Study hard enough, kids, and maybe one day you will become a professional robot hunter. A few years ago, Boston Dynamics set the standard for the field by having people with hockey sticks try to prevent Spot the four-legged robot from opening a door. Earlier, in 2015, distant federal research firm Darpa staged a challenge where it forced clumsy humanoid robots to embarrass themselves on an obstacle course. way out of the competition of the machines. (I once asked you, dear readers, to stop laughing at them, but have since changed your mind.) And now, see, the creators of the Jueying robot dog have learned a fascinating way to take down a human adversary. to fend off who kicks over it or pushes it with a stick.
A team of researchers from Zhejiang University in China – where the Jueying hardware was also developed – and the University of Edinburgh not to educate the Jueying how to recover from an attack, even if they let the robot figure it out. It’s a dramatic departure from how a hardware developer like Boston Dynamics teaches a robot to move, using decades of human experience to hard-code line by line, the way a robot is supposed to respond to stimuli like, um, someone’s foot.
But there has to be a better way. Imagine, if you want, a soccer team. Midfielders, forwards, and a goalkeeper generally all do soccer-like things like running and kicking, but each position has its own specialized skills that make it unique. For example, the goalkeeper is the only person on the field who can take the ball with his hands without being yelled at.
Traditional methods of training robots would require meticulous coding of all those specialized behaviors. For example, how should the actuators – motors that move a robot’s limbs – coordinate to make the machine spin like a midfielder? “The reality is that if you want to send a robot into the wild to perform a wide variety of different tasks and missions, you need different skills, right?” says Roboticist Zhibin Li of the University of Edinburgh, corresponding author of a recent article in the journal Science Robotics describes the system.
Li and his colleagues began training the software that would run a virtual version of the robot dog. They developed a learning architecture with eight algorithmic “experts” that would help the dog develop complex behaviors. For each of these, a deep neural network was used to train the robot’s computer model to achieve a particular skill, such as trotting or straightening itself if it fell on its back. If the virtual robot tried something that brought it closer to the goal, it got a digital reward. If it did something non-ideal, it went digital error. This is known as reinforcement learning. After many such guided trials of trial and error, the simulated robot would become an expert at a skill.
Contrast this with the traditional line-by-line way of coding a robot to do something as seemingly simple as climbing stairs –this actuator spins so much, this other actuator spins so much. “The AI approach is very different in the sense that it captures experience, which the robot has tried hundreds of thousands or even millions of times, ”says Li. So I can create all possible scenarios in the simulated environment. I can create different environments or different configurations. For example, the robot can start in a different position, such as lying on the floor, standing, falling over, and so on. “