智元主办AGIBOT WORLD CHALLENGE 2026收官 比赛结果公布

This article covers the conclusion of the AGIBOT WORLD CHALLENGE 2026, an embodied intelligence competition hosted by Zhiyuan alongside ICRA 2026, where results across all tracks have been officially released. Key takeaways are summarized below:

1. Event overview: The competition drew 526 industry-university-research teams from 27 countries and regions worldwide, with three core technical tracks: Inference & Manipulation, World Model, and Whole-Body Control. It adopted a format of standardized online evaluation plus on-site real robot validation, with the organizer providing open-source data, a simulation platform and robot hardware as development infrastructure for participants.

2. Competition results: The World Model track was won by a joint team from the Institute of Automation, Chinese Academy of Sciences and Amap. The Inference & Manipulation track champion was the vivo team, while the Xiaomi Robotics team took first place in the Whole-Body Control track. More than 100 teams passed the official benchmark assessment threshold.

3. Upcoming plans: Zhiyuan will launch a permanent simulation ranking server, continuously iterate evaluation benchmarks and toolchains, open up relevant resources for industry-wide collaboration, and lower development barriers for the sector.

This competition provides valuable insights on technology trends and brand positioning for brands active in the humanoid robot industry, summarized below:

1. Technology and consumer trend reference: The core R&D focus of the current embodied intelligence humanoid robot industry lies in three areas: inference and manipulation, world model, and whole-body control. A key industry pain point is that simulation training results cannot directly transfer to real-world environments, and task execution stability in real scenarios has become the core metric for technology commercialization and product competition.

2. Brand marketing reference: As an official ICRA co-hosted competition that attracts top global industry-university-research teams, competing and placing well in this event allows brands to quickly build their technical brand influence and connect with cutting-edge global R&D resources.

3. R&D cost optimization reference: Zhiyuan has opened up public R&D resources including open-source data, a simulation platform and real robot hardware. Brands can leverage these resources to cut their own algorithm R&D and testing costs, and accelerate product commercialization.

For sellers operating in the humanoid robot track, the key opportunities and insights from this competition are summarized as follows:

1. Track opportunity outlook: Embodied intelligence humanoid robots have now entered the validation and commercialization phase. Mobile manipulation requirements for real consumer scenarios such as retail stores have become a core testing direction, and commercial products targeting these scenarios offer clear growth opportunities. Sellers that lay out relevant technology early can capture first-mover advantage.

2. Accessible resources: Zhiyuan has opened up resources including an open-source dataset, the Genie Sim 3.0 simulation platform, and the Genie G2 real robot hardware, with a permanent simulation ranking server set to launch. Small and medium-sized sellers can use these public resources to avoid the high upfront cost of building in-house testing environments, significantly lowering R&D barriers and speeding up algorithm iteration.

3. Technology direction guidance: The core barrier to commercialization today is transfer stability from simulation results to real environments. Sellers should prioritize improving robots' adaptability to complex and dynamic physical environments, which is the key opportunity for small and medium-sized sellers to outpace incumbents.

For factories across the humanoid robot supply chain, key insights from the competition are as follows:

1. Product design and manufacturing demand: Humanoid robot R&D is increasingly reliant on standardized real robot testing, driving growing demand for standardized humanoid robot hardware bodies and unified hardware bases. The competition uniformly adopted the Genie G2 humanoid robot as the testing platform, confirming that standardized hardware is a common industry R&D requirement.

2. Accessible business opportunities: Industry demand for open-source testing resources and public R&D platforms is booming. Zhiyuan will continue iterating its full-stack toolchain and open up ecosystem collaboration around hardware, data and simulation platforms. Factories specializing in component production and complete robot assembly can partner with leading firms to fulfill ecosystem demand and secure stable long-term orders.

3. Digital transformation insights: Humanoid robot R&D has formed a full-stack digital R&D workflow spanning "data – simulation – real robot testing". Factories can adopt this standardized R&D model to advance the digital upgrade of their own production and R&D, reduce R&D and testing costs via unified tools, and improve new product iteration efficiency.

For service providers in the embodied intelligence humanoid robot sector, key industry insights from the competition are as follows:

1. Industry development trends: Embodied intelligence evaluation has evolved from focusing on single-model metrics in the early stage to multi-dimensional assessments covering task execution stability in real scenarios, adaptability to physical environments, and long-horizon task completion capability. Industry demand for standardized, reproducible, cross-comparable evaluation services is growing rapidly, creating significant market opportunity.

2. Core customer pain points: A pervasive pain point in humanoid robot R&D today is high overall development costs. Small and medium-sized teams and startups struggle to access standardized real robot testing environments. In addition, there is a large performance gap between simulation testing results and real robot operation, making the transfer of algorithms from simulation to real robots extremely challenging, a problem that plagues most R&D teams.

3. Viable solution directions: Service providers can build standardized public testing service platforms aligned with the full "data – simulation – real robot" R&D and evaluation workflow, to provide low-cost on-demand testing and validation services for small and medium-sized R&D teams, directly addressing current industry pain points.

For industry platform operators in the robotics sector, operational insights from this competition are summarized as follows:

1. Current core industry demand for platforms: R&D teams globally need unified, reproducible full-stack R&D and testing platforms covering the entire workflow from data to simulation to real robot testing to lower development barriers. Currently, the market lacks sufficient supply of standardized public platforms meeting these requirements, creating room for strategic layout.

2. Referenceable platform operation model: Zhiyuan's competition combined automated online evaluation with offline real robot validation, used two self-developed evaluation benchmarks and unified testing hardware, effectively solving the long-standing problem of disconnect between simulation evaluation and real-world deployment. This evaluation model is a valuable reference for platforms.

3. Guidance for future operations: After hosting competitions, platforms can convert accumulated resources into permanent operations, launch public ranking servers, continuously add new testing tasks and diversified evaluation benchmarks, and attract more industry-university-research teams to join and enrich the ecosystem. Platforms should also design tasks aligned with real-world commercial scenarios to avoid the common pitfall of overprioritizing simulation metrics over actual deployment capability.

For researchers in the embodied intelligence field, key research insights from the competition are as follows:

1. Latest industry trends: Global embodied intelligence R&D currently focuses on three core areas: decision-making from task understanding to action execution, interactive predictive modeling of the physical world, and whole-body coordinated control for continuous robot movement. All three correspond to core technical links for humanoid robot commercialization. The entire industry has shifted from pure technology demonstration to focusing on real-scenario deployment capability, and industry-university-research collaboration has become an increasingly common R&D model.

2. Newly identified core challenges: The core difficulty in embodied intelligence evaluation today is that high performance in simulation does not equate to stable performance on real robots. Numerous uncontrollable variables in real environments, such as object position shifts, grasping errors and occlusion, all reduce task completion rates, and a standardized full-stack evaluation system is still under development.

3. Guidance for future research: The evaluation benchmarks, open-source datasets, simulation platform resources and hybrid online-offline evaluation model open-sourced for this competition provide a reusable foundational framework for future research, and also highlight core research directions including simulation-to-real transfer and real-scenario task stability for researchers to explore.