en.Wedoany.com Reported - AGIBOT is leveraging its AGIBOT World Challenge 2026 to shift the evaluation of embodied artificial intelligence from virtual environments to real-world testing scenarios. Held during ICRA 2026, the competition attracted 526 research teams and corporate teams from 27 countries.
The Shanghai-based robotics company designed the contest around a growing focus in the industry: whether AI systems that perform well in simulated environments can achieve equivalent results when deployed on physical robots. Instead of relying solely on benchmark scores generated in virtual environments, the challenge incorporates real robots, real tasks, and standardized evaluation methods.
Finalists were required to complete tasks using the AGIBOT G2 humanoid robot during the live finals held in Vienna. This approach places greater emphasis on factors such as stability, adaptability, and long-term task execution capabilities, which are critical for real-world deployment but often difficult to measure through simulation alone.
The competition is divided into two categories, targeting different aspects of embodied intelligence. The Reasoning-to-Action track examines how robots understand instructions, perceive environments, formulate plans, and execute tasks in physical settings. This represents an expansion of AGIBOT's previous manipulation-focused evaluations, broadening the assessment from simple task execution to the complete reasoning and action pipeline. The World Model track focuses on prediction and environmental understanding, requiring participating teams to build systems capable of predicting how physical environments change in response to robot actions and sensor data.
Participants came from universities, research institutions, startups, and technology companies, with over 100 teams exceeding the competition's baseline performance requirements. In the Reasoning-to-Action track, participants trained models using the AGIBOT WORLD open-source dataset and tested them via Genie Sim 3.0. The benchmark measured multiple capabilities, including language understanding, spatial reasoning, disturbance handling, atomic manipulation skills, and zero-shot transfer performance. The PrismBot team from vivo secured first place, RP-VLA from Shanghai RoboParty took second, and GreenVLA placed third.
A more practical addition to the event was the supermarket benchmark, jointly developed by AGIBOT and Dexmal. This test placed robots in a retail environment, requiring them to navigate shelves, locate products, pick up items, transport them, and place them in designated positions. Teams had to operate under real-world constraints such as shelf height limitations and random product placements. Participants controlled physical robots remotely via APIs, allowing organizers to evaluate algorithm performance under real deployment conditions. The benchmark also introduced challenges such as item drops and grasp failures, creating conditions closer to real-world robot operations.
The World Model competition produced different winners. The joint team NeoVerse-ABot from the Institute of Automation, Chinese Academy of Sciences, and Amap CV Lab took first place. The PAI@IAII team from the Institute of Industrial Artificial Intelligence, Chinese Academy of Sciences, placed second, and the Loop team from the University of Science and Technology of China placed third. This track focused on evaluating how effectively AI systems simulate and predict physical world interactions.
Beyond the competition, AGIBOT also released a broader development framework designed to support embodied AI research. The tool stack includes the AGIBOT WORLD open-source dataset, the Genie Sim 3.0 simulation platform, and the AGIBOT G2 humanoid robot platform. These tools are co-designed to help developers train, evaluate, and validate robotic AI systems from simulation to physical deployment. The company stated that resources developed through the challenge will be fed into its ongoing benchmarking and open-source initiatives, with future plans including launching online simulation leaderboards, adding new test tasks, and expanding benchmark coverage.
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
