Tech
'Knowledge Absorbing' Robots: Should The Human Race Be Concerned?
Jarold
First Posted: Oct 15, 2016 04:30 AM EDT
Brown University professor Stefanie Tellex came up with the theory of what if robots could learn from other robots in her study "Million Object Challenge". Most jobs humans would like robots to do such as factory assembly, aiding soldiers on the front lines, and assisting bedridden patients are still currently not possible due to the robot's limited recognition of common objects.
As for us humans, we find no trouble doing simple tasks like picking simple objects or folding our socks because we've gone through a process called "big data collection" during our childhood. For AI to do the same types of routine tasks, they would also need access to heaps of data on how to handle and manipulate even the most common object. But where does that data come from? Usually it comes from painstaking software programming where engineers feed all information into the robot's software. But Tellex's theorized that these data could ideally be fetched from other robots.
The goal of Tellex's theory, "Million Object Challenge," is for research robots around the globe to learn common daily tasks and objects then upload those learned information to their data cloud. The information uploaded by one robot to the cloud can then be accessed and analyzed by other robots across the globe.
Research robots that use a standard framework for programming such as ROS, makes projects like these possible. Once a single machine learns a specific task, it can then pass the gathered information on to others-and other machines can send back information containing a much refined instructions given to subsequent machines. Tellex stated that the data created by the AI upon learning a given task can be compressed to just about five to ten megabytes, small enough to easily traverse across the global internet.
The University professor Tellex, was also an early partner in other robotics projects like the RoboBrain, which demonstrated how AI could learn from another's experience through data collection and sharing. Brain of Things CEO, Saxena stated that such progress in robotics and AI seem incremental now, but in the coming years, we can expect to see "an explosion in the ability of robots," as more researchers continuously contribute to these type of research.
See Now:
NASA's Juno Spacecraft's Rendezvous With Jupiter's Mammoth Cyclone
©2024 ScienceWorldReport.com All rights reserved. Do not reproduce without permission. The window to the world of science news.
More on SCIENCEwr
First Posted: Oct 15, 2016 04:30 AM EDT
Brown University professor Stefanie Tellex came up with the theory of what if robots could learn from other robots in her study "Million Object Challenge". Most jobs humans would like robots to do such as factory assembly, aiding soldiers on the front lines, and assisting bedridden patients are still currently not possible due to the robot's limited recognition of common objects.
As for us humans, we find no trouble doing simple tasks like picking simple objects or folding our socks because we've gone through a process called "big data collection" during our childhood. For AI to do the same types of routine tasks, they would also need access to heaps of data on how to handle and manipulate even the most common object. But where does that data come from? Usually it comes from painstaking software programming where engineers feed all information into the robot's software. But Tellex's theorized that these data could ideally be fetched from other robots.
The goal of Tellex's theory, "Million Object Challenge," is for research robots around the globe to learn common daily tasks and objects then upload those learned information to their data cloud. The information uploaded by one robot to the cloud can then be accessed and analyzed by other robots across the globe.
Research robots that use a standard framework for programming such as ROS, makes projects like these possible. Once a single machine learns a specific task, it can then pass the gathered information on to others-and other machines can send back information containing a much refined instructions given to subsequent machines. Tellex stated that the data created by the AI upon learning a given task can be compressed to just about five to ten megabytes, small enough to easily traverse across the global internet.
The University professor Tellex, was also an early partner in other robotics projects like the RoboBrain, which demonstrated how AI could learn from another's experience through data collection and sharing. Brain of Things CEO, Saxena stated that such progress in robotics and AI seem incremental now, but in the coming years, we can expect to see "an explosion in the ability of robots," as more researchers continuously contribute to these type of research.
See Now: NASA's Juno Spacecraft's Rendezvous With Jupiter's Mammoth Cyclone