Imagine a robot dog that can not only walk and run, but also jump, climb walls, and even crawl under obstacles. This isn't science fiction anymore. Researchers in Switzerland have developed a new way to train robots to move with the agility and grace of a parkour athlete.
Four-legged Robots Get a Parkour Upgrade
Robots with four legs, inspired by the movement of dogs and other animals, have been around for a while. They're already used for various tasks, like inspecting dangerous areas or helping people with disabilities. But these robots haven't quite been able to match the impressive athleticism of real animals. This new research is changing that.
A team from ETH Zurich in Switzerland wanted to bridge the gap between how robots and animals move. They focused on a robot named ANYmal, a 100-pound machine built by the company ANYbotics. Their goal? To teach ANYmal how to move like a human parkour athlete.
What is Parkour?
Parkour is a sport where participants move from one point to another as quickly and efficiently as possible. This often involves jumping, climbing, and crawling over obstacles along the way. It takes a combination of strength, speed, and quick thinking to navigate a parkour course, which can be set up in an obstacle course gym or even in a city environment.
The researchers successfully trained ANYmal to complete a basic parkour course, moving at a speed of about 6 feet per second.
How Did ANYmal Learn Parkour?
A video released by ETH Zurich shows ANYmal in action. The robot climbs a small wooden staircase, leaps over a gap, and lands on another platform. It then charges forward, dives down to crawl under an obstacle, and quickly pushes itself back up to climb a vertical crate. The impressive part? ANYmal can even complete the course when the obstacles are rearranged in a different order.
So how does ANYmal "see" and move through its environment? It uses built-in laser sensors to create maps of its surroundings. These maps help the robot plan and execute its path autonomously. Four lightweight legs powered by 12 motors propel ANYmal forward.
The Secret Ingredient: A Neural Network
The researchers used a special computer program called a neural network to improve ANYmal's movement. This network is made up of three parts, each focusing on a specific task: movement (locomotion), perception (sensing the environment), and navigation (planning its path).
The navigation part of the network was designed to understand ANYmal's physical abilities, like walking, jumping, and crouching. With this knowledge, ANYmal can automatically adjust its movements depending on the obstacle it encounters. The result? A robot that can quickly identify and react to different obstacles, allowing it to traverse complex terrain.
Inspiration from a Parkour Enthusiast
One of the researchers on the project, Nikita Rudin, is actually a parkour enthusiast himself. His passion for the sport played a role in the development process. "Before the project started, some researchers thought legged robots couldn't improve much more," Rudin said in a recent interview. "But I knew there was more potential."
Learning from the Best: Humans and Simulations
The researchers trained the neural network by showing it examples of human parkour athletes completing courses. ANYmal then learned its new skills through trial and error in simulated environments. Finally, the robot was tested on a real obstacle course, where it used its simulated training to successfully navigate the challenges.
Beyond Looking Cool: Real-World Applications
While ANYmal's parkour skills are impressive, the researchers have bigger goals in mind. They believe these advancements can be applied to search and rescue robots, allowing them to jump over debris or squeeze into tight spaces to reach victims. In the future, these skills could even be used on robots exploring the rocky terrain of the moon or other planets.
For now, ANYmal joins the ranks of other impressive robots like Boston Dynamics' Spot and Atlas, blurring the lines between machines and athletic animals. This research is a big leap forward in robot agility, with the potential to improve robots used in a variety of real-world applications.