‘Humanoid robot soccer players will win’

By 2050, soccer-playing robots expect to beat the winners of the ‘World Cup of Human Soccer’.

At the 3ME faculty, home to the university’s mechanical engineers, the Delft Biorobotics Laboratory (DBL) developed its first biped walking robot, named ‘Stappo’, in 1995. Over subsequent years, DBL researchers developed many other prototype robots, including ‘Bob’, ‘Mike’ and ‘Max’, which are now all proudly displayed at the laboratory.

These robots evolved from basic research in ‘Passive dynamic walking machines’, which are machines that can walk down slight slopes
without using motors or electronic controls; instead, they’re propelled by gravitational force exerted on their robotic limbs. As DBL’s research advanced, however, more sophisticated robots were built, with motors and electronic controls aiding their walking motions. ‘Leo’, the latest addition to DBL’s ‘robot family’, can stand up by itself if it falls down. The ultimate aim of all this biologically-inspired robot design research is to develop human-like walking machines that can walk as naturally as human beings.

Touring the Delft Biorobotics Laboratory, one is suddenly confronted by a hanging, headless robot with bunches of wires running through its body. “That robot is named TUlip and is being developed by the Dutch Robotics team”, says Sebastiaan Kiemel, a second-year MSc bio-mechanical design student and the team manager of Dutch Robotics. Dutch Robotics is an initiative by TU Delft, TU Eindhoven, UTwente and Philips to create teen-sized humanoid robots. The robots, called teen-sized because of their height (100 to 120cm), serve two purposes: primarily as part of ongoing university research to make robots walk as naturally as humans, and as prototypes capable of participating in the annual RoboCup competitions, in which robots attempt to play soccer.
The RoboCup Federation is an international research and education initiative that uses a universal challenge – in this case, robots playing soccer – to foster artificial intelligence and robotics research, in which a wide range of technologies can be studied and integrated. The first RoboCup soccer competition was held in July 1997, in Nagoya, Japan. Although the RoboCup Federation chose to use robots playing soccer as its central pursuit, the robotic innovations resulting from this high-level robotics research will ultimately be applied to socially important problems and industries. In other words, it’s not about the game per se, but soccer does have many aspects that next generation technologies need to embrace, such as teamwork, real-time perception and decision-making, and high-level motion control.
Nevertheless, RoboCup’s official goal is: “By 2050, a team of fully autonomous humanoid robot soccer players will win a soccer game, complying with the official FIFA rules, against the winner of the most recent World Cup of Human Soccer.” 

Bottleneck
RoboCup has five different categories of competition, each with different goals: Simulation League, SmallSize Robot League, MiddleSize Robot League, Humanoid League, and Standard Platform League. Humanoid League in turn has three different robot classifications: KidSize (30-60cm height), TeenSize (100-120cm) and AdultSize (130cm and taller).

Robots with human-like bodies and human-like sensing compete in the Humanoid League. “Dutch Robotics is a student project and we strive to do it with students”, Kiemel says. Dutch Robotics’ first milestone was to build TeenSize humanoid robots that competed in RoboCup Soccer 2008, in Suzhou, China, where the team’s robot, named ‘Flame’, unfortunately lost in the quarterfinals to ‘NimbRo’, a robot developed by the University of Bonn (Germany). Last summer, Dutch Robotics competed in Graz, Austria, with a new robot, named ‘TUlip’, which performed better but also lost in the quarterfinals to ‘CIT brains’, a robot developed by the Chiba Institute of Technology (Japan).

“This year we’re further developing these robots and will try to win the TeenSize Humanoid League competition in Singapore”, Kiemel says. Each year the RoboCup Federation changes the rules to make the game more like a real soccer game, with the teams facing more complex design requirements that require building more advanced robots. For example, the TeenSize Humanoid League initially featured a penalty kick competition for teams whose platforms focused explicitly on dynamic walking. In 2008, the penalty kick competition was replaced by a more challenging ‘Dribble and Kick’ competition. Kiemel: “And now, in 2010, for the first time the TeenSize soccer competition will feature teams of two autonomous robots competing against each other.”
Dutch Robotics team members typically devote about six to ten hour per week to the project, Kiemel says, and the team’s main goal at the moment is to qualify for RoboCup’s screening round, held later this month. “To qualify our robot needs to be able to stand up, locate and walk to the ball, and kick the ball”, Kiemel explains. “This is a 3TU project, so each university is now focused on getting the best from their respective fields of expertise. TU Delft mainly focuses on the mechanics and software, while TU Eindhoven handles the control and UTwente helps with simulations.”

The Dutch Robotics team however is still looking for software developers to help frame software architecture and guide others currently involved in the project. The team does have some sponsors, including Sogeti, which will write the software for the vision control, but would like more. “So we’re also looking for volunteers who can help market TUlip and bring in more sponsors”, Kiemel adds.

Kiemel notes that a strength of the Dutch Robotics team is the professors involved in the project, as well as the PhD students who provide good background support. Kiemel: “On the technical side, we’re focusing a lot more on smart mechanics, rather than heavily relying on process and control, which is a major advantage of our design. The design becomes more stable, reliable and energy efficient if the robot can walk similar to humans.”
Although robotics is rapidly progressing, it is 2010 and robots can just about kick a soccer ball. Is beating a human soccer team by 2050 really achievable? “Of course it’s hard to make predictions about this, but if we look to the past we see that certain developments can suddenly advance very quickly”, Kiemel says. “But one crucial thing that must happen is to create a reliable and robust walking gait. Many robots can walk, but the robustness of a robot dealing with unknown situations, like, for example, recovering from a shove, is a real bottleneck.”

Although Kiemel admits that actually beating a human soccer team by 2050 seems unlikely, he cheerfully concludes that “if a team of eleven humanoid robots at least compete against a human soccer team by 2050, that would be quite an impressive achievement in itself.”

Students interested in joining the Delft Robotics team should contact Sebastiaan Kiemel at info@dutchrobotics.net

Reinout Rutte, 36
Universitair docent geschiedenis bij Bouwkunde

Ik dacht vandaag eens iets gezonds te nemen. Mijn voorkeur gaat uit naar frikadellen en zo. Of soep. Soep vind ik ook lekker. Maar dat is onhandig mee te nemen naar de overkant, waar ik les geef. Deze lunch kan ik goed stapelen. Het broodje heb ik genomen omdat het een verrassingsbroodje leek. Ik kon niet bedenken wat er op zat, maar het zag er goed uit. Even proeven. Oh, het is iets vissigs. De vernieuwde bouwkundekantine is heel goed. Iedereen doet nu extra zijn best, lijkt het. Ik neem nooit meer iets mee van thuis, het is hier veel te lekker. Een pluim daarvoor.