Does lights-out manufacturing mean no workers at all?

Rarely. A few facilities run genuinely unattended for stretches, but most plants that use the approach automate specific cells or shifts while keeping skilled people for setup, changeovers, maintenance, and the judgment work automation cannot do. The realistic goal is more unattended hours, not zero people.

What industries use lights-out manufacturing?

It fits high-volume, low-variety work with a stable process: CNC machining of consistent parts with robotic tending, injection molding and plastics that already run continuously, and packaging with consistent formats. The common thread is predictability, so the cell can run for hours without a human decision.

What technology does lights-out manufacturing require?

A full stack: reliable machines and a stable process at the base, then automated material handling and buffers, machine vision and sensors, supervisory control (PLCs, SCADA, MES) that can stop safely, and predictive analytics with remote alerting. A gap in any layer ends the unattended run.

Why isn't lights-out manufacturing more common?

Because variability breaks it, material handling is hard to automate reliably, cells can stop safely but rarely fix themselves, and the capital only pays back at high volume. For most plants a supervised cell beats an unattended one on total cost, so lights-out stays the exception.

How do you start moving toward lights-out?

Start at the bottom of the stack, not the top. Make the process reliable and visible first, know which machines stop and why, what true OEE is, and where the process varies, because that instability is what sinks an unattended run. Robots and vision pay off only on a stable, well-measured process.

What is a dark factory?

Dark factory is another name for lights-out manufacturing: a facility that can operate without lighting because no people are present to need it. In practice, fully dark factories are rare, and most so-called dark factories still use partial automation with people overseeing production.

Lights-Out Manufacturing: Fully Automated Production

What unattended production actually means, where it works, the honest prerequisites, and the limits that keep dark factories rare.

Lights-out manufacturing is fully automated production that runs with little or no human presence on the floor, the name comes from the idea that machines do not need the lights on. Robots, machine vision, sensors, and controls handle loading, making, checking, and unloading, so a line or cell can run unattended through nights and weekends.

It is a real capability in a narrow set of plants and a costly mistake in most others. This guide covers what lights-out actually means, where it works, the honest prerequisites, and the limits that keep it rare, so you can tell the difference between an achievable unattended cell and an automation project that will quietly bankrupt a line.

What is lights-out manufacturing?

Lights-out manufacturing is production that continues without operators present. It sits at the far end of an automation spectrum. At one end, a person runs a machine by hand. Add fixtures and a cobot and you have assisted work. Automate a full cell and a supervisor tends several. Push all the way and you reach unattended operation: the machines run, correct minor faults, hold quality, and stop safely on their own when they cannot.

The automation spectrum. Lights-out is the far end, reached in stages, not in one jump.

The idea is old. Heavy automation pushes date back to the mid-twentieth century, and General Motors invested enormous sums in automated plants in the 1980s, a program remembered as much for its cost overruns as its ambition, and a useful caution that automation for its own sake destroys money. The clearest working example is the robot maker FANUC, which has run lights-out cells producing robots since 2001, with machines building machines unattended for stretches measured in days. That is the honest picture: proven, but concentrated in a few high-volume, highly repeatable operations.

Where does lights-out manufacturing actually work?

Lights-out works where the product is stable, the volume is high, and the variety is low. The unattended hours have to be predictable, which means the process cannot depend on human judgment to keep running. Plot it on volume against variety and the fit is obvious: the top-left corner, high volume and low mix, is where unattended production pays off. The high-variety corners are where it breaks.

Where lights-out fits. The unattended case is strongest at high volume and low variety.

Typical good fits are CNC machining of stable parts (a robot tends the machines and swaps pallets overnight), injection molding and plastics runs that already run continuously, and packaging operations with consistent formats. The common thread is that the work is predictable enough to run for hours without a decision only a person can make.

Look closely at what those cases share. The part geometry does not change mid-run. The raw material arrives in a form a robot can grip the same way every time. The quality check is something a camera can make objectively, not a judgment an inspector makes by feel. And the failure modes are known, so the cell can be programmed to recognize and stop on them. Take any one of those away, introduce a part that warps, a material that varies, a defect only a human notices, and the unattended window collapses back to a supervised one. The best lights-out candidates are often processes a plant has already run for years, because that history is what proves the process is stable enough to trust overnight.

What does lights-out manufacturing require?

Unattended operation is not one technology; it is a stack, and a gap anywhere in the stack ends the unattended hours. The prerequisites are demanding and specific:

The lights-out stack. Each layer depends on the ones below it; unreliable machines undo everything above.

Reading from the bottom up, an unattended cell needs each of these:

Reliable machines and a stable process. This is the foundation and the one most often skipped. Automation multiplies whatever reliability you start with; a process that jams every few hours cannot run unattended. Predictive maintenance and disciplined upkeep come first, not last.
Automated material handling and buffers. Robots, gantries, or AGVs and AMRs to load raw stock and clear finished parts, plus enough buffer capacity to feed the run for its whole unattended window.
Machine vision and sensing. Cameras and sensors to confirm parts are present, correctly oriented, and in spec, so the cell can detect a bad state instead of making scrap for eight hours. See machine vision systems.
Supervisory control that can stop safely. PLCs SCADA and coordination logic that sequences the cell and, crucially, brings it to a safe stop on any condition it cannot handle. A cell that cannot fail safely cannot run unattended.
Fault detection and remote alerting. Condition monitoring and analytics that catch drift early and notify someone off-site, so a small problem at 2 a.m. becomes a phone alert rather than a wrecked night of production.

Notice the order. Companies love to start at the top with robots and analytics. The plants that succeed start at the bottom, with machines that do not break and a process that does not vary.

What are the realistic limits of lights-out manufacturing?

The honest limits are why lights-out remains the exception. Most factories that describe themselves as automated still run partial automation with skilled people on every shift. Four constraints do most of the limiting:

Variability breaks it. Every product change, material inconsistency, or non-standard part is a decision, and unattended cells cannot make judgment calls. High-mix work fits poorly.
Material handling is the hard part. Making the part is often easier to automate than reliably feeding raw material and clearing finished goods for hours without a jam.
Fault recovery is limited. A cell can stop safely, but it rarely fixes itself. Unplanned downtime during unattended hours can mean the line sits idle until morning unless someone is alerted and responds.
The economics are demanding. The capital is large and the payback depends on filling those unattended hours with saleable output. Below a volume threshold, a supervised cell beats an unattended one on total cost.

There is also a staffing reality that the phrase "lights-out" hides. Unattended cells do not remove people so much as move them. Someone still programs the robots, sets up the fixtures, maintains the machines, and responds when the 2 a.m. alert fires. The skill mix shifts from operating machines to keeping automated cells healthy, which is often a harder role to hire for, not an easier one. Plants that treat lights-out as a headcount-elimination project usually discover they have traded a floor of operators for a smaller number of specialists they struggle to find.

The realistic goal for most plants is not a dark factory. It is extending unattended runtime at the margins, an extra hour after the shift leaves, a weekend of machining on stable parts, while keeping people for the judgment work. That is a spectrum you move along, not a line you cross.

How do you move toward lights-out without a moonshot?

You earn unattended hours; you do not buy them. And you cannot automate a process you cannot see. Before robots and vision pay off, most plants need to know, with real data, which machines actually stop and why, what their true OEE is, and where the process varies, the very things that decide whether an unattended run survives the night.

That visibility is the unglamorous prerequisite under the glamorous stack. It is where an operational layer earns its place: read the signals your PLCs already produce, capture the downtime reasons and quality notes that today live on paper, and turn them into a clear picture of reliability and loss. That is the wedge Harmony takes as a manufacturing operating system connect what you already run with no rip-and-replace, so you can find and fix the instability that would otherwise sink an unattended cell, the way plants like CLS replaced paper logging with live visibility. From there, adjacent tooling, machine learning for maintenance AI quality control and broader smart-factory and agentic AI capabilities, has the clean data it needs to safely extend runtime. Lights-out, where it makes sense, is the last step of that path, not the first.

Lights-out manufacturing by the numbers

Context for how far automation has actually spread, from the industry's primary statistical body:

The average global robot density reached 162 robots per 10,000 manufacturing employees in 2023, more than double the level of seven years earlier (International Federation of Robotics, World Robotics).
The Republic of Korea leads at roughly 1,000+ robots per 10,000 employees with the United States, Germany, and Japan far behind, a reminder that even the most automated economies are nowhere near people-free (International Federation of Robotics, robot density by country).
Genuinely unattended plants remain exceptions, not the rule: FANUC has run lights-out robot production since 2001 but most "automated" factories still staff every shift.

Read those honestly: robots are spreading fast, but density in the low hundreds per 10,000 workers is a long way from lights-out. For nearly every plant, the win is more automation and more unattended hours, not zero people.