDrive annotations for driver cognitive-distraction detection across road, weather, and time-of-day settings | driving; internal-state-reasoning; affect; safety; multimodal-reasoning | egocentric-video; gaze; annotations |
GIST | toolkit | 2026-04 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2604.15495 | https://arxiv.org/abs/2604.15495 | null | null | Grounded Intelligent Semantic Topology pipeline that turns mobile point clouds into semantic navigation topologies for search, localization, zone classification, and egocentric route instructions | navigation; assistive-systems; spatial-localization; semantic-mapping; instruction-generation | point-clouds; maps; text; egocentric-video |
Gaze-to-Guidance Assistants | model | 2026-04 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2604.08062 | https://arxiv.org/abs/2604.08062 | null | null | Controlled study of gaze-grounded multimodal LLM assistance using egocentric video with gaze overlays to infer reading difficulty and personalize retrospective help | assistive-systems; gaze-reasoning; personalized-qa; cognitive-state-aware-assistance | egocentric-video; gaze; text |
VueBuds | toolkit | 2026-03 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2603.29095 | https://arxiv.org/abs/2603.29095 | null | null | Camera-integrated wireless earbuds for egocentric vision, streaming low-power binocular monochrome views to on-device VLMs for scene understanding, translation, visual reasoning, and text reading | ar-sensing-navigation; wearable-interaction; assistive-systems; on-device-perception | egocentric-video; monochrome; wearable-sensing; text |
ROHIT | model | 2025-12 | arXiv | watch | null | 2,025 | https://arxiv.org/abs/2512.07394 | https://arxiv.org/abs/2512.07394 | null | null | Reconstructs Objects along Hand Interaction Timelines from HOT3D and EPIC-KITCHENS stable-grasp clips, propagating object pose across pre-contact, held, and released phases | object-reconstruction; hand-object-interaction; 3d-reconstruction; tracking | egocentric-video; hand-pose; object-pose; 3d |
Single-View Mesh Rotation Stress Test | benchmark | 2026-06 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2606.22987 | https://arxiv.org/abs/2606.22987 | null | null | Controlled roll/pitch/yaw stress-test protocol for single-view mesh reconstruction on Aria Digital Twin and Franka wrist-camera sequences, exposing view-dependent layout and depth failures | 3d-reconstruction; digital-twin-evaluation; robot-learning; benchmark | aria; wrist-camera-video; 3d; object-pose |
The N-Body Problem | model | 2025-12 | ECCV 2026 Workshop | watch | null | 2,025 | https://zhifanzhu.github.io/ego-nbody/ | https://arxiv.org/abs/2512.11393 | null | null | VLM reasoning task for predicting feasible N-person parallel execution from a single egocentric video, evaluated on 100 EPIC-KITCHENS and HD-EPIC videos | procedural-reasoning; spatial-reasoning; planning; video-language | egocentric-video; text; qa |
World Action Models | survey | 2026-05 | arXiv | watch | null | 2,026 | https://arxiv.org/abs/2605.12090 | https://arxiv.org/abs/2605.12090 | null | null | Survey and taxonomy of world action models that unifies VLA, predictive world modeling, robot teleoperation, portable human demonstrations, simulation, and internet-scale egocentric video | world-modeling; vla; robot-learning; survey | survey; egocentric-video; robot-actions |
FutureNav | model | 2026-06 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2606.30367 | https://arxiv.org/abs/2606.30367 | null | null | Unified world-action modeling framework for vision-and-language navigation, jointly optimizing action prediction, inverse/forward dynamics, and future spatial-state prediction from egocentric agent observations | navigation; world-modeling; vla; spatial-reasoning; robot-learning | egocentric-video; text; spatial-features; trajectories |
ZR-0 | model | 2026-06 | arXiv | watch | adjacent | 2,026 | https://github.com/RUCKBReasoning/ZR-0 | https://arxiv.org/abs/2606.30552 | https://github.com/RUCKBReasoning/ZR-0 | null | 2.6B VLA model trained with dense embodied chain-of-thought supervision over ProcCorpus-60M to align cross-embodiment manipulation representations without inference-time reasoning overhead | vla; robot-learning; manipulation; cross-embodiment-transfer; reasoning | video; text; robot-actions; trajectories |
VLK | model | 2026-06 | arXiv | watch | adjacent | 2,026 | https://vision-language-kinematics.github.io/ | https://arxiv.org/abs/2606.30645 | null | null | Vision-language-kinematics supervision pipeline generating 48K synthetic humanoid loco-manipulation trajectories with paired egocentric renderings, instructions, and whole-body kinematic targets in reconstructed scenes | humanoid-control; loco-manipulation; vla; robot-learning; synthetic-data | synthetic-video; egocentric-video; text; trajectories; humanoid-motion |
HUMEMBR | model | 2026-06 | IROS 2026 | watch | adjacent | 2,026 | https://arxiv.org/abs/2606.30404 | https://arxiv.org/abs/2606.30404 | null | null | Human-Centered Memory for Embodied Robots, building structured long-term human-routine memories for embodied question answering and routine-conditioned navigation on physical robots | memory-reasoning; navigation; embodied-reasoning; question-answering; robot-learning | robot-observations; memory; text; trajectories |
WARP Retarget | model | 2026-06 | arXiv | watch | adjacent | 2,026 | https://warp-retarget.github.io/ | https://arxiv.org/abs/2606.29940 | null | null | Whole-body-aware retargeting pipeline that converts offline human demonstrations into precise mobile-manipulator trajectories using shoulder-elbow-wrist geometric solving and lazy mobile-base control | robot-learning; imitation-learning; retargeting; mobile-manipulation; cross-embodiment-transfer | human-demonstrations; human-motion; trajectories; robot-actions |
ContactWorld | benchmark | 2026-06 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2606.13877 | https://arxiv.org/abs/2606.13877 | null | null | Vision-tactile world-model benchmark across 12 contact-rich robot manipulation tasks, comparing wrist-view, front-view, point-cloud, and tactile force-field representations | world-modeling; tactile-learning; robot-learning; manipulation; benchmark | wrist-camera-video; exocentric-video; tactile; point-clouds; robot-actions |
ACWM-Phys | benchmark | 2026-05 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2605.08567 | https://arxiv.org/abs/2605.08567 | null | null | Action-conditioned video world-model benchmark for generalized physical interaction, spanning rigid, kinematic, deformable, and particle dynamics in a controllable simulator | world-modeling; physical-action-understanding; benchmark; evaluation | synthetic-video; robot-actions |
3DG-VLN / UAV-VLN-FOV | benchmark | 2026-06 | arXiv | watch | adjacent | 2,026 | https://github.com/xuefanfu/3DG-VLN | https://arxiv.org/abs/2606.20045 | https://github.com/xuefanfu/3DG-VLN | null | 2,717 high-resolution target-visible UAV trajectories with front/downward egocentric observations and continuous 3D waypoint annotations for see-and-reach VLN | navigation; spatial-reasoning; benchmark; robot-learning | egocentric-video; trajectory; text; annotations |
JAXenstein | benchmark | 2026-05 | arXiv | watch | adjacent | 2,026 | https://arxiv.org/abs/2605.19926 | https://arxiv.org/abs/2605.19926 | null | null | JAX-based Wolfenstein 3D renderer for accelerated reinforcement-learning benchmarks in visual first-person tasks under partial observability | reinforcement-learning; benchmark; virtual-egocentric-reasoning | first-person-video; synthetic-video |
DenseStep2M | dataset | 2026-04 | arXiv | watch | adjacent | 2,026 | https://huggingface.co/datasets/mingjige/DenseStep2M | https://arxiv.org/abs/2604.26565 | null | null | Training-free procedural-video annotation pipeline producing about 100K videos and 2M dense instructional steps, with zero-shot transfer evaluations on egocentric, exocentric, and mixed-perspective domains | procedure-understanding; captioning; temporal-grounding; retrieval; benchmark | video; captions; text |
RoboPearls | toolkit | 2025-06 | arXiv | watch | adjacent | 2,025 | https://arxiv.org/abs/2506.22756 | https://arxiv.org/abs/2506.22756 | null | null | Editable 3D Gaussian video-simulation framework for robot manipulation, evaluated on RLBench, COLOSSEUM, Ego4D, Open X-Embodiment, and real-robot scenes | world-modeling; robot-learning; video-generation; manipulation | synthetic-video; robot-actions; 3d |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.