ROBOTICS CLUSTER
Massachusetts%20Robotics%20Cluster%20Report%20Final
Massachusetts%20Robotics%20Cluster%20Report%20Final
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
AI and machine learning technologies<br />
and techniques are being used<br />
to make robotic systems more<br />
intelligent<br />
been used in support of decision making, object identification, vision processing, speech translation, navigation,<br />
motor control, sensor integration, and other functions, as well as facial and emotion recognition.<br />
6.8.3.2. Deep Learning<br />
Deep learning, a class of machine learning techniques, has received much coverage from the technology and<br />
business media as of late, and for good reason. The technology, which employs multi-level (“deep”) neural<br />
networks to create inferencing systems for pattern and feature detection in large datasets, has proven critical<br />
for commercial applications ranging from speech and music recognition, to industrial process control and<br />
product recommendation. Deep learning techniques continue to improve, as do results.<br />
6.8.3.3. Research and Investment<br />
Deep learning research is robust and ongoing, and investment in the technology is strong. Some of the<br />
world’s leading technology firms are now invested in the technology. Google’s purchase of U.K. AI startup<br />
DeepMind Technologies in 2014 for an estimated US$400 million exemplifies this. So, too, does Facebook’s<br />
launch of its own AI laboratory in 2014, and IBM’s launch of its Watson Group (also in 2014). Both research<br />
groups are focused on deep learning methods (among other research topics). IBM also purchased AlchemyAPI,<br />
a provider of AI-based text analysis and computer vision cloud services, in March 2015, and has other<br />
cognitive computing and deep learning initiatives underway. Microsoft’s Project Adam, the goal of which is to<br />
enable software to visually recognize any object, also exploits deep learning techniques.<br />
6.8.3.4. Cloud and Local<br />
Deep learning methods are computationally intensive, requiring a great deal of computing resources. Software<br />
typically executes on powerful processors running on banks of highly optimized servers. As a result,<br />
devices running applications that use deep learning methods must typically access services residing in the<br />
cloud. Many efforts are currently underway to provide deep learning capabilities without the necessity of<br />
off-device processing in the cloud.<br />
6.8.3.5. Deep Learning and Robotics<br />
AI and machine learning<br />
technologies and techniques are<br />
being used to make robotic systems<br />
more intelligent.<br />
Many of the capabilities enabled by deep learning methods are critical for robotics systems. Examples include<br />
computer vision, facial recognition, and natural language processing. Object recognition is especially<br />
important. A number of techniques have been employed to assist robots in recognizing objects, with deep<br />
learning methods central to many of them. Commercial companies are now coming to market with deep<br />
learning solutions designed specifically for robotics systems.<br />
6.8.4. Robot Operating System (ROS) and Open Source Solutions<br />
Developing and deploying robotics systems are difficult, time-consuming, and error-prone, and as a result,<br />
robotics innovation and commercialization efforts have been slowed or stalled. To address the problem,<br />
ROS, open source system software for robotics, was developed (Quigley et al., 2009). ROS includes software<br />
libraries, tools, and a run-time environment specifically designed to ease the burden of creating advanced<br />
robotics applications.<br />
www.abiresearch.com<br />
THE MASSACHUSETTS <strong>ROBOTICS</strong> <strong>CLUSTER</strong><br />
54