Venture in development

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.

Domestic and commercial cleaning
Hospitality and food-service workflows
Retail and product handling
Warehousing and fulfilment
Facilities and building operations
Human-object interaction
Repetitive manual processes
Difficult or unusual edge cases

How the model could work

From task need to structured dataset.

01

Define requirements

Clarify the task, environment, capture mode, metadata and success criteria.

02

Identify environments

Find real-world settings with suitable workflows, layouts and participants.

03

Secure consent

Set participation agreements, usage rights and governance expectations.

04

Capture demonstrations

Collect repeated examples through video, first-person footage or other agreed modes.

05

Structure and check

Annotate task sequences, interactions, outcomes and data quality issues.

06

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.

Which tasks robotics companies will pay to collect
One-off datasets versus ongoing data supply
Required volume, diversity and sensor requirements
Annotation, metadata and quality expectations
Ownership, licensing and usage rights
Acceptable collection costs
Suitable partner environments
Privacy, consent and governance requirements

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