Humanoid robots at RoboCup have become less prone to falling and can be deployed without people following them, first reported by Robohub. But researchers said humanoid robots are rarely the best choice for applications today, and that a wheeled robot can be simpler for basic tasks such as moving a block.
This year’s international competition had 55 humanoid robot soccer teams and a demonstration game of 11 vs 11. Biswas said the trustees’ annual match against the robots, once trivial, is becoming less easy for them.
Biswas linked the gains to better hardware, algorithmic advances and training methods. Most modern humanoids use learned controllers for walking, including models trained in simulation and transferred to the real world.
RoboCup has simplified its soccer leagues and plans to have only humanoid league soccer in 2028. Biswas said integrating control and perception in one robot is worth exploring from a scientific perspective.
RoboCup’s soccer plans center on humanoids, while Biswas’s block-moving example favors a wheeled base. For a position-based task such as moving a block, he said a wheeled robot is better because its stable base and fewer degrees of freedom make planning easier; a humanoid adds forces, velocities and accelerations.
Biswas said the best advances are over-optimized for specific scenarios, and that these solutions are not robust or general-purpose; he raised a different brush or lighting as examples of changes to test. Hauert said videos can show robots that are remotely controlled or finely tuned, not machines ready for an unfamiliar environment.
Dechter said a humanoid might create a better psychological impression in healthcare assistance, while Hauert saw a possible role for humanoids in caring settings if they are not deceptive. Hauert said people may expect more from humanoids because of their form, and that less-capable cartoon-like robots might be safer in those settings.
Biswas said teleoperation, in which a human controls a robot to produce training data, is useful but insufficient on its own to deliver general-purpose robots. He argued that task complexity grows combinatorially and compared the data-collection push with self-driving companies’ earlier belief that HD maps alone would solve driving.
RoboCup’s challenge to defeat the human world champions by 2050 is, Biswas said, an open conjecture.
