Skip to content
RFL_GLOBAL
中文

DATA FIRST. MODELS NEXT.

DATA

Proprietary POV hardware captures human experience at industrial scale. Curated streams for manipulation, interaction and long-horizon learning.

DATA PRODUCTS
3
FRONTIER RESEARCH
3
30 FPS
1080p
POV PIPELINE // ACTIVE
01CURATED POV STREAMS

DATA PRODUCTS

  • MANIPULATION TASKS: HAND-OBJECT INTERACTIONDS_001

    HAND-OBJECT INTERACTION

    MANIPULATION TASKS

    High-density POV clips of real human manipulation: cooking, cleaning, picking, placing, tool use, and more.

    • dense trajectories
    • action-rich
    • 1080p/30fps
  • HUMAN-ENV INTERACTION: PERCEPTION & AFFORDANCEDS_002

    PERCEPTION & AFFORDANCE

    HUMAN-ENV INTERACTION

    POV videos with object cues, spatial context, and natural scene dynamics from real home/work environments.

    • affordances
    • context-rich
    • indoor workflows
  • LONG-HORIZON EPISODES: MULTI-STEP SEQUENCESDS_003

    MULTI-STEP SEQUENCES

    LONG-HORIZON EPISODES

    Extended POV episodes with multi-step actions, environment changes, mistakes, and recoveries.

    • complex planning
    • error-recovery
    • temporal-reasoning
02STATE-OF-THE-ART BENCHMARKS

FRONTIER RESEARCH

  1. 01

    POV-LEARN

    IMITATION LEARNING

    First-person video data pipelines for scalable imitation learning. Enabling robots to learn from human demonstrations.

  2. 02

    ACTION-VLA

    VLA MODELS

    Vision-Language-Action model integration research. Building bridges between perception and robotic control.

  3. 03

    DEXTEROUS-X

    ROBOTIC MANIPULATION

    Advanced manipulation priors from egocentric video datasets. Teaching robots fine-grained motor skills.

03RESTRICTED // DATA CATALOGUE

DATA CATALOGUE

Browse samples, schemas and licensing for every dataset. Sign in with an institutional account to request access.