WRIGHT-PATTERSON AFB, Ohio --- The Autonomous Research System (ARES) may not look like “Johnny Five,” the famous robot from the 1986 movie “Short Circuit,” but this robot’s ability to integrate robotics, artificial intelligence (AI) and data science is altering materials research in a big way at Air Force Research Laboratory.
The AFRL Materials and Manufacturing Directorate’s ARES can design, conduct and evaluate experimental data without human intervention, revolutionizing the materials research process as it is today.
“To our knowledge, ARES is the first of its kind to link autonomous robotics, artificial intelligence, data science and in situ experimental techniques for materials development,” said Dr. Benji Maruyama, a senior materials research engineer at AFRL’s Functional Materials Division. “Not only does it allow us to be faster and smarter in how we do experiments, we can get to a scientific understanding in a shorter amount of time.”
Traditional materials science research is a time-consuming, human-centered process that takes a certain kind of individual with the knowledge, patience and understanding to design, conduct, analyze and interpret experimental data, and then decide what to do next, Maruyama said. A typical research team may only conduct one or two experiments per day using traditional research routines.
ARES, on the other hand, can complete upward of 100 experiments per day, expediting the materials discovery process.
“We are in the dark ages in the way we do experiments, yet we are inventing such high-tech materials. There is a disconnect between the research process and the high-end technology output,” Maruyama said. “ARES combines the best of hardware experimentation, and modeling and simulation with an AI planner that proposes what to do next. We can get feedback faster.”
ARES’ robotic expertise was tested by Maruyama’s team in the field of carbon nanotube growth, an area of materials research that is traditionally poorly controlled and not very well understood. Carbon nanotubes are extremely valuable in materials science, as they are strong, light weight and have an ability to conduct heat and electricity. Nanotubes can be used in a number of different applications, from airplane wings to lightweight, flexible conductor wires, ballistic materials, computer chips and even for drug delivery.
ARES conducted more than 600 experiments in autonomous mode, with the computer “brain” determining experimental conditions to achieve an objective maximum growth rate for the nanotubes. Human scientists set the objective growth rate, which ARES used to execute the research. Each new experiment performed by the robot resulted in new knowledge, which ARES incorporated into the design of future experiments. As the number of experiments increased, the results became more constant, converging on predicted growth rates for the carbon nanotubes, indicating the AI system learned to grow carbon nanotubes and applied the intelligence with scientific success.
Though ARES is capable of conducting scientific research autonomously and can generate rapid results, the role of the researcher remains extremely important, said Maruyama.
“ARES will not replace humans, but rather the success of ARES depends strongly on the partnership between the human researcher and the robotic system -- a human-machine trust,” he said.
ARES frees the researcher from tedious bench-level experiment activities, such as instrument preparation, monitoring and cleaning, and allows them to undertake the creative, insightful, higher-level thinking that can lead to new discoveries, Maruyama said.
“The beauty is that it makes us more efficient. We are able to be faster and smarter in how we do experiments and can get to a new state of understanding,” he continued.
While ARES proved itself in carbon nanotube growth, autonomous research robots have the potential for use in a number of scientific research areas. Kevin Decker, a software engineer from UES, Inc., is working with the ARES team to program the AI software to allow ARES to be a generic research tool, enabling it to work on other materials research problems.
In the future, the direction of ARES will be to explore chemical and physical phenomena autonomously.
“There are multiple types of machine intelligence that work for different areas and specific problems,” Decker said. “We are working to develop software that incorporates multiple different types of AI that will allow us to determine the most suitable strategy for an experimental problem.”
According to Maruyama, ARES is a “disruptive tool” that is changing the research ecosystem.
“Research is core to what we do in the Air Force. We are trying to cause a disruptive improvement to the process of research wherein not only can we do research 100 times faster, but 100 times smarter and more economically,” he said. “We ask ourselves, ‘How can we reengineer the research process to make research better and more cost effective?’”
As ARES shows, robots and machine intelligence may be the answer.