PARIS: A new research from the University of Maryland involving the use of visual stimulus to train robots has resulted in new culinary cooking bots being taught by videos on YouTube.
The project – funded by DARPA as a part of their Mathematics of Sensing, Exploitation and Execution program – involves providing robots with sensors and technology advanced enough to learn from and react to objects in their environment. The robots observed the video, examined the objects used in them and identified them according to their use, and then used that knowledge to objects presented to them in the three-dimensional world.
Robots involved in the program were able to identify and select specific objects that had been presented to them only on video and then manipulate them in the same ways they were seen being used in the videos they watched. All this occurred without any additional programming or input from human engineers, proving that these robots could both perceive objects and understand their use – a step above simply recognizing and then identifying the objects for what they are, according to DARPA manager Reza Ghanadan.
Additional steps are now being undertaken by researchers to see if these robots can not just retain this knowledge as to how the objects they identified work but if other robots can be taught by them, allowing this knowledge to be passed on. Mastery of this hurdle could reduce the amount of time it takes for new robots to learn these skills by a significant margin; Ghanadan said that no longer would it take time and expense to program robots to undertake these tasks, as robots capable of learning from others could simply share knowledge. It’s an important step towards the development of technologies that would benefit the military repair and logistics areas, the DARPA manager added.
