Autonomous Experimentation for the Future of Science
Science at the speed of nature
At Pacific Northwest National Laboratory (PNNL), the labs of the future are being built today.
Because the science of tomorrow will not look like the science of today.
Intelligent data systems.
AI agents.
Robotics.
All play a role. But speed alone isn’t going to help us meet our ambitious goals.
PNNL’s Center for Robotics and Autonomy is accelerating scientific discovery at speed and scale with the desired level of human oversight. We’re building in reproducibility and trustworthiness from day one.
The human question starts it all
It all starts with the human drive to solve scientific problems and find solutions that make a difference.
This isn’t new. What’s new is our ability to combine human problem-solving with tools that integrate prior research, experimental data, and scientific knowledge to identify promising experimental pathways.
We are developing agents that support researchers deciding which experiment to run, which experimental conditions to test, and how to interpret results.
At PNNL, researchers have been using large language models to integrate years of experimental data with published research to help conceive new experimental approaches and decide which experiments to do next.
The leap is connecting that reasoning to experimental execution in the physical world.
The ARCADIA platform: uniting AI and robotics with human direction
Meet ARCADIA: Agentic Robotics and Curated AI Data for Intelligent Autonomy. It unites robotic lab systems, AI agents, and an AI-ready data environment so experiments can be planned, executed, analyzed, and refined in an adaptive experimental loop, with scientists able to guide, review, or intervene whenever their expertise is needed.
The ARCADIA interface allows natural language programming of robot movements in the laboratory, as shown in this demo. (Video GIF courtesy of Nick Cramer | Pacific Northwest National Laboratory)
The ARCADIA interface allows natural language programming of robot movements in the laboratory, as shown in this demo. (Video GIF courtesy of Nick Cramer | Pacific Northwest National Laboratory)
Robotic systems can execute experimental workflows consistently and precisely, reducing variability and helping make results reproducible. People have used robots for decades, but what’s changed is how quickly we can now instruct them. Not long ago, programming robots for lab tasks meant painstakingly writing code for every movement, every time, work that could take months or even years to integrate across a full system.
Now researchers can use natural language to tell ARCADIA where to find materials for experiments, how to mix them, where to move them, how long to wait between steps, and when to analyze the results.
Just as important: PNNL is building the safety culture needed for predictable, reliable robotic systems that can operate safely alongside people.
The intelligence behind the experiment
What can’t be as easily seen are the AI agents inside ARCADIA that are working alongside researchers. SciLink is one of them, and it coordinates multi-agent workflows, from idea generation and experimental design to simulation, analysis, and decision support.
Think of it as a coordinated team that can draft a plan, run the workflow, watch results come in, and recommend what to do next. In other words, autonomy isn’t replacing scientific judgment; it’s amplifying it, strengthening it.
But does it work?
The challenge: isolate a valuable metal from industrial waste streams, which are a complex “chemical soup.”
Here’s how the workflow played out. Scientists started by describing what’s in the waste stream. Then AI agents dug into published literature and rapidly generated an actionable experimental recipe for separating the target material.
Next, the liquid-handling robot ran 96 experiments exactly as specified. Humans helped prep samples for the final chemical analysis in these early runs—but the key is what happened afterward.
SciLink didn’t just optimize chemistry; it also delivered an early read on whether a method was economically feasible and scalable, using low-cost commodity chemicals already common in industry.
All in a matter of days.
Scaling across PNNL
We are building on this success to scale autonomous experimentation across the lab.
That means making an investment.
PNNL is backing the Center for Robotics and Autonomy with sustained internal funding to build the platforms, infrastructure, and expertise needed to scale.
Science researchers can trust
We understand that transforming scientific experimentation can’t be done in isolation. And we know the best solutions happen when we combine the best ideas with the best technology.
Our industry partners are building out enterprise-scale solutions:
Microsoft Discovery
The goal of PNNL's partnership with Microsoft Discovery is to bring the power of Microsoft’s AI and high-performance computing capabilities to the science problems being solved at PNNL. The collaboration is using AI to dramatically accelerate the discovery of new energy storage materials through novel experimentation design and by shrinking analysis of massive volumes of scientific data from years to weeks. In biosystems design, Microsoft Discovery is plugging directly into PNNL’s laboratory automation infrastructure to launch self-driving scientific workflows that autonomously design, run, and fine-tune biological experiments in real time.
Amazon Web Services
PNNL has partnered with Amazon Web Services (AWS) to develop the fundamental data infrastructure needed for autonomous science and experimentation. Through this collaboration, PNNL deployed a data lakehouse that contains AI-ready data from instruments across the Laboratory, accessible to AI agents and autonomous systems. This work represents a key step in enabling autonomous scientific discovery at PNNL.
Ginkgo Bioworks
Ginkgo Bioworks and the Environmental Molecular Sciences Laboratory (EMSL) at PNNL are working together to advance two complementary microbial research capabilities: the already-built Anaerobic Microbial Phenotyping Platform and the Microbial Molecular Phenotyping Capability (M2PC) now being designed. The anaerobic platform automates traditionally slow, manual experiments with oxygen-sensitive microbes, enabling more scalable, reproducible studies of how these organisms grow, consume resources, and produce compounds. Building on that collaboration, Ginkgo is also working with PNNL to design M2PC as a broader, high-throughput microbial phenotyping capability that will generate standardized, AI-ready datasets for discovery, modeling, and community use. Together, these efforts combine PNNL and EMSL’s expertise with Ginkgo’s capabilities in modular laboratory automation and bioengineering to accelerate research for energy, fuels, waste-to-value processes, and the U.S. bioeconomy.
Oregon State University
PNNL and Oregon State University (OSU) have worked together for years to advance research, develop scientific talent, and address complex challenges. The two institutions recently signed a memorandum of understanding to strengthen and expand their long-standing partnership. PNNL and OSU are advancing autonomous scientific laboratories by integrating robotics, AI, computer vision, simulation, human-robot interaction, and intelligent automation to create reusable technologies that enable safer, more capable, and more collaborative autonomous experimentation.
We are integrating validation, verification, transparency, and reproducibility directly into autonomous experimentation. Precise, repeatable execution makes experiments more reproduceable, and reproducibility builds confidence that the results are scientifically defensible.
Because the goal isn’t just more experiments. It’s enabling new scientific discovery in chemistry, materials, biology, energy, and national security at the speed of nature.
That's how PNNL is advancing trustworthy autonomous science: autonomous experimentation at speed and scale that produces reproducible results scientists can trust.
Explore how the Center for Robotics and Autonomy is Connecting AI, robotics, and data architecture to make scientific discovery at the speed of nature.
Contact us or follow us on: Facebook / YouTube / Instagram / LinkedIn / X
|
RESEARCH |
PARTNER WITH PNNL |
|
|---|---|---|
Pacific Northwest National Laboratory (PNNL) is managed and operated by Battelle for the Department of Energy
Built with Shorthand