RoboCup’s humanoids have played their first 11-a-side match. This October, some of the
same problems — balance, vision, autonomy and plenty of debugging — come to Maker
Faire Rome.
This summer, twenty-two humanoid robots walked onto a football pitch in Incheon, South Korea. Eleven against eleven. No remote controls. On 5 July 2026, B-Human from Bremen and HTWK Robots from Leipzig played the first 11-a-side humanoid robot football match in RoboCup history. It was an exhibition played after the official competition, on a larger pitch, and lasted one half. B-Human won 4–0. The score was interesting. The fact that the match could be played at all was more interesting.
When RoboCup introduced its first humanoid competition in Fukuoka in 2002, the robots
were still tackling limited challenges rather than playing a full football match. Twenty-four
years later, two teams have managed an 11-a-side game. That is quite a lot of engineering in between. A robot playing soccer maps out the kinematic vectors, wheel velocities, and directional forces required for an autonomous mobile robot to navigate, track a ball, and execute kicks on the field.
Why make a robot play football?
RoboCup began in 1997 with an unusually clear long-term goal: by the middle of this
century, a team of fully autonomous humanoid robots should be able to beat the reigning
human World Cup champions under FIFA rules. It still sounds improbable and that is rather the point. Football is a useful robotics problem because almost nothing stays still. The ball moves.
The other players move. The robot itself is balancing on two legs while trying to work out
where it is and what it should do next. There is no fixed sequence of movements that will get you through a match.
A robot has to locate the ball, understand its position on the pitch, recognize other players,
choose an action, move without falling over and react when somebody else gets there first.
And all of those systems have to agree with one another. Good vision is not much help if the robot cannot turn quickly enough. A beautifully trained kick is useless if the robot has misjudged the position of the ball. Improve the walking system and you may create a new problem for localization. Change a component and software that worked yesterday may need tuning again. For a maker, this is where robot football gets good. You can see the whole system being tested at once–same robot, different football. But things are changing fast.
The 2026 season also brought a significant change to RoboCup. The former Standard Platform League and Humanoid League were brought together in the new Humanoid Soccer League. In the Middle Division, teams including B-Human and HTWK Robots competed using Booster Robotics’ K1 humanoid platform. Having the same basic robot does not mean having the same robot on the pitch. B-Human developed its own motion systems for the K1, including running, shooting and getting back onto its feet. The team used deep reinforcement learning to train those behaviours rather than relying entirely on the manufacturer’s standard movement routines. This is the bit worth paying attention to. Buying a humanoid platform removes one enormous engineering job: you no longer have to design every joint, gearbox and structural part yourself. It does not give you a footballer. That still involves software, training, control, testing, tuning and a great deal of time spent discovering that a behaviour which looked fine in simulation behaves rather differently once motors, batteries, friction and gravity are involved.
B-Human went on to win the RoboCup 2026 Middle Division title, beating HTWK Robots
6–0 in the official final. The two teams then came back for the experimental 11-a-side
match.
The hardware may increasingly come off the shelf. What you make it do is another matter.
Watch what happens before the kick Robot football becomes much more interesting when you stop watching only for goals. Look at the few seconds before a robot reaches the ball.
How did it find it? How does it know where its own feet are in relation to it? Why did it approach from that angle? How quickly can it change direction? What happens when another robot crosses the path it had planned? And when it falls over — which robots still do — how does it decide how to get up and back into the game?
RoboCup comes to Maker Faire Rome
A match bundles together problems in computer vision, localisation, motion planning, bipedal locomotion, machine learning, control and coordination between several autonomous machines. It also exposes something every hardware maker eventually learns: the physical world has opinions. A simulator does not have a connector that has worked loose. Its battery does not fade halfway through a run. Its foot does not slip because the surface is slightly different from the one you tested on. Real machines do all of those things.
That is why watching them fail can be every bit as useful as watching them score.
From 23 to 25 October, Maker Faire Rome will host the first Italian Open RoboCup Humanoid Soccer Tournament, an official RoboCup Regional Event, at the Gazometro Ostiense.
The Rome event will feature the Middle and Large Divisions, with 3-v-3 and 5-v-5 matches
depending on the teams and robot platforms taking part. There will also be technical testing, public demonstrations and a RoboCup Humanoid Soccer workshop. That makes it rather different from watching a polished humanoid demonstration video. The robots have something to do. They need to find the ball, move into position, stay upright, make decisions, cooperate with team-mates and deal with whatever the other side does next. Sometimes that will work.
Sometimes somebody will be reaching for a laptop. Both are part of the story.
Just as important is to talk to the people behind the robots. At a Maker Faire, the machine is only half the exhibit. The other half is usually the person standing beside it, happy to explain why version three failed and why version four has a completely different knee joint. RoboCup teams bring together students, researchers and engineers working across mechanics, electronics, AI, computer vision and control systems. What looks like a simple movement on the pitch may be the result of weeks spent changing a gait, retraining a behavior or trying to understand why the real robot refuses to behave like the model.
That is what makes RoboCup such a natural fit for Maker Faire Rome. There is a competition, certainly. But underneath it is the familiar cycle: Build it. Test it. Find out what is wrong. Change it.Try again. RoboCup’s World Cup goal is still twenty-four years away. Nobody knows whether a robot team will be ready to face the human champions in 2050. For now, there is a better reason to watch. You can see how they are building their way towards it. Maker Faire Rome – The European Edition 23–25 October 2026 Gazometro Ostiense, Rome
ADVERTISEMENT
