Google invented a sports discipline for robots
Posted: Wed Jan 29, 2025 4:48 am
Agility, a dog competition developed in England in the late 1970s, involves a dog running through an obstacle course under the guidance of its owner. Speed, reflexes, and attention to detail are essential to victory, allowing for a better understanding of the abilities of different breeds.
at Google DeepMind have developed a course for four-legged robots called Barkour. The name combines two words: bark and parkour. However, so far, the robot dogs are doing much worse than real animals.
The quadruped robot industry has made great strides in the qatar number data past decade, with Boston Dynamics' Spot being one example. But comparing their successes is difficult because there is no set standard, says Google DeepMind scientist Atil Isken.
He said that unlike other tests, Barkour includes a variety of obstacles that combine different types of behaviors, such as walking, climbing, and jumping. Moreover, the time limit encourages a focus on speed while maintaining accuracy and variety of movements.
Typically, agility courts take up to 743 sq. m, but Barkour's version is only 25 sq. m. Isken and his colleagues chose four types of obstacles: start and finish areas, poles, a slide, and a jumping barrier.
As in traditional competitions, robots lose points if they fail to overcome an obstacle or miss it, as well as if they exceed the 11-second time limit.
To assess the complexity of a route, DeepMind developed two training approaches:
specialized — trains every necessary skill, such as jumping and climbing a slope;
general – uses specialized training simulations.
The first approach showed relatively better results: 25 versus 27 seconds.
However, both groups not only exceeded the time limit, but also lost to two small dogs - a dachshund and a spitz-chihuahua mix, who completed the obstacle course in less than 10 seconds.
"As this test showed, there is still a large difference in agility between robots and animals," the scientists concluded.
The robots’ performance didn’t live up to expectations, but the team believes it shows there’s still room for growth and improvement. Isken hopes Barkour will become an industry standard because the experiment is easy to replicate.
In the future, the DeepMind team wants to explore the role of human partners in such competitions.
MORE INTERESTING:
Mini Cheetah robot dogs play soccer and do backflips.
The Unitree Go1 robot dog from Unitree Robotics can become a jogging companion
Boston Dynamics robot dogs performed at the Coperni brand show
at Google DeepMind have developed a course for four-legged robots called Barkour. The name combines two words: bark and parkour. However, so far, the robot dogs are doing much worse than real animals.
The quadruped robot industry has made great strides in the qatar number data past decade, with Boston Dynamics' Spot being one example. But comparing their successes is difficult because there is no set standard, says Google DeepMind scientist Atil Isken.
He said that unlike other tests, Barkour includes a variety of obstacles that combine different types of behaviors, such as walking, climbing, and jumping. Moreover, the time limit encourages a focus on speed while maintaining accuracy and variety of movements.
Typically, agility courts take up to 743 sq. m, but Barkour's version is only 25 sq. m. Isken and his colleagues chose four types of obstacles: start and finish areas, poles, a slide, and a jumping barrier.
As in traditional competitions, robots lose points if they fail to overcome an obstacle or miss it, as well as if they exceed the 11-second time limit.
To assess the complexity of a route, DeepMind developed two training approaches:
specialized — trains every necessary skill, such as jumping and climbing a slope;
general – uses specialized training simulations.
The first approach showed relatively better results: 25 versus 27 seconds.
However, both groups not only exceeded the time limit, but also lost to two small dogs - a dachshund and a spitz-chihuahua mix, who completed the obstacle course in less than 10 seconds.
"As this test showed, there is still a large difference in agility between robots and animals," the scientists concluded.
The robots’ performance didn’t live up to expectations, but the team believes it shows there’s still room for growth and improvement. Isken hopes Barkour will become an industry standard because the experiment is easy to replicate.
In the future, the DeepMind team wants to explore the role of human partners in such competitions.
MORE INTERESTING:
Mini Cheetah robot dogs play soccer and do backflips.
The Unitree Go1 robot dog from Unitree Robotics can become a jogging companion
Boston Dynamics robot dogs performed at the Coperni brand show