Robot dog has learnt to walk on practically any surface
Developers trained the four-legged robots based on 4,000 virtual clones on all sorts of terrain
Quadruped robot designers like Boston Dynamics have taken great care to develop systems that can move through a range of terrain. For the right amount, you can buy a robotic dog that can kick, get back up and keep going.
Researchers from UC Berkeley and Carnegie Mellon developed a technique for teaching robots to traverse challenging terrain. The list includes slippery, uneven ground, stairs and curbs.
Scientists trained the robots using simulators, sending 4,000 virtual clones on their voyage across a variety of terrains.
The researchers say that, by using this method, they were able to achieve six years’ worth of robot walking experience in just 24 hours. The neural network updates the information gained during the simulations. The device can modify its legs and adapt to its environment quickly, which can drastically reduce the cost of robots.
The system works similarly to cats and other animals: four-legged animals have memories that allow them to keep track of their front and rear legs. The system can also work in low light conditions and can climb steps that are of its own height.
