The Economist | from the print edition
Sporting robots are still slow. But their inventors are making rapid strides
ONLY a few of the many thousands who applied to carry Britain’s Olympic torch in the relay that finishes on July 27th made the final selection. But one application (albeit unsuccessful) stood out. James Law, a computer scientist at the University of Aberystwyth in Wales, nominated iCub, his metre-high humanoid robot.
Competition to build athletic machines is intensifying as greater processing power makes them nimbler and handier. Disembodied arms at the University of Tokyo pitch baseballs, and bat them back, with uncanny accuracy. Engineers in Munich have built a pool-playing automaton that can pot five balls in a row. Their counterparts at Zhejiang University in China boast a pair of humanoid table-tennis players. (They can rally with a human for 145 strokes, and with each other for about 180.)
Sporting robots delight schoolchildren and excite students, says Peter Stone, an academic in Austin, Texas, who helps to organise the RoboCup—an annual competition for robot footballers. But they also spur innovation. The same software that helps miniature robots form sports teams could co-ordinate larger rescue robots in a disaster zone. Games that require robots to take humanoid form spur research into machines that might one day care for the elderly or run errands for the infirm.
Some sporting robots have more immediate practical uses. Robotic arms that hit perfect drives help golfing firms produce ever-better clubs. A martial-arts robot at Loughborough University batters prototype protective sports gear. The United States Bowling Congress uses a machine named E.A.R.L. to measure whether ten-pin facilities are up to scratch. These appliances are valued for their consistency, but uncanny precision is a disadvantage in a real game. E.A.R.L. gets less precise over ten frames (its ball wears out the surface of the lane by always hitting the same spot).
Even “ageing amateurs” can defeat today’s batch of footballing automatons, admits Mr Stone. But cleverer breeds of robot will learn from watching humans, thinks Subramanian Ramamoorthy at the University of Edinburgh. New technologies are generating ever more data about flesh-and-blood sportsmen—much of which might be fed into machines. Basketballs made by 94Fifty, an American start-up, are packed with sensors which measure how accurately players pass, dribble and shoot. Cameras owned by SportVU, an Israeli firm, identify and monitor players on sports fields. The company uses missile-tracking software to crunch the huge array of statistics that their moves provide.
Such omniscience could make artificial athletes unbeatable—if their hardware were up to scratch. But without wheels, most automatons are still painfully slow. The tiny athletes who took part in the world’s first robot marathon, in Osaka, took more than two days to finish the course. Usain Bolt, the world’s fastest man, reaches 28 miles (44km) per hour; but the fastest robot on legs, a doglike creature built by Boston Dynamics, runs at only around 18mph. The swiftest two-legged machines, like one built by scientists at the University of Michigan, now travel at about 7mph. That is about the speed of a jogging human.
iCub’s successors may gain revenge eventually: entrants in the RoboCup hope their teams could beat humans by 2050. For now their machines have more brains than brawn—though at least they have no penchant for embarrassing partying when the games are over.