February 13, 2014      

It?s been a long time coming, but Cambridge, Mass.-based startup Neurala has finally been awarded a patent for its ?brain-like? intelligence for robots.

The technology, which Neurala began developing in 2004, attempts to mirror the human brain?s continuous feedback process to enable robots to process multiple inputs and commands simultaneously, making them smarter and able to quickly adapt to their surroundings.

The company first started working on the deep-learning technology in 2004, filing patent 8,648,867 (pdf), the company?s first, on Sept. 25, 2006.

Neurala?s first two products with the patented technology will be flying robots. The first product is due out by the end of 2014.

Startup Profile: Deep Learning Tech from Neurala Competes with Google

?Our invention makes it possible for robots and other devices to use artificial intelligence in situations in which execution time is critical,? Massimiliano Versace, CEO and co-founder of Neurala, said of the technology that runs on graphic processing units (GPUs). ?It will be fundamental for our effort to build brains for robots that interact with the world and with humans in real-time.?

Roger Matus, Neurala?s VP of products and markets, said the patent ensures the company will be able to deliver on its promises to customers, which currently include NASA and the U.S. Air Force, which together have paid Neurala $1.7 million. Neurala is currently developing software for the Mars Rover, due by July 2015, that allows it to operate autonomously.

Inside the ?Brain Technology

Matus said Neurala is trying to mimic the entire way the human body works, stressing the importance GPUs will play in the future of robotics and AI.

?People have been trying to do faster and faster processes to mimic the human brain. They create larger computers that do more instructions per second, but no matter how fast they get, they aren?t close to what we need,? said Matus. ?They can?t do what a mouse can do, never mind the human brain.?

GPUs were originally designed to run computer games and 3D graphics. But since Artificial Neural Networks (ANNS) and AI are drastically faster on GPUs, ?anyone who wants to make an autonomous robot has experimented in this area.?

?What we noticed early on is that the human brain is very slow. But it?s massively parallel and does a lot of things at the same time,? Matus said. ?The GPU is similar to that for computers and is a fundamentally better way to drive robots.

?To give an example, if all you?re doing is dealing with vision, to do it the old way you?re going dot by dot on the camera to analyze every pixel. With GPUs, you can look at the entire image at once and do the processing much more effectively.?

Applications for the ?Brain? Technology

Matus wouldn?t divulge into specifics about the first two products with the technology, but the company sees “strong use for collision avoidance. Being able to see things in real time and avoid moving objects will be very important.?

Matus also said the work being done with NASA and the Air Force involves GPU computing, but neither endeavor will be completed by the end of 2014.

Neurala won the contract from the U.S. government to develop software for the Mars Rover by demonstrating the possibilities of what could be accomplished with GPU computing. One of the main issues with exploring Mars with robots, Matus said, is it takes between 15-45 minutes for a signal to get to and from Mars.

?The speed rate, bandwidth, is about the same as a dial-up line,? he said. ?You can?t drive a robot on Mars, the robot has to make individual decisions. And you can?t send someone out there to change the battery, so using lasers to measure distance is totally out because it uses too much battery.

?We?re applying GPU technology to the Mars Rover so it can look around its environment and figure out exactly what it sees, for example, ?here?s a rock, a nice rock, this is a really shiny rock.??

Matus also said Neurala will focus on self-localizing that can determine where they are on a map and figure out how to get to where they need to go.