The top news stories from Asia and the Pacific
Provided by AGP
Global Showdown: 526 Teams from 27 Countries Compete in the AGIBOT WORLD CHALLENGE @ ICRA 2026 Online Round
Shanghai, May 13, 2026 (GLOBE NEWSWIRE) -- Global Showdown: 526 Teams from 27 Countries Compete in the AGIBOT WORLD CHALLENGE @ ICRA 2026 Online Round
On April 30, the results of the AGIBOT WORLD CHALLENGE @ ICRA 2026 online round were officially announced. Hosted by AGIBOT, the competition was open to global research teams working on embodied intelligence, robotic foundation models, and world models. It attracted 526 teams from 27 countries and regions, including top universities, research institutions, technology companies, startups, and individual developers, establishing itself as a premier international competition platform in the fields of embodied intelligence and world models.
As one of the most prestigious academic conferences in robotics, ICRA has long brought together leading robotics research institutions, university laboratories, and industry pioneers worldwide. Held under the ICRA framework, the AGIBOT WORLD CHALLENGE provides a more professional, fair, and verifiable competition platform. It also reflects AGIBOT’s deep engagement with the global robotics research community and highlights its international influence in shaping evaluation standards for embodied intelligence.
Industry’s Only Dual-Core Technical Tracks:
From “Executing Tasks” to “Understanding, Predicting, and Acting”
Unlike previous competitions that largely focused on offline algorithms or single-capability evaluations, AGIBOT, drawing on its deep insight into the evolution of embodied intelligence, pioneered the industry’s exclusive dual-core technical tracks: the Reasoning to Action track and the World Model track. The two tracks address two key challenges in the evolution of robotic capabilities: how robots translate task understanding into executable actions, and how they develop the ability to predict changes in the physical world.
The Reasoning to Action track focuses on evaluating a model’s capabilities in logical reasoning, step decomposition, and action decision-making in complex tasks. Rather than assessing whether a single action can be completed, it examines whether a robot can form a more reasonable action sequence based on environmental information and task objectives. The World Model track focuses on a robot’s ability to model scenes, object interactions, and future states. For robots operating in open environments, recognizing the current state is not enough; the key is determining “what will happen next” and adjusting strategies accordingly.
Together, the two tracks point to a core challenge in the real-world application of embodied intelligence: robots must evolve from passively executing instructions into autonomous systems capable of understanding their environment, planning tasks, predicting outcomes, and completing closed-loop actions.
Global Teams Compete as Online Round Rankings Are Released
Competition in this year’s online round was intense. In the Reasoning to Action track, Team GreenVLA from, Sber Robotics Center ranked first with a score of 0.885. Team SynapX placed second with a score of 0.848. Team RP_VLA from RoboParty ranked third with a score of 0.847, while Team Aether AI from the United States placed fourth with a score of 0.846. Also entering the top 10 were the individual developer team robot_level2, Team Embodied horse from South China University of Technology, Team EchoMind Master from Pyromind Dynamics, Team kp-robot from Ucap, vivo’s Team PrismBot, and Team No.89757 from Futuring Robot.
Competition in the World Model track was equally fierce. Team NeoVerse-ABot, jointly formed by the Institute of Automation, Chinese Academy of Sciences and Amap, topped the leaderboard with a score of 0.829. Team PAI@IAII from the Institute of AI for Industry, Chinese Academy of Sciences, Team Loop from the University of Science and Technology of China, Team Wild Path from North China Electric Power University, and Team VIPL-GENUN from the Institute of Computing Technology, Chinese Academy of Sciences ranked second through fifth. In addition, Team PrimeMind from Chongqing University, Team WorldCrafter jointly formed by Tianjin University and University of Chinese Academy of Sciences, Team The One from Alibaba Inc. and Team SHU from Shanghai University, as well as Team AmapWorld, also delivered strong performances to enter the global top 10.
From the composition of the rankings, this online round demonstrated remarkable diversity. Universities, research institutes, technology companies, startups, and individual developers competed on the same stage, showing that embodied intelligence is no longer an isolated academic direction or industrial application, but an emerging frontier attracting worldwide technological focus and investment..
From “Competition Arena” to “Benchmark”:
Defining New Industry Standards for Robotic World Models
The robotics industry has long faced a gap between simulation and real-world deployment: AI models often perform well in simulation environments but show performance degradation and limited generalization when deployed on physical hardware and practical scenarios. To address this industry bottleneck, AGIBOT introduced an integrated full-stack toolchain for the AGIBOT WORLD CHALLENGE @ ICRA 2026, combining real-world robot data, simulation assets, evaluation benchmarks, and baseline models.
This online round adopted a fully automated evaluation system. Powered by AGIBOT’s self-developed world model benchmark EWMBench, the simulation benchmark GenieSimBenchmark , and open-source datasets, it standardized the evaluation process and ensured reproducible results, addressing the industry’s long-standing challenges of inconsistent evaluation metrics and difficult cross-comparison. AGIBOT also open-sourced the Genie Sim 3.0 simulation development platform and the ACoT-VLA reasoning-enhanced baseline model, providing global researchers with a lower-barrier environment for development, training, and validation.
The core value of this full-stack toolchain lies in moving beyond the traditional approach of judging performance purely by offline algorithm scores, enabling competition results to better reflect robots’ real task execution capabilities. Based on a unified technical platform and transparent automated evaluation rules, different technical approaches from global teams can be fairly compared, scientifically validated, and rapidly iterated.
With the online round results announced, the competition will now move into preparations for the offline finals. This June, finalist teams will compete in Vienna on robotic tasks that more closely resemble real-world deployment.. Compared with the online round, the offline finals will place higher demands on model stability, generalization, task execution quality, and adaptability to real-world scenarios.
Yao Maoqing, Partner, Senior Vice President, and President of the Embodied Intelligence Business Unit at AGIBOT, said that the AGIBOT WORLD CHALLENGE @ ICRA 2026 aims to be more than a competition. It seeks to place the rapidly evolving key capabilities in embodied intelligence on a more open, professional, and reproducible international platform for validation. By combining the competition, benchmarks, toolchains, and open-source models, AGIBOT hopes to enable developers from different countries, institutions, and backgrounds to exchange technologies, validate solutions, and identify problems under the same set of rules, jointly helping the industry form clearer technical consensus.
Advancing research and setting standards through competition. AGIBOT WORLD CHALLENGE @ ICRA 2026 is gradually establishing a quantifiable, generalizable industry evaluation benchmark for robotic world models, oriented towards real-world deployment. Looking ahead, AGIBOT will continue working with the global academic community, industry partners, and developer ecosystem to accelerate the large-scale deployment of core embodied intelligence technologies from the laboratory to the real world.
William Peng media@agibot.com
Legal Disclaimer:
EIN Presswire provides this news content "as is" without warranty of any kind. We do not accept any responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright issues related to this article, kindly contact the author above.
