Deep Learning for Robots

Artificial Intelligence, Neocortex, Robotics / 08.08.20140 comments

Check out this interesting article on advances in deep learning for robots: Robots Helped Inspire Deep Learning and Might Become Its Killer App?

It’s worth comparing/contrasting between deep learning techniques and Neocortex. Neocortex is Universal Robotics’ patented machine learning platform, based on a seven-year development effort between NASA and Vanderbilt University. Even though the technology is currently employed on Robonaut 2 on the International Space Station, Universal’s focus with Neocortex is material handling tasks. By learning to recognize new objects or recognize previously seen objects that have changed, Neocortex machine intelligence brings flexibility to material handling automation.

Neocortex certainly may include deep learning (hierarchal neural networks), but it is far more. Let’s take a quick look.

Similarities. Both Neocortex and deep learning can do “unsupervised learning” that doesn’t require data to be pre-categorized. The more and the better data available to train their artificial intelligence, the smarter both their robots get over time. In both cases, the robots teach themselves which features are important or pertinent in the raw data they are ingesting.

Differences. Whereas deep learning may need 50,000 images to understand an object, Neocortex only needs 5 – 50 images. Neocortex is not inextricably linked to deep neural networks the way deep learning is. Neocortex is also a patented process for cybernetic activities. Whereas deep learning is a discrete process involving patterns of categories, Neocortex is a continuous process that draws its detailed granularity from ongoing interactions with the physical environment.

As a result, Neocortex machine intelligence can very quickly identify objects from machine vision it hasn’t seen before and inform the robot on the next steps. Based on this functionality, Universal experts (80% with masters or doctorate level training in robotics, machine vision and machine intelligence) have built robust industrial flexible material handling applications that have been tested at Fortune 500 companies.

Hob Wubbena

Tags: , , ,

Leave a Reply

Your email address will not be published.

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>