September 26, 2020

This is the robot that changes its own shape according to the task it executes | Technology

This is the robot that changes its own shape according to the task it executes | Technology


Innovations in the world of robotics They happen in a frantic way. Three of the challenges that most fascinate the scientific community today are getting machines to act more and more autonomously, perform multiple tasks and adapt to different environments. A group of American researchers has developed for the first time a prototype robot capable of all three at the same time. This model can fulfill actions in an unknown environment, for example, identify an object, pick it up and take it to a designated area or go up some stairs to deposit an envelope in a box, after configuring its architecture according to the objective that has been requested and the obstacles you encounter. This is shown by three experiments described in a study published on Wednesday in Science Robotics.

"There may be robots capable of climbing stairs well, but they would not know how to pass through a crack," says one of the authors, Jonathan Daudelin from the United States, as an example in teleconference from the United States. Cornell University (New York State). To overcome this problem, for about 25 years, different research groups have been exploring how to use modular defined robots, adds Mark Yim, co-author and professor at the University of Pennsylvania. "These machines are systems formed by repeated robotic elements (modules) that are themselves robots, but can connect with each other to create different structures with different functionalities," explains Tarik Tosun, another coauthor and alumnus of Yim. This feature makes them "very versatile and flexible," he adds.

Modules of the robot.
Modules of the robot.

Daudelin says that, thanks to this ability, modular robots "have the potential to adapt to what they perceive in their environment." The researcher adds that this "works very well for autonomous systems in unknown environments," he adds. These considerations have led him and his colleagues to design an intelligent central system that integrates algorithms and sensors capable of governing the action of cubic robotic modules and modify it as the perception of what surrounds them changes, explains Daudelin. The result is the first prototype that can reconfigure its shape, for example, moving from a vehicle to a machine that collects objects, and adapting to different tasks without human control.

Three tests

In the article published this Wednesday, the authors describe three tests. In the first, the robot was assigned the task of finding two small objects, one pink and one green, placed in a room between different obstacles. The ultimate goal was to pick them up and take them to an area marked in blue, located at another point in the room. In the second, he had to deposit another object in a box placed on top of some stairs. The third experiment provided for the robot to place a sticker on a point of a box marked in pink.

To achieve these objectives, the machine had to explore the environment with its cameras and sensors, design a 3D map of that environment and decide how to configure the modules, each equipped with elements such as wheels and magnets, in the most efficient way to solve the mission designated. The modules can be disconnected from the central axis of the robot and move freely as indicated by the intelligent system installed on it through a wi-fi connection. The action of reconfiguration of a module took about one minute to complete, say the authors of the experiments. In previous tests conducted in other studies, it could take between five and 15 minutes, contrast.

"The algorithm is based on a" library "of behaviors and possible conformations of the robot," explains Daudelin. "Each record represents an action, such as moving or grabbing an object, corresponding to the required shape and the environmental conditions necessary to execute that action," he explains.

A difficult mission

Achieving that modular robots are autonomous is a goal of great complexity, the authors say. "A big difference with regard to robots of a single form is that there are multiple robotic elements, in our case many, independent of each other", explains Daudelin. "That's why there are many more possibilities of error. If an element fails, the entire system may be dislodged, "he says. Tosun remembers that the algorithm that regulates the machine has to choose "not only what he will do, but also what form he will assume to complete his task".

Detail of a module.
Detail of a module.

The engineer says that, during the tests, they noticed many situations in which the robot "did not do exactly what was planned" and that some actions needed in their planning "much more strength before the possibility of failure" than others, because they were in practice more frequently. For all these reasons, Daudelin believes that there are still years of research before a robot with these characteristics is available in the market. "Many improvements in technology are needed to increase the robustness and reconfiguration skills of modular robots," he says.

In spite of the difficulties, the creators of the new prototype assure that the autonomous modular robots could be useful in sectors like the operations of search and rescue or in military actions. "In these applications, the robot could find a variety of different terrains and other unknown environments that require the transformation of its form to adapt," says Dudelain. In military missions, he adds, you can add the inconvenience of explosives damaging parts of a machine. "A modular robot can have an advantage, because you only need to change the damaged modules and not the entire structure," he explains. His partner Tarik Tosun assures that they could also help with daily household chores.

Next generations

Jamie Paik, director of the Reconfigurable Robotics Lab of Lausanne (Switzerland), considers the work of these researchers "impressive and novel", because "it really teaches examples of how modular structures can react and adapt to a new environment and new tasks in an autonomous way". For this researcher, the three plans on which the model they have developed (the autonomous modules, the intelligent control center and the warehouse of preprogrammed actions) allow the system to concentrate "not so much on each individual detail of the behavior of each module , but rather in knowing how to react to the situation ".

This example offers the scientific community a basis to explore "how to reconfigure a robotic format according to how the environment changes", says Paik. The engineer believes that it will be easier to carry out concrete applications in a house than outside, because it is easier to explore the conformation and distribution of objects. "In outdoor environments, more sensors are missing, allowing you to do more things," he says.

Understanding where the line between achievable tasks and unrealizable tasks is is one of the questions that researchers put in this area. Hada Kress-Gazit, a professor at Cornell University and also co-author of the study published this Wednesday in Science Robotic, offers a suggestion: "This question may inspire the design of the next generations of robotic modules, or ideas on how to make two modular robots collaborate to do something different."

.



Source link