by Credit: Joseph Humphreys, University of Leeds
With its stumbling gait and the preliminary steps, it could be a baby animal that learns to maneuver through unknown terrain.
But this is Clarence, which was taught by animal, which was taught to navigate a dog or a horse through the world, and the first in the world to adapt to unknown surfaces.
It is to be hoped that it could be used in dangerous environments such as bombing or dangerous search and rescue operations where it is too risky to send in a person.
Tests show that it can successfully mix through confused roots, mix through loose wooden foams, mix up in muddy fields and even negotiate unequal stacks of boards.
Joseph Humphreys, a postgraduate researcher at the University of Leeds, who contributed to the development of artificial intelligence (AI) in the development of Clarence in the development of Clarence, said: “The matrix is similar to NEOS skills in martial arts to his brain, but he is not lost physical training in the real world.
“We tested the robot in the real world on surfaces that he had never experienced before, and he successfully navigated them all.
“It was really worth seeing to see how it adapted to all the challenges we made and to see how the animal behavior we studied had almost become the second nature for it.”
Most robots are cleverly to follow a specific task on familiar surfaces, but fight when the environment changes. The researchers have overcome the problem by conveying their system with natural strategies for movement movement.
The researchers were inspired by dogs, cats and horses, the experts for adapting to different landscapes and quickly change the way they save energy, maintain balance or react quickly to threats.
The program also teaches the robot how it can change between trot, running, limitation and more, just as mammals do in nature.
It means that Clarence not only learns to move, but also learn how to decide which gang is to be used, when you change and how to adapt it in real time, even on the site that she has never met before.
Credit: Joseph Humphreys, University of Leeds
It only took nine hours for the robot to find out its new skills much faster than the days or weeks that most young animals take to safely cross different surfaces.
Professor Chengxu Zhou, Senior Author of the Study from the University College London, said: “This research was driven by a fundamental question: What if Leged -Robots are doing the way animals do? Instead of training robots for certain tasks, for certain tasks that they adapt to the adjustment of their compensation, coordination, coordination and energy efficiency wanted.
“By embedding these principles in a AI system, we have enabled robots to choose how you can move based on real-time conditions and not on preprogrammed rules. This means that you can control unknown environments safely and effectively, even those whom you have not met beforehand.
“Our long-term vision is to develop embodied people in one person in a humanoid robot, which adapt, adapt, adapt, adapt and interact with the same resistance with the same fluidity and resistance.”
The team says that the performance is a big step forward to make more adaptable and able to act with real challenges, in dangerous environments or in position, to be in dangerous environments or to be able to make access difficult.
A robot that can navigate unknown, complex terrain opens up new opportunities for the use of disaster, planetary, agricultural and infrastructure inspections.
In the future, the team hopes to lend dynamic skills such as the long -range jump, climbing and navigating steep or vertical terrain.
Research was published in the Nature Machine Intelligence magazine.
