再RoboCup, July 19th 2014 P66 (ロボットサッカー)

f:id:nprtheeconomistworld:20200204082913j:plain


RoboCup, July 19th 2014 P66 (ロボットサッカー)

 

Messi v the Machine was how some commentators touted the World Cup final, inspired by the disciplined way the German team dismantled Brazil in the semi-finals. But despite such caricatures of Teutonic precision, German players are only human. So as the latest edition of RoboCup, a competition for robot soccer players rather than flesh-and-blood ones, kicks off on July 19th in Joao Pessoa, Brazil's easternmost city, a question that will be on many minds is:when will real machines conquer the sport? When the first RoboCup was held, in 1997, those who launched it set a target of 2050 for engineers to produce a humanoid robot team that would rival the champions of the older competition. Judged by the plodding clumsiness of some of the RoboCup players, that goal might seem farfetched. But it is easy to underestimate how quickly robotics is improving. Self-driving cars and delivery drones, which seemed hopelessly futuristic just a decade ago, are now topics of serious business interest. By comparison with the corporate investments of the likes of Google in electric cars, the teams competing in this year's RoboCup - more than 150 of them - have shoestring budgets. But the tournament includes features that the organisers hope will accelerate innovation without the incentive of cash. One is a clever combination of competition and co-operation. Leading up to the playoffs, teams prepare new strategies and fine-tune their hardware and software in secret. Immediately after the finals have been played, however, all must publish their methods, thus raising the bar for everyone the following year. Another feature is that there are limits to how far teams can push their hardware, to encourage them to develop smart route to victory, rather than using mere brute force. A further twist is that RoboCup is not one competition but many. These range from a little league of miniature cylinders on wheels, like the “Star Wars” character R2-D2, in which each entire team is controlled by one computer using input from overhead cameras, to a fully limbed humanoid league, akin to R2-D2's faithful companion, C-3PO. In the humanoid league, which is further divided into three sizes of robot - kid, teen and adult - each android has its own independent on-board sensors and artificial-intelligence software. Moreover, for those who cannot be bothered to get out the spanners and the soldering iron, there is a virtual leagues as well. Competition there can focus on improving the software needed for the rapid planning of the best move at each step of the game, without having to worry about the vagaries of the hardware. And there is also a junior league for school-age geeks, some of whom will no doubt join the RoboCup major leagues when they go to college, their skills already finely honed. To sense the state-of-the-art in robot football, you need to go behind the scenes. Last week, hidden in a windowless office, Joydeep Biswas of Carnegie Mellon University and his colleagues were making final adjustments to their team, which competes in the little league. While robots in humanoid league are still lumbering and prone to error, the speed and accuracy of those in the little league are stunning. Their electromechanical kickers can fire the league's orange golf balls at eight meters per second (a well-aimed pass in regular football travels at about twice that speed). In fact, they could easily shoot harder if it were not hardware-control regulations that set a maximum speed. Mr Biswas, a graduate student, works for Manuela Veloso. She helped found RoboCup and her group has won the most finals titles in the little league. In 2009 Dr Veloso and her colleagues decided to share with their competitors the vision software that had let their team win a streak of RoboCups. This helped establish the now-mandatory open-source approach that has rapidly raised the quality of the competition. “In the past couple of years,” Dr Veloso opies, “one of the big changes is that we are starting to analyse real football tactics and strategy, to devise our own.” A paper her group published earlier this year lays out how their CMDragons team observed and exploited the defense tactics of opponents, luring them away from positions where they could prevent goals. This approach, dubbed “coerce and attack”, has parallels in professional playbooks. Other research groups are getting equally sophisticated, and teams from Australia, China, Iran and Thailand, among other countries are regularly placed high in several leagues of the competition - in contrast to their national reputations on real pitches. In the early years of RoboCup, there were huge differences in quality between the teams. No longer. The best of the little league routinely finish their ten-minute-long games with the low scores characteristic of well-matched human teams. Indeed, Dr Veloso's squad came in second last year, after a penalty shoot-out following a 2-2 game. Nor need only the players be robots. In a step that many of FIFA's critics may admire, Dr Veloso and her team are developing automated referees. That will not stop some teams from exploiting decidedly human traits, such as fouling by forcefully bumping into another robot. But it may result in more effective enforcement of things like the maximum kick-speed rule. What fascinates Dr Veloso most about RoboCup is the execution, during the game, of moves that had not been deliberately inserted into the algorithms controlling the robots. She is ebullient about an unexpected three-way pass and chip, worthy of a minor Messi and his Argentine teammates. Such unanticipated plays are examples of emergent behaviour, a hallmark of artificial intelligence at its highest level, and something she reckons RoboCup teams in all leagues will produce a lot more of with each passing year. So is 2050 an unrealistic deadline for robots to beat the best humans at football? Half a century is roughly the time that separates ENIAC, America's first electronic computer, from Deep Blue, the IBM machine that beat chess grandmaster Garry Kasparov in 1997. Judged in that light, RoboCup's goal does not seem absurd. Indeed, the question may be whether, come 2050, there are still any human football players around who have not been prosthetically enhanced in some way, making the cyborgs. RoboCup v RoboCup, anyone?