A team of scientists from Zurich, Switzerland, have developed a robotic dog that can be trained with real-world tricks using artificial intelligence.
The robot, called ANYmal, is able to move and respond faster than any other four-legged robot can, thanks to its “neural network system from which the robot would be able to quickly learn and adapt”. It uses a method called reinforcement learning, still relatively new in this field, where the robot can learn through something akin to trial and error, repeating various tasks over and over until it gets it right, and in good speed. In fact, it was able to beat the previous record from a robot about 25% faster.
ANYmal is also able to withstand rough kicks and even stand back up if it falls. A video showing scientists testing the dog was released as proof of this. Its limbs responded so fast, adjusting based on the kick, that it rarely ever tipped or fell over. According to the Swiss scientists, if ANYmal does tip over, the dog has a so-called “roll-over motion” which allows it to quickly recover from a fall. This “roll-over motion” has reportedly never been observed before in any other four-legged robot.
With its ability to survive a fall, scientists say that four-legged robots could be really useful to humans in the future, especially in rescue missions. “Legged robots may one day rescue people in forests and mountains, climb stairs to carry payloads in construction sites, inspect unstructured underground tunnels, and explore other planets. Legged systems have the potential to perform any physical activity humans and animals are capable of.”