Scientists teach ping-pong robots to master spin
Humans can master the art of spin and outwit the machines - but maybe not for much longer
The world could be just two years away from the first serious showdown between a robot and a top-flight ping-pong player with a mainland breakthrough in high-speed vision in robots.
Scientists with a state-funded project at Zhejiang University said they hoped not only to beat the world's ping-pong best but also to apply the technology to industrial and military robots.
Professor Xiong Rong with the university's Institute of Cyber-Systems and Control said the team had found a way to track an ordinary ping-pong ball in full spin. Using data such as the ball's rotational speed and axis, their robot was able to make a precise estimate on the ball's movement after striking the table.
Spin on the ball is one of the most important weapons in table tennis, and a nightmare for robotic scientists to track. Robotic teams worldwide have been trying to come up with various prototypes of ping-pong robots for a couple of decades, but none could deal with the spin, as the estimates of the ball's speed or trajectory were often way off.
Xiong's team came up with a creative solution. They followed the brand logo, which is printed on every ping-pong ball, with high-speed, high-definition cameras, and by analysing the movement of the logo, they obtained high-quality data on the physics of the ball's spin.
Each camera produced more than 120 images per second, and their computer analysed hundreds of high-resolution photos taken in the blink of an eye to keep up with the intense pace of a game.
"The method is beautifully simple in theory but difficult in practice," Xiong said.
The challenge is complicated by the team's desire to use technology within easy reach.
"The optimal situation, of course, would be to employ a giant computer such as Tianhe II [the world's fastest supercomputer] to process the images, and it would be great to use a camera capable of capturing thousands of photos per second, but we don't want to build an astronomically expensive robot that can't move around the table.
"We want to come up with a system that can be run from a personal computer, and can be used in an environment with normal lighting that would be comfortable to a human player but not too dim for extremely high-speed cameras, which need lots of light for the short exposure time of each frame," Xiong said.
That will mean upgrading the team's two robots, Wu and Kong. The robots, named after the monkey king in Journey to the West, can already make straight shots, and the researchers plan to spend the next two years upgrading the technology to give Wu and Kong a killer spin return.
Part of the problem has been that the coating material on each pad varies, albeit slightly, which affects numerous parameters in their computer models. Even fluctuations in room temperature affect the physical properties of a paddle.
The team are also writing some sophisticated code in the control software so the robots could "learn from each failure", Xiong said.
"While a human player might be negatively affected by a miss, the robot will only improve from it," she said. "We are also giving it the ability to analyse the body movement of a human opponent and use the data to predict the ball's trajectory."
Once the upgrades are done, Wu and Kong might be able to compete against a high-school champion, Xiong said.
She said the race for ping-pong robot technology was intense. Germany is close behind China, and teams in countries such as Japan and the United States could catch up at any time.
She said some of the technology used in their ping-pong robots was already being used in the production lines of some Chinese factories, doing precision jobs that were previously the realm of people.
The technology will significantly improve efficiency and cut labour costs in the manufacturing sector, such as in the car industry, she said.
Professor Liang Zize, a researcher in another ping-pong robot project with the Chinese Academy of Sciences' Institute of Automation, said Xiong's team had made a remarkable achievement because getting a robot to apply spin was one of the biggest challenges in the field.
"The technology requires some very sophisticated algorithms for high-speed vision analysis. Lots of research has been done, but the problem still hasn't been solved," he said.