Anyone can download an autonomous “research robot” from the Air Force Research Lab.
In the struggle to triumph over America’s adversaries by surpassing them in innovation – a struggle that has all the hallmarks of a Cold War despite President Biden’s claims to the contrary – it is vital to increase the speed at which physical laboratory experiments can be performed and repeated.
Air Force Research Laboratory (AFRL) scientist Dr. Benji Maruyama reminds his peers and the public that “Research is an extremely slow process. It takes a lot of time to be in a lab and do experiments. With the aim of reducing this time and speeding up laboratory experiments, Dr Maruyama and his colleagues at AFRL have developed a stand-alone “research robot” operating system that anyone can download for free.
Outside of the largest and most sophisticated laboratories, the experimental process is largely unchanged from the 19th century. The famous Japanese scientist, Professor Hiroaki Kitano, compares contemporary laboratory experimentation to pre-industrial revolution agrarian processes, physically and cognitively manual and extremely slow.
Earlier this year, Dr Kitano published the Turing Nobel Prize Challenge, a call to the global science and technology community to see if humans can build an “artificial intelligence scientist” who can perform Nobel Prize-worthy research by 2050. Kitano’s call for the application of AI to laboratory research calls on AFRL’s Maruyama.
“I think that’s a great way to sum it up,” says Maruyama, ARES OS team leader. “Today’s farmers use tractors to do manual labor, but they also use AI and data systems to improve their [productivity]. If you go to a chemistry lab, we always mix things up in beakers and heat them on a hot plate. Researchers do these things manually. Why is this still a thing?
Maruyama and his AFRL cohorts took on a related challenge ten years ago: designing a tool that could accelerate the pace of discovery through laboratory experiments. In Maruyama’s case, the slow pace of discovery has been a hindrance in his work on carbon nanotubes, cylindrical molecules that can be used to develop ultra-strong, lightweight materials that possess highly conductive electrical and thermal properties.
The result is the Autonomous Research System, an open source software operating system (ARES OS). It is available for download – free of charge – from AFRL.
Using the artificial intelligence and autonomy algorithms built into ARES OS, researchers can transform experiments into autonomous “research robots” capable of directing and conducting their own research.
Dr Maruyama first sought to apply automation to speed up his own materials research, by growing carbon nanotubes in the lab. He and his colleagues quickly recognized that the development of a stand-alone tool could also provide information to shape and self-direct monitoring experiences.
“Wouldn’t it be great if you could experience more like playing Battleship?” thought Maruyama. “You can start looking for a ship, but once you hit something you don’t keep [searching] randomly. You say, ‘Something happened there, so let me go up to that ship and try to find something else there.’ You choose your next move based on your previous results.
This is essentially what ARES does, providing a basis for designing experiments, answering research questions, running experiments, and analyzing the results, which in turn helps design a new set of experiments. In the past, researchers wrote similar automation code just to do a job, says Maruyama, a unique experimental process. Such a code had limited adaptability to new experiences. This is where ARES differs.
ARES OS includes a kernel or universal module with plug-ins that connect to experimental tools or scientific instruments such as electron microscopes. “A big part of the challenge was creating the software to communicate with the hardware,” says Maruyama.
Researchers can write their own software modules to link their specific lab hardware (air flow sensors, chemical sensors, lasers, catalysts, micrometers, flow controllers, even virtual simulators) via ARES OS plug-ins to their own research experiences.
Once this link is established, the software can autonomously design experiments and perform experiments quickly and accurately, learning how to achieve the research objective using AI.
Once he and other AFRL researchers started using ARES, they went from about one physical / material experience per day to 100 experiences. The benefits of speeding up experimentation by 100 are obvious, but it wasn’t the biggest gain, according to Maruyama. “It was really the ability to use artificial intelligence and autonomy algorithms to design the next experiment, to target the next research goal.”
AFRL currently has four different fully autonomous ARES robots operating for various programs. ARES OS is evaluated by ten partners, including Carnegie-Mellon University, University of Missouri, and several business groups. The tool is still in its infancy, says Maruyama. A Boston University professor quantified ARES versus traditional experimentation campaign design in his own research and achieved a 60-fold increase in speed.
“Our goal now is to make it easier for everyone to build their own autonomous research robots,” says Maruyama. AFRL also shows that ARES has applications beyond basic research in areas such as additive manufacturing.
“The secret that people don’t talk about with additive manufacturing,” says Maruyama, “is that it takes a lot of tweaking for these printers to work. If someone hands you brand new ink [i.e. print material] that you are not familiar with, it may take a week or two to figure out how to print with it.
ARES can autonomously speed up the calibration process for new hardware media and AFRL already has a printer running the platform it calls “AM ARES”. The lab is currently working on a cheap $ 300 AM ARES printer for students, schools, etc.
There are similar autonomous research robots, including the University of Toronto’s “Chem OS” and the University of Glasgow’s “Ϗ DL”, but they are not universally accessible.
The importance of ARES lies in the fact that it effectively lowers the barrier for researchers by giving them an operating system for which they only have to write software modules adapted to the particular equipment they are working on. use. As positive as it sounds, even though it may help scientists innovate faster than their rivals, it is reasonable to wonder if making it widely available for download does not simply give American adversaries an advantage?
“It’s an enigma,” admits Maruyama. “If you keep it to yourself you can have an advantage, but only if you are able to go faster than your opponent. With partners, we can do our [tool] go faster than if we kept it to ourselves.
The risk of disseminating such open source IP in the wild seems to be directly related to the determination of scientific groups, governments, strategic research institutions and non-state actors to use it. AFRL apparently thinks it’s a risk to take.
“A lot of people understand this and think it’s late,” Maruyama observes. “As we collectively start to create more tools like this, it leans on itself, making research faster and faster. “