After learning how to kick-proof, the robot dog also learned to fall and get up.

 After learning how to kick-proof, the robot dog also learned to fall and get up.

Netease Technologies News Jan. 23, according to foreign media reports, ANYmal, a dog-like robot developed by the Laboratory of Robotic Systems at ETH Zurichuniversity, Switzerland, can maintain balance when subjected to external forces. In other words, when the researchers kicked it with their feet, the robot would turn over and stand up again.

It is reported that ANYmal is a dog-like robot weighing about 30 kilograms with four legs, which is mainly used for industrial inspection of oil and gas bases. Its shape is similar to that of quadruped robots developed by Boston Dynamics and MIT.

It is undoubtedly a difficult task to design programs and algorithms for controlling robotic animals, and it is still a challenge for robotic experts. Robots have many moving parts. Sensors, contacts and cameras can transmit a lot of information at the same time. Robots must be able to quickly screen these information in order to walk smoothly or stay upright.

Researchers at the Robot Systems Laboratory are trying to make this process easier.

They are using computer simulation data to train ANYmals complex behaviors, such as walking, running and turning over from falls. Through a self-developed simulation platform, researchers can simulate more than 2000 actions at one time using a common desktop computer, and then transmit the accumulated data to the robot. This data transmission helps train ANYmal to move with less energy and torque, and to walk 25% faster than before.

But thats not all. When ANYmal is knocked down, the simulation data can keep it completely in balance. The team calls it an extremely complex recovery task. In fact, the analog data helps ANYmal roll over with ease.

If a robot can stand up from any kind of falls, it will be more valuable if it needs to cross uneven terrain. Another advantage of training with simulated data is that the cost has been reduced, but the development time is also increasing.

Researchers also published papers detailing how an efficient, fast, automatic four-legged robot rescues people trapped in forests and mountains, hauls things up stairs, inspects unstructured underground tunnels, and explores other planets. (Han Bing)

Source: Responsible Editor of Netease Science and Technology Report: Wang Fengzhi_NT2541