01CABLE / CONNECTOR
Cable & connector manipulation
Routing, insertion, fastening, wire-harness handling and connector assembly.
Discuss a data need INDUSTRIAL SKILL DATA FOR PHYSICAL AI
Capture the industrial skills robots still need to learn.
Task-specific real-world training data from skilled manufacturing workflows across Asia — built around the physical capability your model is missing.
SHANGHAIOn-the-ground starting point
ASIA-FIRSTManufacturing environments
CUSTOMSpecification before collection
LIVE COLLECTION CONCEPTTM / 001
OBJECT / WORKFLOWINDUSTRIAL MANIPULATION
HUMAN SKILL→STRUCTURED DATA→ROBOT LEARNING
CABLE ROUTING•CONNECTOR INSERTION•ELECTRONICS ASSEMBLY•TOOL USE•MACHINE INTERACTION•FLEXIBLE MATERIALS
THE DATA GAP
More hours are not the goal.
Useful physical experience is.
Real-world robot learning is constrained by data quality, task coverage and access to the physical situations models must master.
TaskMetra is focused on high-information industrial skills that are difficult to reproduce inside a robotics lab — and on collecting them only when a customer specification calls for them.
REAL-WORLD CONTEXTManufacturing complexity
Manufacturing complexity
is part of the data.
Tools. Fixtures. Variation. Failure modes. Sequencing. Human technique.
WHAT MATTERS
01Task relevanceCollect the behavior the model actually needs.
02Operator skillLearn from people who perform the work professionally.
03VariationCapture meaningful changes in objects, setup and execution.
04Quality gatesDefine what makes an episode useful before scaling volume.
WHAT WE CAPTURE
Industrial skills,
not generic video.
Our wedge is the physical work where dexterity, sequencing, tools, deformable objects and real production variation matter.
02PRECISION ASSEMBLY
Electronics & small-part assembly
Fine manipulation, sequencing, alignment, fastening and component handling.
03TOOLS / FIXTURES
Tool use
Skilled operator interactions with tools, fixtures and manufacturing equipment.
04MACHINE TENDING
Machine interaction
Loading, unloading, setup, fixture handling and operator–machine workflows.
05DEFORMABLE OBJECTS
Flexible materials
Textiles, films, packaging, cables and other deformable objects.
06LONG-HORIZON
Kitting, packing & inspection
Long-horizon workflows combining perception, decisions and manipulation.
WHY TASKMETRA
Start with the missing skill.
Build the data around it.
We do not want to win by being the cheapest source of video hours. The collection begins with a model gap and a real industrial behavior.
Skilled operators
Demonstrations from people who perform physical tasks professionally — not actors approximating the work.
Workflow realism
Tools, objects, sequencing, variation and failure modes from manufacturing-relevant environments.
Asia-first access
Built from Shanghai with a focus on difficult-to-access manufacturing workflows across China and Southeast Asia.
Specification first
We scope task, modality, diversity and acceptance criteria before collection. No generic dataset pushed onto the customer.
GENERIC COLLECTIONavailable collectors → maximum hours → dataset
VS
TASKMETRAmodel gap → industrial skill → targeted collection
HOW IT WORKS
Your model gap becomes
a collection program.
Define
Tell us the physical skill, environment, embodiment and model constraint.
Design
We turn the gap into tasks, operators, environments, sensors and acceptance criteria.
Capture
Task-specific demonstrations are collected with qualified operators in relevant environments.
Validate
Episodes are checked against the agreed protocol, provenance and quality gates.
Deliver
Structured, documented data is prepared for your training or evaluation pipeline.
CUSTOMER→SPECIFICATION→COLLECTIONNever the other way around.
DATA MODALITIES
Built around your
data specification.
Collection architecture is scoped to the model requirement. We do not pretend one sensor stack fits every robotics problem.
Egocentric RGBMulti-view RGBRGB-D / depthTask & action labelsSuccess / failure statesObject metadataHand / pose informationProcess & SOP contextAdditional sensing and robot-specific collection can be scoped with the customer.
EXAMPLE PROGRAM / 01Cable & connector
Cable & connector
manipulation
FROM MODEL GAP TO DATA
A concrete collection
can look like this.
Illustrative program for skilled human demonstrations relevant to robotic cable routing and connector insertion.
POTENTIAL VARIATION
- 30 skilled operators
- 20 connector types
- Multiple cable stiffnesses
- Different orientations
- Success + failed attempts
POSSIBLE OUTPUTS
- Multi-view / egocentric video
- Task segmentation
- Success / failure labels
- Object metadata
- Operator action sequences
FOR MANUFACTURING PARTNERS
Turn industrial know-how
into a data asset.
We are building partnerships with approved manufacturing environments where selected workflows can be captured for Physical AI research and training — with defined scope, permissions, governance and commercial terms.
Become a data partnerSUPPLYSkilled industrial workflowsChina · Southeast Asia
→DATA LAYERTASKMETRAScope · Capture · Validate
→DEMANDPhysical AI teamsTraining · Evaluation
EARLY PARTNERSHIPS
Tell us what your robot
cannot learn.
Tell us the task, behavior or environment where your model is data-constrained. We'll determine whether a targeted real-world collection program in Asia could help.
Discuss the collection
FOUNDER
Eimantas Vabuolas
Founder, TaskMetraBuilding the data layer connecting Physical AI with the world's industrial skills.
Based in Shanghai, focused on China & Southeast Asia.
CONTACTeimantas@taskmetra.com+86 187 2120 3615+370 670 34400 (Viber / WhatsApp / Telegram)WeChat: Eimantas1
Scan to add on WeChat