Seeing a white spiky tube cleaning a table or window is striking. But that the same machine – made up of a rolling base, an articulated arm and the equivalent of a robotic head to appear friendly – also picks up trash or opens a door is a technical marvel. And this is what Alphabet (Google’s parent company) is achieving with its Everyday Robots project.

Now there are a hundred of these prototypes, which give an air to the friendly Pixar Wall-e robot, operating on the Google campus in Mountain View (California). They move autonomously and perform different maintenance tasks in buildings. The project is part of X Development, the branch of Alphabet that brings together a jumble of cutting-edge research initiatives. Until recently Everyday Robots hadn’t left its lab environment, so this is a litmus test.

The experiment does not end here. Now Google robots clean the tables on campus, for example. But the goal is much more ambitious. Everyday Robots mission, they point out from X Development, is to develop “a general-purpose robot that can operate autonomously in unstructured environments.” In plain words, “they want them to be used for many different applications,” says Pablo Varona, a professor at the Autonomous University of Madrid specializing in Computer Languages ​​and Systems. “They would be robots that have to move and that have to interact with their environment. And the environment is not defined a priori. It is a technology that advances with the intention of managing uncertainty ”.

In robotics it is easy to build a machine that always has to move in the same environment. All parameters will be known and all circumstances can be foreseen. “The environments where autonomous robots have typically been used are simple and controlled,” says Fran Bellas, a professor at the University of A Coruña who specializes in robotics. “Unknown elements do not appear, the sensory information is stable and not very noisy. The robot operates autonomously but in an unrealistic environment ”.

To operate in unstructured environments, where there are unforeseen obstacles or people, Everyday Robots machines are equipped with an arsenal of sensors. They have several cameras to identify the surrounding elements through object recognition. A LIDAR (laser imaging detection and ranging), similar to the one used in the autonomous cars of Waymo (another Alphabetl subsidiary), allows them to know the distance of objects and their shapes. This is coupled with a great capacity for information processing and refinement of the motor part.

“A very important part is movement control. And what this company seems to be doing are robots that move in an environment safely and with precision ”, points out Varona. And he adds that this is a delicate aspect: “Especially when the robots are already larger than, for example, a vacuum cleaner. When they have a mass and a volume, great care must be taken so that these robots interact safely, that they do not break things and also that there are no accidents with people ”.

Home robots for everything

A Roomba vacuum cleaner crawling around a house cleaning dust is safe. But given the aging of the population, more sophisticated machines are being promoted as assistants in the home from robotic engineering. These are of another depth. “Vacuum cleaners are already smart, but there are other more sophisticated tasks that robots can perform. And there are even more advanced ones that have to do with robots that can interact with people or that can collaborate with people to carry out a job ”, explains Varona.

It may be robots that work in a factory side by side with people. But they can also be robots that assist the elderly at home. For this it is essential that the machines learn to perform different tasks. They also need to adapt to the environment in real time, react to the unexpected. All this is close to the concept of general or strong artificial intelligence, which would consist of a system capable of learning and performing any task, as a human could.

“This company, from what they have published, wants to give greater impetus to everything that has to do with the learning capacity of their robots. They build a generic robot platform and that platform has the ability to learn in different circumstances. It is the same platform, the same sensors, but it has learning mechanisms to adapt to any environment ”, says Varona.

For Fran Bellas, the global approach of Everyday Robots is to make robots that can learn by themselves and from a human to solve different tasks autonomously: “It seeks to solve the problem of lifelong learning in robots, to provide them with intelligent capabilities similar to those of humans ”. However, he points out that there are many unknowns to solve before reaching a general artificial intelligence.

The creation of objectives by the robot autonomously, obtaining an optimal representation of the environment that is not predefined or the use of long-term memories that allow information to be reused and to carry out gradual learning are issues that still need to be solved. “I think they will definitely make progress in home robotics. And we will have better robots in our homes helping with more tasks. But from there to being truly autonomous, able to learn by themselves and operate in truly unstructured environments, that is still missing ”, concludes Bellas.

You can follow EL PAÍS TECNOLOGÍA at Facebook and Twitter or sign up here to receive our newsletter semanal.