en.Wedoany.com Reported - vivo's PrismBot robotics team won the "Reasoning to Action" track at the ICRA 2026 AGIBOT World Challenge held in Vienna. As part of the IEEE International Conference on Robotics and Automation, the competition attracted 526 teams from 27 countries and regions. The vivo team secured first place with a decisive advantage in the real-robot finals, demonstrating its technical approach's ability to translate task understanding into action decisions in complex real-world environments.
The vivo robotics team also ranked among the top three globally in the full-body control track, further showcasing the company's R&D strength in embodied intelligence. These achievements reflect vivo's strategic direction: building perception systems based on imaging technology and developing an AI-driven robot "brain" to bring intelligent capabilities into the physical world. Hu Baishan, President and Chief Operating Officer of vivo and President of vivo Central Research Institute, elaborated on this strategy at the Boao Forum for Asia Annual Conference 2026.
The AGIBOT World Challenge is one of the most demanding international competitions in the field of embodied intelligence, emphasizing real-world deployment and using rigorous real-robot testing to evaluate system performance. The "Reasoning to Action" track focuses on the core challenge of translating task understanding into action decisions, combining online simulation evaluation with offline real-robot testing in Vienna. Scoring is based on task completion rate in real environments, long-term stability, and generalization capability in complex scenarios. Participating models must independently complete intent understanding, task decomposition, sub-goal sequencing, and anomaly recovery, then translate decision processes into physical execution via robotic arms. The vivo robotics team developed a training and inference framework centered on keyframe optimization and contrastive learning. Keyframe loss weighting helps the model learn critical action points more effectively, while contrastive learning narrows the semantic gap between text-based instructions and physical action execution.
In the full-body control track, robots must autonomously pick up specified products in a realistic supermarket scenario and place them into a shopping cart, handling diverse product categories, changing spatial layouts, and complex semantic understanding and action generalization. The vivo robotics team applied the same reasoning-centric technical system to this scenario, using keyframe loss weighting to improve grasping accuracy and contrastive learning to better identify grasping directions, thereby achieving a top-three finish. This result validates the transferability and engineering robustness of vivo's technical approach across different task types.
vivo has chosen the home as the starting point for its robotics R&D, as home environments demand high levels of long-term task execution, dual-arm collaboration, fine manipulation, reasoning, and decision-making—capabilities closely aligned with those tested at ICRA 2026. The vivo robotics team is progressing from teleoperation toward autonomous intelligence, gradually enhancing the executability and scalable validation of complex tasks. Leveraging its experience in device systems, imaging capabilities, and global product development, vivo is building a robotics capability system that integrates hardware and software and continuously evolves over time.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
