Building the data layer for robots operating in the real world.
An early-stage venture exploring how structured, consented human task data can help robotics and embodied-AI companies train and evaluate systems in real environments across Asia.
Task
Human demonstration
Context
Rooms, tools, objects
Rights
Consent and usage scope
Structure
Sequences and metadata
Quality
Checks and annotation
Use
Training and evaluation
The problem
Real environments are difficult to reduce to clean training data.
Robots are increasingly expected to operate outside controlled labs. In practice, ordinary environments are inconsistent, crowded, culturally specific and full of small human decisions.
Task variation
People complete the same task with different tools, layouts, sequences and recovery actions.
Human-object interaction
Useful data needs to show how people handle products, spaces, obstacles and mistakes.
Operational collection
Access, consent, annotation, consistency, quality control and rights management all matter.
The venture thesis
Robotics companies do not only need more data. They need the right demonstrations.
The working thesis is that robotics and embodied-AI teams may still struggle to obtain sufficiently varied, structured, legally usable and task-specific real-world data from Asian environments.
This venture is assessing whether a specialised collection network in Asia could provide ongoing task data with clear rights, useful metadata and enough variation to support training and evaluation.
What could be collected
Task areas currently being evaluated.
These are exploration areas, not finished services or available datasets.
How the model could work
From task need to structured dataset.
Define requirements
Clarify the task, environment, capture mode, metadata and success criteria.
Identify environments
Find real-world settings with suitable workflows, layouts and participants.
Secure consent
Set participation agreements, usage rights and governance expectations.
Capture demonstrations
Collect repeated examples through video, first-person footage or other agreed modes.
Structure and check
Annotate task sequences, interactions, outcomes and data quality issues.
Deliver defined data
Provide a dataset or ongoing collection programme. Participants may be compensated depending on the model.
Why Asia
Differentiated environments, commercial density and practical access.
The initial focus is Asia, particularly Hong Kong and nearby markets. Dense cities, logistics, hospitality, retail and facilities operations create varied task environments that may differ from common Western datasets.
Hong Kong may be a commercially connected starting point for testing access, structuring, rights clarity, task specificity and international usability.
Currently being validated
The immediate work is demand discovery.
The goal is not to build a large collection operation on assumption.
Who I want to speak with
Conversations that can sharpen the first pilot.
Robotics companies
Teams seeking task demonstrations, evaluation data or harder-to-source real-world scenarios.
Investors and venture builders
People interested in robotics infrastructure, physical AI and data businesses.
Collection partners
Hotels, cleaners, retailers, warehouses, facility operators and other businesses with suitable workflows.
Researchers and advisers
People with experience in robotics datasets, teleoperation, computer vision, annotation, simulation or data governance.
Founder note
Developed from Hong Kong, with validation before scale.
I am developing this venture from Hong Kong, drawing on my background in AI, digital systems, commercial strategy and building practical services around emerging technology. The immediate goal is to establish where there is real buyer demand, what data is difficult to obtain, and which operating model can deliver it responsibly.
Start a conversation
Working on robotics data, embodied AI or real-world automation?
I am currently speaking with potential buyers, partners, researchers and investors to test the opportunity and shape the first pilot.
Start a conversation