Industries / Industrial equipment and services / IoT and industrial automation
Digital twins
A digital twin is a computer model of one particular machine, plant or building that is kept in step with the real thing by data from it, so that people can see its condition, try changes on the copy first and predict what the real one will do.
2002
Michael Grieves presented a model of a physical product paired with a virtual copy of it, the two linked by a flow of data, first at a Society of Manufacturing Engineers (SME) management forum in October 2002 and then at a University of Michigan meeting on product lifecycle management on 3 December 2002, by his own account. He writes that the name "Digital Twin" was later "coined" by John Vickers of NASA.
27
Milestones achieved across 6 domains.
7
To watch for: announced, not yet confirmed outside.
12
Open deliverables. Nobody has shipped them.
In one minute
Where it started
Engineers kept models of particular machines long before the name. From 1958 the US Air Force tracked how much fatigue life each of its aircraft had used, and in 1970 NASA checked rescue procedures for Apollo 13 on ground simulators before reading them to the crew. Michael Grieves set out the twin as a model in 2002, and NASA's technology roadmaps gave it its name around 2010.
Where it stands
The parts are in routine use. OPC UA and MQTT carry machine data, Brick names building data, and standards now fix the structure of a machine's digital record (IEC 63278-1, 2023) and of a 3D scene (OpenUSD Core Specification 1.0, 2025). BMW planned its Debrecen car plant in a virtual copy before it was built. But the US National Academies found in 2024 "a lack of standards in reporting" how far a twin's outputs can be trusted, and most published results, such as the 20% throughput gain Siemens reports for PepsiCo, are the sellers' own figures.
What is open
Building and updating the model is still mostly hand work: turning scans, piping drawings and lists of sensor names into models, and noticing when the real asset has changed. Few twins publish their forecasts before the fact and score them afterwards, and few savings claims are checked at the meter. Phone scanning, Gaussian splats stored in OpenUSD, AI agents that build simulations from video and language models that read engineering data now make these jobs possible to attempt.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Building the model
Building the model means making the virtual copy of one particular asset: its shape, its parts and how they connect, taken from drawings, scans and the makers' data.
4 · 1 · 2
Last achievedOne file format for scenes2025
Watch forSimulations built from phone video
2
openConnecting data
Connecting data means getting readings out of machines and building systems, naming each one so that software knows what it measures, and carrying it to the model.
6 · 2 · 2
Last achievedPrinciples for connected twins2018
Watch forA standard building data model
2
openSimulation
Simulation is the part of a twin that calculates what the real thing will do next, from physics, from past data or from both.
4 · 1 · 2
Last achievedA national review of gaps2024
Watch forWorld models as simulators
2
openUse in operations
In operations the twin helps run the real asset day to day: planning changes, rehearsing procedures and training the people who run it.
4 · 1 · 2
Last achievedA car plant tried virtually2025
Watch forThroughput gains from factory twins
2
openUse in maintenance
In maintenance the twin follows each unit's own use and wear, so that inspections and repairs can be set by that unit's history rather than by one schedule for the whole fleet.
4 · 1 · 2
Last achievedA twin for each vehicle2012
Watch forEngine maintenance by analytics
2
openKeeping it current
Keeping a twin current means updating the model when the real asset changes, and checking again and again that it still matches what is measured.
5 · 1 · 2
Last achievedA legal duty for records2024
Watch forA model synced to controls
2
open
Advances to build with
2 research notes

MIPS Actus
Maintenance inside the joint
MIPS says its Actus chip can drive a robot joint's motor and also run models that predict when maintenance is due, and no independent measurement was found.

Agentic robotics
Readable robot programs
A UC Berkeley lab has AI agents write readable robot programs, which scored below human-written ones in all twelve of its own test conditions on one simulator.
Open in digital twins
3 of 12 open deliverables
Open · Building the model
Plant rooms from phone scans
A pipeline that turns a phone or handheld laser scan of a plant room (the room that holds a building's pumps, boilers, pipes, valves and ducts) into an editable building model in IFC, the open file format for building models. Every pipe becomes a pipe object of the right diameter, joined to its neighbours at the fittings, and each object keeps a link to the photos it came from.
Built with what exists
an iPhone Pro's LiDAR sensor with a scanning app such as Polycam, or a borrowed handheld laser scanner; COLMAP to work out where each photo was taken; gsplat or Nerfstudio for a Gaussian splat of the room, stored with OpenUSD v26.03's splat schema; Open3D to fit cylinders to the point cloud; Meta's Segment Anything 2 to pick out pipes and valves in the photos; IfcOpenShell, an open-source library, to write the IFC file; a coding agent such as Claude Code to join the steps.
Open · Connecting data
Brick models from point lists
A tool that takes a building automation system's export (point names, units, descriptions and a month of trend data) and writes a Brick model, which an engineer then accepts or corrects point by point.
Built with what exists
the Brick schema and the brickschema Python library, with SHACL checks; the Mortar collection of Brick-labelled buildings, which PNNL used for training; features from the trend data (units, range, how a value moves) computed with pandas; a strong language model such as Claude Fable 5.2 to read names and descriptions; Gemini Decision 1, which gives probabilities for fixed-answer questions, to flag the points a person should check (see the Note "A thousand questions, one pass").
Open · Simulation
Forecasts posted before the fact
A small physical rig, such as two water tanks with a pump and valves, and a twin of it that publishes each day a forecast of the next 24 hours of levels and flows, with uncertainty bands, before those hours happen. A script scores every forecast against what was measured.
Built with what exists
OpenModelica for the physics model, exported as an FMU and run with FMPy; a Raspberry Pi with level and flow sensors; Mosquitto, an open-source MQTT broker; InfluxDB and Grafana to store and plot the data; scikit-learn or PyTorch for a data-driven correction to the physics.