News

Intel Labs improves robot learning with neuromorphic computing

Neuromorphic computing has become one of the most important pillars in the world of AI and deep learningand it is also one of the fundamental research and development fronts for Intel Labs. As our regular readers will remember, the chip giant has made significant achievements in this field that mimics the structure and functioning of the human brain. Loihi 2 was one of the most recent, and one of the most interesting, but it was not the only one.

Intel Labs has just introduced a new model for object learning based on neural networks, whose development has been possible thanks to the collaboration of the Italian Institute of Technology and the Technical University of Munich. This new model is especially aimed at the development of robotic assistants that interact with environments without restrictions, among which we can highlight, for example, logistics, health care or care for the elderly.

It goes without saying that this new model represents an important step forward to improve, with maximum efficiency, the capabilities of future robots dedicated to assistance and manufacturing. The use of neuromorphic computing occurs through new interactive online object learning methodswhich allows the robots to obtain the base they need to be able to continue learning to recognize new objects once they have been deployed. In other words, robots will never stop learning.

Intel Labs, together with the Italian Institute of Technology and the Technical University of Munich, successfully demonstrated this interactive learning model using Loihi chip, and the result was a complete success, since they managed to complete the learning of a new instance with a speed and precision similar to those of conventional methods under CPU, but with a consumption 175 times lower. Impressive, no doubt. To perform this demonstration, the researchers implemented a spiking neural network architecture on the Loihi chip that localized learning to a single layer of plastic synapses, and accounted for different views of objects using a process of “recruitment of neurons.” » according to your needs (“on demand”). This meant that, in addition, the learning process will be developed autonomously while interacting with the user.

Yulia Sandamirskaya, Director of Robotics Research at the Intel Neuromorphic Informatics Laboratory, has published very interesting article on this topic that you can read following this link, and has commented:

“When a human discovers a new object, they take a look at it, turn it over, ask what it is, and then is able to recognize it again in all types of environments and conditions instantly. Our goal is to apply similar capabilities to future robots working in interactive environments, allowing them to adapt to the unforeseen and work more naturally alongside humans. Our results with Loihi reinforce the value of neuromorphic computing for the future of robotics.”

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *