A research team led by Penn State engineers has received a $12 million grant from the U.S. Army Research Laboratory to develop what's essentially a digital brain, inspired by how insects and small animals make smart decisions using very limited information.
Saptarshi Das, a professor of engineering science and mechanics at Penn State and the project's principal investigator, explained that many of today's most powerful computing systems depend on supercomputing clusters, data centers, or substantial external power to function. That's a problem for edge devices, meaning computer systems operating disconnected from a larger cloud network or power grid, which often struggle to make smart AI-driven decisions on their own. If successful, this project could help AI-powered drones and ground robots make better decisions, particularly in remote or covert settings where information is scarce, he said.
Das explained that drones may operate in remote areas without any network connection, or may deliberately avoid communicating with the cloud to keep their location hidden. He said the team's proposed system aims to give edge devices the onboard processing power needed to make smart decisions in complex environments, without relying on bulky hardware or large amounts of energy.
The project centers on building an intelligent sensing node that would eliminate the need to convert the raw physical data collected by a drone's cameras and sensors into digital information before it can be processed. Joshua Robinson, director of Penn State's Materials Research Institute, said Das and his team are tackling one of the hardest problems in technology today, making machines think more efficiently while using less power, and described the work as directly relevant to keeping military personnel and national security technology safer.
To pull this off, the team is drawing inspiration from an unlikely source: biological systems, including the tiny brains of animals like locusts and owls. Das noted that these animals' brains function as a kind of edge device themselves, since despite being extremely small, they're able to accurately process audio and visual information because they're so effective at filtering out irrelevant input and focusing on what matters.
The sensing node itself will include sensors capable of picking up the broadband electromagnetic waves given off by electronic devices in the surrounding environment. According to Wooram Lee, an associate professor of electrical engineering and the project's co-principal investigator, these signals will be processed using a specialized structure modeled after the cochlea, the part of the human ear that collects physical vibrations and converts them into signals the brain interprets as sound. Lee explained that rather than processing a broadband electromagnetic signal directly in the digital domain, an approach that tends to be power-hungry and limited in how much data it can handle, the team is instead drawing on the cochlea's architecture to pre-process the signal in the analog domain, an approach that can substantially improve both how much data the system can handle and how efficiently it uses power.
Lee noted that the biggest source of power drain in current drone computing systems comes from converting analog information gathered through optics and sensors into digital data. Analog computing, by contrast, uses physical phenomena like electrical current to perform calculations directly, rather than relying on the binary ones and zeros of digital computing. Powering drones using analog computing instead would eliminate that energy-intensive conversion step and significantly cut overall power needs. To achieve this, the team plans to combine two-dimensional materials, particularly graphene, a honeycomb-structured layer of carbon just one atom thick, with traditional silicon semiconductors, using them to route electrical current into an array of memristors, components that can take in an electrical current and amplify its output.
