ROBOWORK.DATA
STUDIO_CAPABILITIES

Multiview recording profiles and task families.

We operate structured recording setups across five core task categories. We collect synchronised RGB and spatial streams under precise protocol guidelines.

Recording Profiles

Standardized recording configurations.

We capture human demonstrations across three distinct technical tiers based on policy complexity.

Basic Multiview

Ego / Chest Camera + Static Workspace Camera

A standardized chest-mounted or head-mounted ego camera paired with a single workspace-level camera. Continuous, aligned timeline mapping natural-language instructions to success/failure segment tags.

Specs Preview

  • 1080p @ 30 FPS
  • Standard RGB sensor
  • Natural language instructions
  • Start & end states verified

Advanced Multiview

Ego Camera + Front camera + Side camera

Standardized chest camera aligned with dual environmental cameras (front workspace perspective + side overview perspective). Delivers robust coverage of hand-object occlusion, grasp trajectories, and workspace interactions.

Specs Preview

  • 1080p @ 30/60 FPS
  • Color-calibrated workspace
  • Obstruction-free angles
  • Unified sync trigger marker

Spatially Enriched

Multiview RGB + Depth + IMU + Calibration (Optional)

RGB-D depth maps, IMU sensor synchronization, camera pose parameters, and 3D hand/object trajectories. Designed for complex manipulation research requiring depth-aligned coordinates.

Specs Preview

  • RGB-D depth sync
  • IMU sensor alignment
  • Camera calibration matrix
  • 2D/3D hand keypoints
⚠️ Capability Note: Available for selected projects based on technical feasibility and customer requirements.

Task Catalog

Supported task families.

We collect data across structured, low-risk service actions designed to pretrain physical reasoning models.

Cleaning and Surface Interaction

  • Wipe a surface
  • Clear a table
  • Simulated spill response
  • Pick up loose trash
  • Organize cleaning supplies
  • Move cleaning tools
  • Simulated floor-cleaning pass

Object Handling & Sorting

  • Sort objects into containers
  • Group items by category
  • Match objects to labeled positions
  • Organize objects by size
  • Place objects in a defined sequence
  • Separate recyclable materials

Household & Mock-Apartment Tasks

  • Open drawer and place item
  • Organize shelf
  • Place objects in cabinets
  • Fold simple textiles
  • Arrange household items
  • Clear a workspace

Light Logistics & Handling

  • Move parcel from floor to shelf
  • Sort lightweight parcels
  • Place packages into designated zones
  • Load and unload small containers
  • Pick and place labeled objects

Tool & Assembly Demonstrations

  • Select a tool
  • Sort tools
  • Match screws and components
  • Connect non-powered components
  • Insert parts
  • Assemble simple mock components
  • Route cables through holders
  • Organize hardware parts
🚫 Safety Exclusions Policy: We explicitly exclude tasks involving high-voltage electrical contacts, open flames, heavy machinery operation, high elevation, hazardous chemicals, medication dispensing, weapons, or covert recording.

Policy Limits

Direct robot control requires hardware-specific actions.

Human demonstration data focuses on planning, affordance, and temporal segmentation. Direct policy execution on physical systems demands target proprioception.

💡 Human demonstration datasets can support pretraining, task understanding, visual representation learning, planning, and human-to-robot transfer. Direct robot control generally requires additional robot-specific state and action data.

Explore a customized collection protocol

Contact our engineers to map your task verb ontology to our camera positions.

Request a Dataset Pilot