Nvidia is creating robots that can learn through observation
Having hosted its annual International Conference on Robotics and Automation (ICRA), Nvidia has unveiled its machine learning-powered technology that is being used to create robots that can learn to carry out actions by watching humans.
These actions include stacking coloured boxes in a specific order. Powered by Nvidia’s state of the art Titan X GPUs, the neural networks are able to detect different objects and determine the relationships between them, then generating a program whilst observing human interaction with these objects and repeating the sequence of movements.
The technology was developed at Nvidia’s robotics research lab based in Seattle, headed up by Dieter Fox, with the facilities key aims outlined by Nvidia: “The goal of the lab is to develop the next generation of robots that can robustly manipulate the physical world and interact with people naturally.”
In order to test and train this technology, Nvidia used synthetic data from a simulated environment, allowing the company to train the robots much more quickly.
Moving forward, Nvidia will look to enhance the complexity of the tasks that these robots can complete purely through observational learning, having successfully begun to develop such a revolutionary technology.