Startup Patents ‘Whole Brain’ AI Approach
Among the criticisms of current artificial intelligence systems is their inability to handle more than one task such as spatial navigation or object recognition at a time.
“Integrating advanced behaviors into a whole brain system for robots and drones represents the future of artificial intelligence” argues Massimiliano Versace, co-founder and CEO of Neurala Inc. The software startup with ties to NASA and the U.S. Air Force announced this week it has received a U.S. patent for its deep learning neural networks platform that, the company claims, integrates “multiple brain areas.”
The Boston-based startup’s “whole brain” approach seeks to mimic the human brain’s ability to integrate sights, sounds and other senses to make a decision. The patent award covers Neurala’s method for implementing autonomous robotic control. The scheme relies on GPUs and standard processors along with sensors. The mix of processors and sensors depends on the application, the company said.
Versace cited his earlier patent filings related to GPU processing for deep learning and neural networks as evidence the company viewed graphics processor as the “next big thing” in AI development. Noting that major AI players such as Google (NASDAQ: GOOGL) and Microsoft (NASDAQ: MSFT) are just now embracing GPUs to accelerate their AI platforms, Versace asserted that Neurala’s approach goes further.
“Now the rest of the AI world is focused on one specific, small set of neural networks for object recognition, while Neurala is working on the ‘whole brain’ systems that are necessary for intelligent and autonomous behavior,” he argued.
Neurala’s AI research stems from its technology originally developed for the Air Force along and NASA. Versace is credited with improving the autonomous operations of NASA’s Mars rover Curiosity. “We learned from our work with NASA how to execute complex tasks with whole brains using a lean sensor/compute package,” Versace said. The startup’s approach uses relatively inexpensive cameras and its patented software to perform complex tasks. The “native integration” scheme “enables different senses and modules to complement each other’s deficiencies and shortcomings,” the company said in a statement.
The goal now is to commercialize the patented AI technology by simplifying previously complex and expensive processes. Leveraging GPU technologies, the approach could for example eliminate the need for costly radar sensors traditionally used on autonomous vehicles. Applications include self-driving cars, autonomous robots and drones.
To that end, the startup announced a $14 million funding round in January that will be used to commercialize its AI technology.