Manufacturing data for robotics

From the
shop floor to your
training pipeline

TalosHub captures real manufacturing workflows and packages them into structured, white-labeled datasets that robotics teams can train on immediately.

We go into factories and build
the data your robots need

Most robotics training data comes from labs. The real world doesn't look like a lab. It looks like a CNC shop at 6 AM with chip buildup, fixture variance, and operators who've been doing this for twenty years.

TalosHub has direct access to manufacturing facilities. We capture expert operator behavior, machine state context, and exception patterns — then structure everything into training-ready task packs your team can ingest on day one.

Not raw footage. Not generic datasets. Scoped, labeled, segmented, documented, and white-labeled under your brand.

CNC manufacturing facility with operators working at machining centers

One of our partner manufacturing facilities

The full cycle, not just the footage

Task scoping, on-site capture, and complete packaging — from a single manufacturing facility to your repo

We define the workflow, camera plan, machine-state capture points, and exception taxonomy before collection begins. Then we show up, capture, clean, label, package, and hand off.

Machine-State Alignment

Every episode is synchronized with real machine events — door state, chuck state, cycle timing, alarms, HMI context.

Exception Coverage

Misloads, retries, jams, operator interventions, alarm responses. The data that matters for real deployment.

White-Label Packaging

Delivered under your brand. README, schema docs, sample loader, train/val/test splits.

Multi-Facility Access

CNC shops, press/brake, metrology, finishing lines. Repeatable schema across any facility.

Expand Over Time

Add episodes, machine variants, exception classes, or new workflows. Packs grow with your roadmap.

From scoping to handoff

Every step is structured, tracked, and delivered under your brand.

Task Scope — CNC Machine Tending
Workflow Configuration
Workflow Load / Unload Cycle
Machine Haas VF-2
Facility Shop Floor A
Episodes target 50
White-label Enabled
01
Scope

Define the workflow, machine family, and delivery spec. We align on the task boundary, exception classes, and format before anything is filmed.

Live Capture Dashboard
CAM-01
Episodes 34/50
Duration 2:14
Recording
CAM-02
CAM-03
02
Capture

On-site sessions with expert operators and real machines. Multi-view recording synchronized to live machine state events.

Data Labeling Interface
Episode Timeline
Phase 1
Phase 2
Exc.
Phase: Load State: Chuck Open Exception: Misload Action: Insert
episode_042.json
frames1,840
duration3.2s
phaseload_cycle
03
Structure

Segment phases, tag exceptions, align machine states. Every frame is accounted for — labeled, validated, and traceable.

Export & Package
dataset/
├── episodes/ (50 files)
├── labels/manifest.json
├── docs/schema.md
├── sample_loader.py
└── README.md
Train / Val / Test split
70%
15%
15%
04
Package

Labels, splits, schema docs, sample loader — all white-labeled. Ready to drop into your training pipeline on day one.

Delivery Summary
Delivery Package — Ready
50
Episodes
4
Exception classes
97.2%
Quality score
White-labeled
Schema docs
Sample loader
Review scheduled
05
Deliver

Handoff, review, and a gap map for what comes next. You own the data — we document everything for your team to build on.

Built for teams that ship industrial AI

If you're building robots that need to work in real manufacturing environments, you need data from real manufacturing environments.

Robotics Companies

Seed industrial pilots with real factory data instead of lab demos

VLA Model Builders

Expand manipulation coverage with structured, real-world demonstrations

Robot OEMs & Integrators

Reduce deployment pain with task-specific training data and recovery patterns

Industrial AI Research

Access manufacturing-native benchmarks that don't exist in public datasets

Common questions

What kind of robotics training data does TalosHub provide?
TalosHub provides manufacturing-native training data for robotics companies. We capture real operator workflows from CNC machine tending, press/brake operations, metrology, and finishing processes. Every dataset includes multi-view video, machine-state alignment, phase segmentation, exception coverage, and success/failure labels — structured and ready for VLA model training, imitation learning, and robot manipulation research.
How is TalosHub different from other robotics data providers?
Most robotics data comes from lab environments or broad humanoid demonstrations. TalosHub is manufacturing-native — we capture data directly from factory floors with real machines, real operators, and real exceptions. Every task pack includes machine-state context (door state, chuck state, cycle timing, alarms) that lab data simply doesn't have. We deliver white-labeled packages, not raw footage.
What industries and machines does TalosHub cover?
We currently cover CNC machine tending, press and brake tending, metrology and inspection workflows, and finishing and polishing operations. Our capture pipeline extends to any machine-centered manufacturing workflow. We work directly with manufacturing facilities across multiple regions.
Who uses TalosHub data?
Our clients include robotics companies building industrial manipulation systems, VLA model builders expanding their training coverage, robot OEMs and integrators reducing deployment friction, and industrial AI research teams that need real-world manufacturing benchmarks.
What is included in a TalosHub task pack?
Each task pack includes 25–150 usable episodes with multi-view capture, machine-state aligned metadata, phase segmentation labels, 2–8 exception classes with recovery patterns, train/validation/test splits, a sample data loader, schema documentation, and a README — all white-labeled under your brand.

Ready to build with
real factory data?

Tell us the workflow. We'll scope a task pack, capture it on-site, and deliver training-ready data your team can use immediately.

Get in Touch
hello@taloshub.io