January 30, 2014      

Company Name: Neurala
Founder: Massimiliano (Max) Versace, CEO; Heather Ames, Vice President; and Anatoli Gorchetchnikov, CTO
Principals: Roger Marus, VP of Business Devlopment; Gennady ?Genna? Livitz, Lead Engineer
Contact: Roger Matus, VP of Business Development

Why it was founded: Neurala was incorporated a few months after Massimilliano Versace and his labmates drafted a business plan based on research they conducted as graduate students at Boston University.

Product: A consumer iPad app for controlling off-the-shelf toy robots; enterprise software for autonomous perception and navigation in telepresence robots.
Available: Consumer app out in Spring 2014; Enterprise software expected late summer 2014.
What it is: Adaptive learning software for multiple robotics and AI applications
Market niche: Autonomous robotics of all sorts

Funding: More than $1.5 million in funding from NASA and the U.S. Air Force. Recently reported to be seeking a $2 million round, but company declined to disclose its current funding status.
Industry Partnerships: Signed Letters of Intent with iRobot and Teledyne (in the context of the NASA grant). Working with Romotive, Orbotix, and another company on app development.

RBR’s Take: The majority of robotic technologies are still in their infancy, particularly when it comes to autonomous operation. Cambridge-based Neurala is hoping to bring the field into its toddler years with a bio-inspired approach to machine learning. The company is introducing an iPad controller app for use with toy robots this spring, and it expects to introduce an autonomous visual perception and navigation system for enterprise users by summer 2014.

Today, most ?autonomous? technologies require operators to provide robots with detailed information about their environment, required tasks, and more before they can operate independently. Humans, meanwhile, readily apply information from past experiences to new, unfamiliar situations.

How to replicate this ability in a computerized system is a puzzle that researchers across dozens of fields have attempted to solve for decades. But in recent years, deep learning, as it is called, has begun to make real progress by borrowing the architecture of the brain.

Neurala CEO Max Versace

The brain, explains Neurala?s CEO, Max Versace, is essentially a compact parallel compute cluster. The millions of neurons in the brain act as low-power, low-quality processors that all run in parallel. On-board sensors?our eyes or the bones of the inner ear, for example?provide the brain with massive amounts of raw data that can be processed.

Google, with its access to massive quantities of both computing power and raw data (from both digital and, increasingly, physical sources), is among companies that are exploring these methods. It snapped up another startup co-founded by Dr. Geoffrey Hinton, a leading research in the field, in March 2013, and hired noted technologist Ray Kurzweil in January 2013. (Shortly thereafter, Google?s Chinese competitor, Baidu, also opened a deep learning research center in Mountain View, Calif.) The ?Google Brain? has shown some promising results.

Versace says Neurala?s technology attempts to go beyond Google?s methodology by taking the brain analogy one step further. While Google and others separate training modes from performance modes, human learning is a process of continuous feedback. ?The brain integrates multiple functions from the very beginning. You do not have the segmentation algorithm separate from the recognition algorithm,? Versace says. ?These things are very tightly integrated.?

Neurala attempts to mirror the brain?s continuous feedback process. ?You are both learning and performing in our system,? Versace explains. The approach has promise: NASA awarded the fledgling company both a Phase I and a Phase II Small Business Technology Transfer
(STTR) contract, with an eye toward using the technology for training autonomous exploratory robots (such as the Mars Rover) to navigate unpredictable planetary surfaces.

The U.S. Air Force has also contracted with the company for visual data analysis. While Neurala wasn?t permitted to speak about the details of the contract, the USAF is using Neurala?s parallel processing system for visual data on a massive GPU compute cluster. Technology advances have increased the amount of data available to USAF analysts?but the number of analysts on staff hasn?t (and can?t) keep up. Neurala?s system pre-screens the visual information, flagging important or anomalous images for review by human analysts.

For its first commercial applications, Neurala is continuing to focus on basic computer vision and navigation tasks. Instead of military data or virtual regolith, though, Neurala?s initial product?an iPad app for controlling toy robots?will use data from on-board cameras and accelerometers. Users will be able to train any number of ROS-compatible toys (approximately 300,000 of which are in the wild today) and then give them directions using a simple ?point and click? interface, instead of guiding them with joystick-like controls.

Neurala?s enterprise product?a souped-up version of the iPad application?will integrate with telepresence and other enterprise robotics where inexpensive autonomous navigation is required. Together, the two markets could help Neurala amass a larger data set for improving its algorithm?a necessity if the company is to compete with the likes of Google.

That?s why cloud processing is a critical piece of the company?s business model. ?There?s no reason why you shouldn?t combine learning from different robots together,? Versace says. ?You can access the wisdom of the crowd as you have millions of robots deployed to increase an individual robot?s knowledge.?

While the company wouldn?t disclose information about its previously reported efforts to raise a $2 million Series A round, it seems likely that an announcement is coming in the near future. Neurala is working with four (probably telepresence) companies to launch its enterprise application. It has also signed letters of intent with iRobot and Teledyne for commercial applications of its work with NASA. If it can deliver on its goals, in the timeline it has set, I think the company will be in a strong position for commercial success.