Booster Robotics' T1 humanoid robot kicks soccer balls hard enough to dent walls, raising serious safety questions about powerful robots operating near people.

RoboCup is an international competition that promotes and advances robotics and AI through the challenges presented by its various leagues. We got the chance to sit down with Professor Manuela Veloso, one of RoboCup's founders, to find out more about how it all started, how the community has grown over the years, and the vision for the future. I think it would be very interesting to go right back to the beginning and hear how RoboCup got started. What was the initial idea, and how did it get set up? So we are talking about the mid-90s. In terms of the research in those days, it was the beginning of the internet and many AI and computer science researchers were focused on the internet, first on sophisticated search algorithms, on natural language understanding, on information retrieval, and then on software agents and machine learning applied to digital information. From what I recall, there was a smaller group of researchers who were interested in actual, physical robots, and in particular in AI and robotics.

Achieving coordinated teamwork among legged robots requires both fine-grained locomotion control and long-horizon strategic decision-making. Robot soccer offers a compelling testbed for this challenge, combining dynamic, competitive, and multi-agent interactions. In this work, we present a hierarchical multi-agent reinforcement learning (MARL) framework that enables fully autonomous and decentralized quadruped robot soccer. First, a set of highly dynamic low-level skills is trained for legged locomotion and ball manipulation, such as walking, dribbling, and kicking. On top of these, a high-level strategic planning policy is trained with Multi-Agent Proximal Policy Optimization (MAPPO) via Fictitious Self-Play (FSP). This learning framework allows agents to adapt to diverse opponent strategies and gives rise to sophisticated team behaviors, including coordinated passing, interception, and dynamic role allocation. With an extensive ablation study, the proposed learning method shows significant advantages in the cooperative and competitive multi-agent soccer game. We deploy the learned policies to real quadruped robots relying solely on onboard proprioception and decentralized localization, with the resulting system supporting autonomous robot-robot and robot-human soccer matches on indoor and outdoor soccer courts.

Robot decision-making in partially observable, real-time, dynamic, and multi-agent environments remains a difficult and unsolved challenge. Model-free reinforcement learning (RL) is a promising approach to learning decision-making in such domains, however, end-to-end RL in complex environments is often intractable. To address this challenge in the RoboCup Standard Platform League (SPL) domain, we developed a novel architecture integrating RL within a classical robotics stack, while employing a multi-fidelity sim2real approach and decomposing behavior into learned sub-behaviors with heuristic selection. Our architecture led to victory in the 2024 RoboCup SPL Challenge Shield Division. In this work, we fully describe our system's architecture and empirically analyze key design decisions that contributed to its success. Our approach demonstrates how RL-based behaviors can be integrated into complete robot behavior architectures.

This article introduces an open framework, called VSSS-RL, for studying Reinforcement Learning (RL) and sim-to-real in robot soccer, focusing on the IEEE Very Small Size Soccer (VSSS) league. We propose a simulated environment in which continuous or discrete control policies can be trained to control the complete behavior of soccer agents and a sim-to-real method based on domain adaptation to adapt the obtained policies to real robots. Our results show that the trained policies learned a broad repertoire of behaviors that are difficult to implement with handcrafted control policies. With VSSS-RL, we were able to beat human-designed policies in the 2019 Latin American Robotics Competition (LARC), achieving 4th place out of 21 teams, being the first to apply Reinforcement Learning (RL) successfully in this competition. Both environment and hardware specifications are available open-source to allow reproducibility of our results and further studies.

If it is true that good problems produce good science, then it will be worthwhile to identify good problems, and even more worthwhile to discover the attributes that make them good problems. This discovery process is necessarily empirical, so we examine several challenge problems, beginning with Turing's famous test, and more than a dozen attributes that challenge problems might have. We are led to a contrast between research strategies--the successful "divide and conquer" strategy and the promising but largely untested "developmental" strategy--and we conclude that good challenge problems encourage the latter strategy. More than fifty years ago, Alan Turing proposed a clever test of the proposition that machines can think (Turing 1950). He wanted the proposition to be an empirical, one and he particularly wanted to avoid haggling over what it means for anything to think.

Stepping out of the elevator on the seventh floor of Carnegie Mellon University's Gates Center for Computer Science, I'm greeted by an ungainly yet courteous robot. It guides me to the office of Manuela Veloso, who beams at the bot like a proud parent. Veloso then punches a few buttons to send it off to her laboratory a few corridors away. Veloso, a computer science professor at CMU, in Pittsburgh, has worked for over two decades to develop such autonomous mobile robots. She believes that humans and robots will one day coexist, and my robot escort, named CoBot (for Collaborative Robot), is one of her contributions to that future.

When Google's AlphaGo computer program triumphed over a Go expert earlier this year, a human member of the Google team had to physically move the pieces. Manuela Veloso, the head of Carnegie Mellon's machine learning department, would have done it differently. "I'd require the machine to move the pieces like I do," she says. "That's the world in which I live, which is a physical world." If Google can make cars that drive themselves, surely it could add robotic arms to a Go match. Even in 1997, I.B.M. could have given Deep Blue robotic arms in its match against Garry Kasparov.

When Google's AlphaGo computer program triumphed over a Go expert earlier this year, a human member of the Google team had to physically move the pieces. Manuela Veloso, the head of Carnegie Mellon's machine learning department, would have done it differently. "I'd require the machine to move the pieces like I do," she says. "That's the world in which I live, which is a physical world." If Google can make cars that drive themselves, surely it could add robotic arms to a Go match. Even in 1997, I.B.M. could have given Deep Blue robotic arms in its match against Garry Kasparov.

In a disappointing defeat against France, the human German soccer team was knocked out of the European Championship July 7. But at least the nation can take heart that its robots proved victorious in the annual RoboCup robot soccer tournament that took place this past weekend in Leipzig, Germany. The winning team, from the University of Bremen, called "B-Human," beat out the University of Texas, Austin's team--the wonderfully named "Austin Villa"--on penalties after a goalless draw in the final of the tournament, according to The Telegraph. Robot teams play two 10-minute halves on a 9-by-6-meter pitch, autonomously roving about the field trying to pass the ball to teammates and score goals. But unlike the cool efficiency that the German human soccer team is known for, the robots are a bit more ramshackle in their approach to the sport.