Robot Football: RoboCup, Humanoids and ABB

Robot football started as a research problem and now draws crowds, who watch humanoid robots dribble, fall over and sometimes score. This post traces robot football from the first RoboCup in Nagoya in 1997 and its goal of beating the human World Cup winners by 2050, through the robot dog and humanoid leagues, to the humanoid matches in Beijing and Incheon in 2026. It also looks at what the newest research says about how far the robots still have to go. ABB robots show up at the end, in the classrooms where many of these engineers learn their trade.

Two humanoid robots fight for the ball at RoboCup 2026 as the robot in white shoots at goal. Photo: Ayleen Lührsen / Wikimedia Commons (CC BY-SA 4.0)

How RoboCup turned football into a robotics benchmark

In the mid 1990s, artificial intelligence researchers were looking for a problem to follow computer chess, and in 1997 IBM's Deep Blue beat world champion Garry Kasparov. Chess is a closed game where both players see everything. In football the ball moves fast, players block each other's view, and a team has to plan together in real time. A group of professors including Hiroaki Kitano, Minoru Asada, Manuela Veloso and Itsuki Noda founded RoboCup in 1996. The first competition took place in Nagoya, Japan, in 1997, with 38 teams from 11 countries.

RoboCup world champion robots and a trophy on show at Toyota's AMLUX centre in Tokyo, 2007. Photo: Luis Villa del Campo / Wikimedia Commons (CC BY 2.0)

RoboCup set itself a long goal: by the middle of the 21st century, a team of fully autonomous humanoid robots should win a game, under FIFA rules, against the most recent World Cup winners. Early matches were a long way from that. Small wheeled robots pushed an orange golf ball around a table, often guided by a camera hanging over the field. In a separate simulation league, software agents played on a virtual pitch, so teams could work on tactics without building any hardware.

Sony Aibo robot dogs playing football in RoboCup's four-legged league, 2007. Photo: Pablo / Wikimedia Commons (CC BY-SA 2.0)

Legged robots came next. Sony's Aibo robot dogs played in a four-legged league, and later the Nao humanoid from Aldebaran became the shared robot of the Standard Platform League, where teams compete on software alone. A humanoid league for robots on two legs began in 2002. For years those matches were slow and clumsy. Robots paused for long stretches to find the ball and often fell when they kicked, yet universities kept entering because the games tested vision, walking and team coordination all at once.

Nao humanoids of the HTWK Leipzig team on a robot football pitch, 2010. Photo: Die Schreibfabrik / Wikimedia Commons (CC0)

Humanoid robot football in 2026

The second World Humanoid Robot Games ran from 22 to 26 August 2026 at Beijing's National Speed Skating Oval, with 666 teams and 2,056 robots from 16 countries. Football was one of 51 events. B-Human, a team from the University of Bremen and the German Research Center for Artificial Intelligence (DFKI), won two humanoid soccer competitions there and beat the Badger Bots from the USA 15 to 0 in one semifinal. RoboCup 2026 itself was held in Incheon, South Korea, in July.

Bonn's NimbRo-OP2X humanoid plays a Tsinghua robot at RoboCup 2018 in Montreal. Photo: Sven Behnke / Wikimedia Commons (CC BY-SA 4.0)

In September 2026 a Tsinghua University team led by Yushi Wang published a paper in Science Robotics on how its players work. The control system runs on an unmodified Booster T1 humanoid, was trained in simulation and uses only the robot's own cameras. It searches for the ball, chases it and kicks in several directions. Kicks from the front half of the field succeeded 80 to 90% of the time, and the robot reached and kicked the ball up to 64% faster than with rule based control. The team scored 76 goals and conceded 11 across two competitions. With this controller, the Tsinghua Hephaestus team won the adult size humanoid league at RoboCup 2025 and the football at the first World Humanoid Robot Games the same year.

Hamburg Bit-Bots robots in blue playing at RoboCup 2019. Photo: Floriv1999 / Wikimedia Commons (CC BY-SA 4.0)

ABB does not build football robots, and none of its arms plays in RoboCup. Its connection to robot football runs through education. Students who write walking and vision code for RoboCup teams often learn industrial robotics on small arms such as the ABB IRB 120. ABB sells a Robotics Education Package built around the GoFa cobot and pitches the compact IRB 1010 for schools, and both are programmed in RobotStudio. We covered a similar crossover between sport and industrial arms in our post on table tennis robots.

A teacher and students at an ABB Robotics Education Package cell with a GoFa cobot. Photo: ABB

What comes next for robot football

The Tsinghua authors are open about the limits. Their robots track only the ball and the goal, not opponents or teammates, and the paper calls that the next big hurdle. Passing, marking and team tactics all need each robot to know where everyone else is on the pitch. Hardware is improving quickly. At the Beijing games a Chinese humanoid called Tiangong ran 100 metres in under nine seconds, and many robots could get back up after falls. A team that reads a game the way human players do is still years away, and the 2050 date looks ambitious.

A class works with ABB's IRB 1010, which ABB pitches for education. Photo: ABB

For homes, the useful part of robot football is the training it gives. A humanoid that can find a rolling ball, stay balanced while kicking and get up after a fall is practising what a home robot will need around people and clutter. Factories face the same question if humanoids ever share floor space with workers. In both places safety is the open problem, because a heavy robot that topples over near a person can hurt them.

Students program a small ABB IRB 120 six axis arm. Photo: Peterantala / Wikimedia Commons (CC BY-SA 3.0)

Cost and skills limit the sport too. Teams spend a lot on robots and travel, and most players are still university projects. Each season the matches train a new group of engineers in machine vision and control, and in getting several robots to cooperate. Plenty of them go on to program robots in factories, where the same skills apply to arms on a production line. Our classroom robot post looks at how robots are used for teaching.

A DARwIn-OP humanoid, an open platform used by RoboCup teams, in a University of Minnesota lab, 2015. Photo: KaputJonny / Wikimedia Commons (CC BY-SA 4.0)

Innovation

AI solutions for effortless ABB robot control.

Automation

Robotics

ceojohntran@chatwithrobot.net

+84905311611

© 2025. All rights reserved.

qtran1215@gmail.com