It paves the way for the creation of robots that will be able to communicate with their environment, without necessarily having full programming
An advanced robot that can perform various activities, such as cooking, with just voice commands has been created by researchers in the US.
Although it still doesn't understand everything it hears, it is a remarkable achievement in the field of so-called machine learning, paving the way for the creation of robots that will be able to communicate with their environment, without necessarily having full programming.
Researchers at Cornell University in New York, led by assistant professor of computer science Asutos Saxena, have "trained" their robot to understand instructions in the natural language of various people, to perceive what is missing from these instructions, and to have the intelligence to fill in the gaps on its own.
For example, if someone asks the robot to cook something in the kitchen and explains a few steps of the process (e.g. "fill a pot, put it on the "eye", heat the water and, when it boils, throw in the spaghetti"), the robot not only has to understand all of this, but it also has to remember to turn on the kitchen's electrical switch (something that was not explicitly included in the voice instructions).
Thanks to the special software that has been developed and the previous "teaching" of the robot through video simulations (e.g. of the cooking process), the robot, when it hears the word "pot", immediately understands which kitchen utensil it should use and where to find it (thanks to its 3D camera, it finds what it needs on the shelves).
As it learns more and more, all you have to do is tell it to "heat water" and it will perform all the steps in sequence to do it. Even if the robot is moved to another kitchen, where there is no pot, it has learned to look for another utensil on its own in order to heat the water.
Robots around the world are becoming increasingly intelligent, but almost everything depends on the step-by-step instructions that programmers have embedded in their “minds,” that is, in their software. However, since not all humans are programmers and, also, given that new needs constantly arise in practice that a programmer cannot always have foreseen in advance, it is important that robots can be taught on the spot and simply by words. Furthermore, because human language is often ambiguous and humans forget to explain everything clearly, robots must be able to fill in what no one has told them.
Of course, Cornell's robot doesn't "get it" every time. Its success rate (e.g., in making a dessert based on voice instructions) reaches 64%, three to four times better than any previous method. "There is certainly still room for improvement," he stressed.
Source: protothema.gr
