‘Parkour’ robot dog can leap, jump, and crawl its way through complex obstacle courses

Researchers studied some very agile 'freerunners' to make this machine
parkour robot
Researches used a model trained on parkour completions to teach this 100 pound robot to run through an obstacle course. YouTube

Four-legged, dog-inspired quadruped robots have already proven capable of a variety of tasks, from remotely monitoring sports stadiums and guiding blind persons to inspecting potentially hazardous research areas. These robots are generally more agile than their hulking bipedal counterparts, but they have mostly failed to match the fluid grace and athleticism of their furry canine inspirations. Now, a new wall-scaling robot is pushing the boundary of what these quadrupeds are capable of, and it’s doing so with a bit of flair. 

Researchers from ETH Zurich are trying to close the mobility gap between robots and animals with a new highly-mobile robot capable of running, jumping, and crawling its way through obstacle courses. 

The researchers, who published their findings in Science Robotics this week, set out to to teach ANYmal, a 100 pound quadruped robot made by the firm ANYbotics how to mimic human “freerunners” who engage in an underground sport referred to by many as “parkour.” 

In a nutshell, parkour centers around getting from one point to another in the fastest way possible and often involves swiftly crawling and leaping through obstacles along the way. Parkour, which is performed on obstacle courses or even in dense urban areas, requires a combination of athleticism and rapid decision making. ANYmal was up to the task. The newly improved robot was able to complete the  basic parkour course below moving at a clip of six feet per second. 

How did ANYmal learn parkour? 

A video of newly trained ANYmal in action shared by ETH Zurich shows the beefy red robot clambering up a small wooden staircase before leading over a small gap to land on another platform. Without breaking its stride, the robot charges forward then dives down to scramble underneath an obstacle in a motion resembling a scouring insect. ANYmal quickly pushed itself back up so it can climb vertically up another crate slightly taller than its body. The video shows the robot completing the course even when the obstacles are scrambled in different orders. 

ANYmal uses onboard laser sensors to perceive its environment and create maps it can use to autonomously plan and execute a travel path. Four lightweight carbon legs and 12 identical motors propel it towards its target. The ETH researchers set out to improve the robot’s movement by using a neural network composed of three separate modules, each devoted to locomotion, perception, and navigation selectively. 

In their paper, the researchers say they developed the navigation element of ANYmal to understand the robot’s capabilities in walking, jumping, and crouching. Armed with that context, ANYmal can automatically adjust its behavior depending on the type of obstacles prevented. The end result: a robot that can quickly identify and react to a range of obstacles and traverse 


ETH doctoral student Nikita Rudin, one of the researchers working on ANYmal’s improvements, is reportedly himself a parkour enthusiast and tapped into that experience during the research. 

“Before the project started, several of my researcher colleagues thought that legged robots had already reached the limits of their development potential but I had a different opinion,” Rudin recently told Science Daily. “In fact, I was sure that a lot more could be done with the mechanics of legged robots.”

Researchers trained the model on examples of human freerunners who complete parkour courses. ANYmal learned its new skills through basic trial and error in repeated simulated environments. Eventually, the robot was deployed in a physical obstacles course where it used those lessons learned in simulation to complete the course. 

“By aiming to match the agility of free runners, we can better understand the limitations of each component in the pipeline from perception to actuation, circumvent those limits, and generally increase the capabilities of our robots, which in return paves the road for many new applications, such as search and rescue in collapsed buildings or complex natural terrains,” the research said in a statement

Parkour inspiration could improve robots mean for search and rescue and space exploration

Unlike human freerunning, ANYmal’s new skills are intended for more than just looking cool. Looking to the future, researchers believe similar improvements could be applied to search and rescue robots to help them vault over obstacles or scurry into difficulty to reach areas during search and rescue missions. One day, the researchers note, these same type of advancements could possibly even aid in space exploration robots traversing harsh, rocky surfaces of the moon and other planets. The added location could also apply to quadruped robots investigating hazardous areas of the Large Hadron Collider in Geneva, Switzerland. 

Those use cases are, for now, still largely hypothetical. In the meantime, ANYMal will join the likes of Boston Dynamics’ Spot and Atlas robots on the ever growing list of metal machines uncannily performing athletic feats once reserved living beings.