"Robots are clumsy when handling objects" | Technology

Robots are going to provoke, without a doubt, a revolution in our lives. They are already here to perform tasks with which to improve them. But it will not be possible if you are not able to use your hands or grippers - Robotic clamps - for this. Robotics manipulation is booming and more and more experts are working on it. It is the case of Alberto Rodríguez (Lleida, 1981), a Spaniard who is a professor in Massachusetts Institute of Technologys (MIT) and has worked for more than 10 years in the design of robotic hands and algorithms for the manipulation of objects.

"These investigations have a great social and economic impact. For example, from the point of view of the logistics or in relation to the robots assistants of the home ", explains Rodríguez, who has been one of the participants in the IROS 2018 -The largest robot congress in the world-. This Catalan had his first contact with robotic manipulation when he left in 2006 to the Carnegie Mellon University in the United States with a grant from La Caixa. Today, about 10 years later, he assures without any doubt that "robots are still clumsy when it comes to manipulating objects": "They do it blindly without knowing what they are doing".

The robotic options are still far behind what the human hand is capable of, according to the MIT professor. "What makes the human being a master manipulator is the connection between the hand and the brain, which can precisely control the movements and the tactile information that comes from the fingers," he says.

A robotic expert explains the skills of the robot that precedes it and that has 'grippers' or robotic grippers.
A robotic expert explains the skills of the robot that precedes it and that has 'grippers' or robotic grippers. AFP

Even so, there are concrete examples for which robots offer good solutions. "If in an assembly line you have to assemble a million telephones in a week, you can design a clamp that is responsible for taking a specific piece and put it in a specific place," he says. But the difference with a hand of a person is in the flexibility that the human has to manipulate an object without having done it before.


However, the robotic hand also offers certain advantages: "It does not get tired and can be designed to have more strength than a human or to make it bigger or smaller". In addition, once you design one that works, "you can do as many as you want because it's very easy to replicate both the physical and the smart part."

The robotic hands do not necessarily imitate the human ones, but, as Rodríguez points out, they do take inspiration. "Although our hand is complex, we often use simple strategies. For example, if you have an object like a pen in your hand and you want to know where it is exactly, you hit it on the table. That makes you place it and place it correctly. "

A robot orders objects with his articulated arm.
A robot orders objects with his articulated arm. AFP

Some companies like Amazon, in their attempt to fully automate the processes in their warehouses, organize events to encourage research in this sector. The technological giant performs Amazon picking challengue, a competition in which Rodriguez has participated from 2015 to 2017 with an MIT team. It consists of getting the best automatic system capable of ordering objects on shelves alone. While in 2015 they knew all the products they could show them, in 2017 the robot did not know in advance what objects could be: "The only thing they gave him was some image and the weight and volume of them. I had to pick them up and put them in a box emulating what would happen in a warehouse if I had to find an object without having seen it before. "

"In a warehouse with millions of different objects, how do you design hands that are capable of taking, for example, a book from a bookshelf? And what algorithms are necessary for a robot to see those objects and understand their shape and composition? ", Reflects Rodriguez. To answer these questions, several lines of research are open. "One is that of touch sensors. This problem can not be solved unless we have a sensory capacity in the tweezers to be able to know when and what we are touching, "says the MIT professor.

The investigations in this sector during the next years "will be defined by the problems of the society". For example, Rodríguez believes that climate change will complicate food production. "Technologies that help produce food in a more automated way will be very useful." Another problem will be the aging of the population, so the professor says that technologies will be developed to help people stay at home for a few more years: "A robot that helps cook, get out of bed, get into the shower or pick up the TV remote from the floor ". "They are problems that will come and we can not escape them. But manipulation robotics comes with promises for them, "Rodriguez concludes.

Some robotic hands are trained with mathematical models

A robot places material on a shelf during the World Robot Summit in Tokyo.
A robot places material on a shelf during the World Robot Summit in Tokyo. AFP

There are two ways to train a robotic hand. One is to use mathematical models that allow searching all possible interactions in space to see which movement the hand should make.

The other is based on learning and consists in giving a robot the ability to explore which movements are useful or not to pick up objects and manipulate them. "The first is based on machine learning and artificial intelligence and the second, in tools of memorization or control, "says Rodríguez.


Source link