The NimbRo@home team from the University of Bonn has won first place at the RoboCup@Home World Championship in Eindhoven. From July 17 to 21, the service robots competed in the Netherlands against 16 other teams, achieving the highest score in the tests. The team also impressed the jury in the final. The assistance robots, designed for everyday environments, are developed at the Autonomous Intelligent Systems Group of the Institute of Computer Science. The development is led by Prof. Dr. Sven Behnke, Lamarr Area Chair for Embodied AI.
The RoboCup@Home league is the largest annual competition for assistance robots designed to support people in everyday environments. In the open class of the world championship in Eindhoven, 17 teams from eleven countries competed. The robots demonstrated their abilities in ten tests in realistic home environments, including welcoming and introducing visitors, carrying luggage, storing groceries, cleaning the kitchen, and setting the breakfast table.
Particularly challenging were two tests where tasks were communicated through voice commands. The University of Bonn team deployed two mobile robots with human-like upper bodies and omnidirectional wheels. These robots perceive their environment using cameras, laser scanners, and microphones, and control numerous motors and a speaker to autonomously perform assistance tasks. The methods used from artificial intelligence research include image and speech understanding, action and motion planning, and dialogue systems.
In the preliminary round, Team NimbRo@home was set back by a hardware defect in the main robot, placing second behind last year’s winner, Tidyboy from Korea. In the main round, NimbRo overtook Tidyboy. In the final, the top three teams presented self-defined tasks on the theme “Assistance in Preparing Dinner.” The Bonn team used two robots simultaneously, demonstrating object recognition without predefined categories, gesture recognition, and interaction with users through a speech dialogue system. They also showed mobility in an apartment, object gripping, and integration with multimodal base models for generating recipe suggestions based on recognized ingredients. For this performance, NimbRo received the highest score from both the @Home league and industry jury members. NimbRo won the competition with 8,852 points, ahead of Tidyboy (7,495 points) and SocRob@Home from Portugal (6,901 points).
“Assistance robots will contribute in the future to allowing people in need of help to live independently for longer in their familiar environments,” said Prof. Dr. Sven Behnke. “That our team has prevailed against strong international competition highlights the excellence of robotics research in Bonn.”
The success of Team NimbRo underscores the importance of research in the field of Embodied AI at the Lamarr Institute. Embodied Artificial Intelligence refers to AI embedded in physical systems, such as robots, that can interact with their surroundings. This research brings together various disciplines, including computer vision, environment modeling, prediction, planning and control, reinforcement learning, physics-based simulation, and robotics. Embodied AI enables robots to learn and continuously improve their behavior through exploration and interaction with their environment.
