< img src = "/uploads/blogs/da/70/ib-FS10edbu1_1cff606d.jpg" Alt = "work is more effective when they see a task on video"/> ~ ~ < P > Cornell University presented a technology capable of changing the approach to training machines. New development using & nbsp; artificial intelligence & nbsp; allows work to perceive and repeat people's actions by watching them only half an hour. The system was named Rhyme and its potential is already actively discussed in the scientific environment.
< p >According to researchers, the main difficulty in working with & nbsp; work & nbsp; there is a difference between human behavior and the algorithmic nature of machines. Where a person moves coordinated and with intuitive accuracy, robot is often limited to a set of rigidly given actions. It was this discrepancy that prevented machines from efficiently taking on video events.
< p > how & nbsp; noted & nbsp; a team of developers, any deviations from standard trajectories previously led to the failure of the task.
< p > Rhyme algorithm has become an attempt to bypass this restriction. The system not only recognizes the human actions in the video, but also compares them with the already known movements that the robot has managed to master earlier. For example, if the device has previously learned to lift a spoon, then when you watch a scene where a person takes a mug, he will be able to use the knowledge gained to reproduce a similar action.
< p > As explained in the research group, a new approach resembles an automatic translation.
< P > In particular, if the robot sees how the cup is taken and put in a sink, it may not know exactly how to perform the whole sequence, rely on the experience of such actions & ndash; Moving other objects or manipulations with similar OB & Rsquo; Ents.
< P > Interestingly, successful training requires a minimum amount of information. Previously, the process of training in robotics took long hours under the control of the operator, now only one video recording for 30 minutes is enough. As scientists have added, the effectiveness of robots trained with Rhyme has increased in half compared to previous methods.