Self-Driving Labs Are Turning AI Into a 24/7 Scientific Research Partner

Self-driving laboratories are combining artificial intelligence, robotics, automated instruments and real-time data analysis to create a closed scientific loop in which machines can propose experiments, run them, analyze the results and select what to test next. U.S. research centers including NC State University, Oak Ridge National Laboratory and the National Laboratory of the Rockies are developing these systems for chemistry and materials discovery, with the potential to compress research cycles that once took years into weeks.

For centuries, scientific discovery has followed a familiar rhythm: a researcher develops a hypothesis, designs an experiment, prepares the materials, runs the test, studies the results and then decides what to try next.

That cycle is now being redesigned around artificial intelligence and robotics.

Across U.S. research institutions, self-driving laboratories are emerging as automated scientific environments capable of connecting experimental design, robotic execution, measurement and data analysis into a continuous feedback loop. Instead of simply automating individual laboratory tasks, these systems can use the outcome of one experiment to determine what experiment should come next.

The technology does not mean that scientists have disappeared from the laboratory. Rather, researchers define the scientific problem and objectives while AI systems and robotic platforms increasingly handle repetitive experimental exploration.

The result is a new model of scientific research in which laboratories can operate for extended periods with far less manual intervention.

What Is a Self-Driving Laboratory?

A self-driving laboratory, or SDL, combines three major capabilities: automated laboratory hardware, software that coordinates instruments and experiments, and algorithms that make data-driven decisions about subsequent experiments.

A typical cycle can look like this: AI proposes a set of experimental conditions, robotic equipment prepares samples, instruments characterize those samples, software analyzes the results, and the AI uses the new data to select the next experiment.

That creates a design–make–test–analyze–learn loop.

Researchers can therefore explore large experimental spaces without manually preparing every sample or entering every measurement. Recent scientific reviews describe SDLs as increasingly capable of autonomous experimental design, execution and interpretation, although scalability and general-purpose operation remain significant challenges.

From One Experiment at a Time to Thousands of Possibilities

Chemistry and materials science often require researchers to test many combinations of ingredients, temperatures, concentrations, reaction times and processing conditions.

Even a seemingly simple experiment can produce a large search space.

The National Laboratory of the Rockies, for example, describes research workflows in which changing several experimental variables can produce hundreds of separate experiments. Its researchers are developing robotic systems for thin-film semiconductors and catalytic nanomaterials so that these repetitive experimental sequences can be performed with greater speed and consistency.

At Boston University, autonomous research systems have already been used to perform thousands of experiments in materials research. Researchers describe robotics and machine learning as a way to conduct large numbers of small-scale experiments that would otherwise require substantial human time and material.

This is where the economics of scientific experimentation begin to change.

A human researcher can only physically prepare and evaluate a limited number of experiments during a working day. A robotic system can repeatedly perform standardized operations, measure samples and continue the experimental cycle without fatigue.

The U.S. Is Building Larger Autonomous Research Platforms

One of the most significant developments in 2026 is the push to move self-driving laboratories beyond individual research groups.

North Carolina State University received a $20 million, four-year National Science Foundation award to establish the SPEED initiative — Self-driving Platforms for Expedited Experimental co-Design in solution phase chemistry and materials science. The program is designed to combine AI, robotics and autonomous laboratory infrastructure to accelerate the discovery of molecules and materials.

Its initial research areas include catalysts for chemical manufacturing, semiconductor materials for energy and electronics, and photocatalytic materials.

NC State describes its automated laboratory infrastructure as the largest system of automated labs at a U.S. academic institution, with experiments capable of running around the clock.

The significance is larger than simply having more robots.

The goal is to create an experimental infrastructure in which every result becomes data for the next decision.

Oak Ridge Is Operating More Than a Dozen Self-Driving Labs

Oak Ridge National Laboratory is also expanding autonomous science.

According to ORNL, more than a dozen self-driving laboratories are operating across the Tennessee national laboratory. These facilities combine robotics, sensors and automation to perform laboratory operations ranging from liquid handling to sample processing, while AI makes at least some experimental decisions.

This model turns the laboratory into something closer to an automated production system for scientific knowledge.

Human researchers still define the problem, establish constraints and evaluate scientific meaning. Machines handle much of the repetitive physical experimentation and data collection.

That distinction is important because today’s self-driving laboratories are not universally independent scientific entities. Their autonomy varies significantly depending on the hardware, software, scientific field and level of human supervision.

AI Is Becoming the Decision-Making Layer

Robotics alone does not create a self-driving laboratory.

A conventional automated laboratory might execute a predetermined sequence of 100 experiments. A self-driving laboratory can potentially decide which of those experiments should happen next based on what it has already learned.

This is where machine learning and AI become particularly important.

An algorithm can identify promising regions of a chemical or materials search space, select experiments that could provide the most useful information, observe the results and update its predictions.

In catalyst research, for example, researchers at NC State are developing a human-AI-robot platform capable of autonomously designing, synthesizing, characterizing and rapidly screening heterogeneous catalysts through closed-loop experimentation. The project is supported by nearly $3 million from the U.S. Department of Energy’s ARPA-E program.

The laboratory effectively becomes an experimental learning system.

Why Materials Discovery Is a Major Target

Materials science is particularly suited to autonomous experimentation because researchers often need to explore combinations of compositions and processing conditions.

A new semiconductor, catalyst, battery material or polymer may depend on several variables simultaneously. Testing them manually can take months or years.

Self-driving laboratories can instead divide the search into many small experiments, measure the results and progressively concentrate resources on the most promising combinations.

A 2026 perspective in npj Robotics describes SDLs as systems capable of exploring large, high-dimensional experimental spaces through robotic experimentation and algorithmic decision-making.

This could be especially important for technologies such as advanced electronics, energy storage, catalysts, pharmaceuticals and sustainable chemical manufacturing.

The “Thousands Per Week” Claim Needs Context

The idea that AI-powered laboratories can perform thousands of experiments in a week is technically plausible in high-throughput and highly automated environments, and U.S. research programs are explicitly pursuing that level of experimental acceleration.

However, it would be misleading to suggest that every U.S. self-driving laboratory currently designs, tests and validates thousands of completely new chemical compounds every week.

Different systems have very different throughputs.

For example, Chemical & Engineering News reported in 2026 that Radical’s autonomous materials laboratory was producing around 70 alloys per week, while running multiple characterization stages on those materials.

Other platforms can execute thousands of microscale experiments when the experiments are highly standardized or when screening rather than completely new material synthesis is involved.

So the more accurate picture is that self-driving laboratories are creating the infrastructure needed to scale experimentation toward thousands of tests and iterations, rather than universally producing thousands of entirely new compounds every week.

The Laboratory Could Become a Scientific Factory

The most important change may not be the number of experiments itself.

It is the possibility of converting scientific discovery into a continuous process.

A conventional laboratory might operate according to a researcher’s schedule. An autonomous laboratory can potentially continue collecting data overnight, identify promising results and prepare the next experimental batch before researchers return.

NC State researchers describe the objective of their new national test bed as compressing discovery of functional materials and molecules from years toward weeks in selected research areas.

That does not mean every scientific discovery will suddenly take weeks. Complex experiments, biological validation, scale-up, safety testing and independent confirmation can still require substantial time.

But it changes the earliest and often slowest part of discovery: finding which possibilities deserve closer investigation.

Humans Are Still Critical

Despite the name, self-driving laboratories are not replacing scientific judgment altogether.

Researchers still need to define meaningful questions, establish safety boundaries, interpret unexpected observations and determine whether a result is scientifically significant.

There are also technical challenges around reliability, interoperability, data quality and reproducibility. A recent Nature Reviews Chemistry review identifies scalability, generalizability and complete experimental provenance as major requirements for the next generation of SDL infrastructure.

In other words, the future laboratory may not be AI versus scientists.

It may be scientists working at a higher level while AI and robotics handle increasingly large portions of the experimental search.

A New Era of Automated Discovery

The emergence of self-driving laboratories represents a broader shift in how scientific research is conducted.

AI is moving beyond analyzing information on a computer screen. Robotics is moving beyond isolated laboratory automation. Together, they are beginning to create systems that can think through an experimental strategy, physically perform the experiment, measure what happened and use the result to decide what comes next.

For chemistry and materials science, that could dramatically expand the number of experimental possibilities researchers can investigate.

The laboratory of the future may therefore look less like a room where scientists perform one experiment after another and more like an intelligent scientific engine — continuously generating data, learning from failures and successes, and searching for materials and molecules that humans may never have had the time to discover manually.

FAQ

1. What is a self-driving laboratory?

A self-driving laboratory is an automated research facility that combines AI, robotics and laboratory instruments to design, conduct and analyze experiments with limited human intervention.

2. How do self-driving labs use AI and robotics?

AI analyzes experimental data and decides which experiments could be useful next, while robotic systems prepare samples, operate equipment and collect measurements. The results are then fed back into the AI system.

3. Can self-driving laboratories discover new materials and chemicals?

Yes. They can systematically test large numbers of chemical combinations and material properties, helping researchers identify promising compounds, catalysts, semiconductors and other materials more quickly.

4. How much faster are self-driving labs than traditional laboratories?

The improvement depends on the experiment and the level of automation. Because robots can conduct repetitive experiments continuously and AI can select subsequent tests automatically, some research programs aim to reduce discovery cycles from years to weeks in selected areas.

5. Are self-driving laboratories replacing human scientists?

No. Current self-driving laboratories are primarily designed to augment scientists. Researchers still establish research objectives, safety limits and scientific questions, while AI and robotics handle increasingly large portions of experimentation and analysis.✕Compare with AITOPIA