Multi-Step Household Task Data for Home Robots
Multi-step household task demonstrations with task graphs, object states and recovery labels for home robots and long-horizon VLA. Design a program with Shaip.
Dataset Specifications
| Target Buyer Teams | Home Robotics Lead; VLA Research Lead; Robot Policy Team; Product Engineering |
| Representative Tasks | Meal preparation, cleaning, laundry, organizing, table setting and appliance interaction |
| Capture / Modalities | Egocentric + exocentric video; optional depth, audio, hand pose and object-state sensors |
| Annotations / Metadata | Task graph, subgoals, object states, action narration, success/failure and recovery steps |
| Delivery Formats | MP4 + JSON/Parquet task graph; LeRobot-compatible episodes only where robot state/action exists |
| Scale (Illustrative)* | 500–3,000 demonstration hours |
| Participants / Operators | 200–1,000 participants |
| Tasks / Scenarios | 50–150 tasks |
| Objects / Sites / Views | 20–100 environments |
| Recommended Engagement Scope | 100–250 hours across 10–20 tasks and 5–10 environments |
A standardized extension of Shaip's egocentric, exocentric and procedural capture operations.
View the related case study →Where This Dataset Is Used
This household robot dataset is built for teams training home robots and long-horizon VLA models on realistic multi-step tasks such as cooking, cleaning, laundry and tidying. Home robotics companies and embodied AI labs use this household task dataset for task planning, subgoal decomposition, error recovery and long-horizon manipulation policy learning. If you’re sourcing home robot training data, a procedural task dataset or embodied AI household data with task graphs, object states and recovery labels, this dataset provides demonstrations that mirror how tasks actually unfold in real homes.
Request a Sample of the Household Long-Horizon Task Demonstration Dataset
Tell us about the robot product or Physical AI system you’re building and we’ll respond with sample access, specifications and a proposal matched to your task.
"*" indicates required fields
