Projects
In progress · 2026
FloorTruth
Manufacturing Agent Verification Protocol
A protocol and testbed for verifying AI agents on the factory floor. Agent execution logs are checked against user-supplied ground truth and constraints, and task outcomes are separated from human assistance. The current form is a set of synthetic scenarios and a log auditor — not a certification of real processes, quality or safety.
In progress · 2026
DAYBREAK
From the screen to the street — an embodied showcase
An AI agent rehearses in a digital twin, then moves into the world on a small robot: it builds a body, migrates onto it, and walks out. The twin is built first and the physical run follows; the points where people help are recorded, not hidden.
Dataset · 2026
Industrial HRC-Bench
Real-world reference data for industrial human–robot collaboration
Captured at the KTL Robot Testing & Certification Center with LiDAR, RGB-D, motion capture, robot state and 360° imagery in sync, alongside a Unity digital twin pipeline that generates proximity and pose ground truth. Presented at Unite Seoul 2026.
Ongoing
Agent workflows for manufacturing AI
Twin · commissioning · physical robots · task verification
We build agent workflows for manufacturing AI: an agent plans and acts in a digital twin, is commissioned against the twin before touching hardware, drives physical robots, and has its work verified against ground truth and constraints. The pieces are shared across FloorTruth, DAYBREAK and the twin pipeline.
Research
Digital Twins & Synthetic Data
If you can't collect it, generate it — from what you can measure. We anchor a Unity digital twin to real measurements so that physics, rendering, sensor simulation and ground-truth labels share one simulation clock.
Human–Robot Collaboration
The data we actually need — a worker approaching a robot, partly occluded, crossing a safety zone — barely exists in public form. We build it: a real-world reference benchmark and a calibrated pipeline that generates proximity and pose ground truth.
Real-to-Sim
We built the twin. It still didn't match. Residual gaps — the camera, and the ground truth — are found and closed one by one: Ground, Calibrate, Generate.
Manufacturing AI
Digital twins for logistics, foundry and EV assembly lines — a fuller portfolio will follow here.