It is called ANYmal and, as its name suggests, it operates as such. It is a robot that has four legs, the size of a dog and with a lot of autonomy. Locomotion – the action of moving from one place to another – is one of the most difficult skills to develop in machines. Normally, they learn to walk in familiar controlled environments to avoid crashes or falls, but researchers at the ETH Zurich Polytechnic School in Switzerland have made ANYmal face real-world challenges on its own. The latest version of this robot can climb stairs, walk on very narrow beams and navigate natural environments with vegetation, rocks, snow and all kinds of slippery and complex surfaces that it does not know. The creators have published their progress in the magazine Science Robotics.
To achieve its objectives, ANYmal overcame courses full of obstacles on 76,000 square meters of land designed by the researchers. The circuit demanded a lot of precision in walking or a great amount of robustness to overcome it without falling. “We reconstructed difficult environments that would likely be found at natural disaster sites, where debris might break down further when stepped on, and at construction sites, where there are oil stains that create slippery surfaces,” details Fabian Jenelten, professor of robotics at the ETH Zurich and lead author of the study.
This robot is capable of operating autonomously, meaning it can make decisions and adapt to its environment without constant human intervention. To do this, it is equipped with a variety of cameras, proximity sensors and gyroscopes that allow it to perceive its surroundings. Their lidars (laser devices) accurately measure distances and create maps in real time. In addition, this machine can detect gas leaks and carry up to 10 kilos.
Quadruped robots learn to walk in two ways: with deep learning methods that give them skills to deal with different terrain or with traditional methods that achieve more precise movements. Thus, they end up being either robust or accurate, just one of the two, and when faced with the real world, they tend to fail. Jenelten’s team decided to address this weakness in robotics by combining the best of both worlds: the precision of traditional methods and the robustness of deep learning. Thus, ANYmal achieved proprioception, that is, the awareness of the position of his body—which humans and animals rely on—to move. “In our work, we try to combine the two aspects, allowing quadruped robots to conquer new environments that seem to be relevant for future applications,” explains Jenelten.
ANYmal is not one of a kind. Your distant relative Spot, from the company Boston Dynamics, is a quadruped robot with very similar characteristics. Ford has acquired some that already run around its factories to track and prevent possible breakdowns on vehicle assembly lines. Then, the defense company Sword Defense Systems installed in these robots a precision rifle that hits targets more than a kilometer away and has thermal vision to shoot in the dark, while others Spots They have moved to the countryside of New Zealand to be sheep herders.
The inventors of ANYmal believe that in a couple of years they will be able to actively assist humans in rescue missions, for example, following an earthquake or explosion, where, with the help of lidars, they will create maps that anticipate landslides or unstable routes. Looking to the future, Jenlets explains, “these robots have enormous potential to completely replace humans in dangerous and potentially lethal environments.”
Robots: more than a replacement, an assistance
Rodolfo Haber, director of the CSIC Center for Automation and Robotics, suggests considering that these robots “more than a replacement for humans, they will be of assistance and accompaniment.” In large cities, he explains, they can fulfill support functions for people with hearing, visual and other disabilities. And machines will not necessarily replace workers: “A robot can help a person who works in construction, for example, reach 60 years of age stronger,” explains Haber.
Locomotion with legs “was very fashionable” some years, according to Haber. In Spain, for example, CSIC researchers created machines with six legs that performed mine detection tasks. Many quadruped robot projects were abandoned, but the researcher believes that advances in artificial intelligence can improve the models, so many companies and industries are going to take them up once more and there will be a large market at an industrial and social level.
Both the creators and the CSIC robotics expert rate the advances in ANYmal locomotion as valuable for the future of legged robots. Given these progress and the constant evolution of technology and artificial intelligence, it is likely that in a few years it will be common to walk the streets among four-legged robots, which assist people in all types of tasks.
to continue reading
_
