"Robot+" total prize pool reaches up to $530,000! Zhiyuan hosts another robotics competition, embodied intelligence kicks off the "Model Battle"

This report (chinatimes.net.cn) was written by reporter Shi Feiyue from Beijing.

“Our original intention for organizing this competition was not primarily for commercialization; our greatest wish is to promote the development of the entire embodied intelligence industry ecosystem through this format,” stated Shen Yongjian, Director of Ecosystems and Solutions at Zhiyuan Genie, to media including reporters from the Huaxia Times on March 25. Recently, the AGIBOT WORLD CHALLENGE ICRA2026, hosted by the company, has opened servers for two competition tracks and will hold an offline finals with real machines on June 1.

In the past two years, with the iteration of AI technology and robotic body technology, a wave of competitions has been spawned: robots running marathons, fighting, playing soccer… various stages for “showing off muscles” have emerged, but Zhiyuan’s chosen path is somewhat different, focusing the competition on the model level. However, in the eyes of industry insiders, no matter how the form of the competition changes, the underlying logic remains clear: technology and ecosystem are the stories on the surface, while exposure and influence are the true goals beneath.

Model capabilities urgently need improvement.

Previously, the global humanoid robot free fighting league initiated by Zhongqing announced a ten-kilogram, approximately 10 million yuan pure gold belt for the final champion team, and now Zhiyuan has set a total prize pool of up to $530,000, indicating that manufacturers are willing to invest in this matter.

The AGIBOT WORLD CHALLENGE ICRA2026 is divided into two tracks: Reasoning to Action and World Model. The Reasoning to Action track aims to evaluate the model’s reasoning and action execution capabilities, including online simulation and offline real machine segments, with competition topics designed in conjunction with practical projects and the Genie Sim 3.0 simulation platform, focusing on common scenarios such as logistics, industry, supermarkets, catering, and home environments; the World Model track focuses on the core capabilities of embodied world models, specifically accurately modeling physical environment dynamics based on robot actions. Participants must train video generation models based on the AGIBOT WORLD dataset, generating interactive videos of robots in 10 real operational scenarios based on actual observations and action signals.

For the robotics industry, models are currently the most pressing shortfall that needs to be addressed, with many manufacturers shifting their R&D focus to embodied intelligence models: The prospectus recently disclosed by Yushu Technology shows that its core technology “General Humanoid Robot Embodied Large Model” is mainly composed of two technical directions: “World Model-Action (WMA) Large Model” and “Vision-Language-Action (VLA) Large Model”; on March 25, Xinghai Tu released research results on the World Model called Fast-WAM, which completely abandons the traditional inefficient world model paradigm of “imagine first, execute later”.

Recently, Yushu Technology founder Wang Xingxing stated that in the future, when a robot large model can complete 80% of tasks in about 80% of unfamiliar scenarios solely through language and text instructions, that moment will mark the “GPT moment” for embodied intelligence.

At Zhiyuan, Shen Yongjian revealed that the world model has always been an important research direction for the company, which has invested numerous excellent scientists and engineers to continue exploring. From a strategic priority perspective, its importance is on par with VLA and related technical routes. “Our core goal is to create a truly coherent and effective world model in the field of physical intelligence. The key ‘nourishment’ for achieving this goal is real machine data from real scenarios. Therefore, this year, while increasing investments on the algorithm side, we will accumulate one or even two orders of magnitude more data on the data side to lay a solid foundation for the leap in the capabilities of the world model.”

Shen Yongjian also candidly stated that talent in the entire embodied intelligence industry is still relatively scarce, especially high-quality and top-tier algorithm talent. This competition will definitely allow some teams to shine, and Zhiyuan hopes to attract some industry talent to the company through the competition.

In addition, “We hope to allow more people to experience the entire process from data, models to real machine implementation through the competition and witness how robots solve real problems in real scenarios. With such firsthand experience, they will have a more fundamental understanding of the development and application of embodied intelligence. This is exactly the original intention behind our competition,” Shen Yongjian told our reporter.

Intensive emergence of robot competitions.

In the tech field, whether it’s embodied intelligence, AI large models, or the previous AI wave, competitions, contests, and human-robot battles have always played an important role as a driving force.

However, according to Wang Chao, founder of the Wenyuan Think Tank, aside from promoting technological development, from a communication logic perspective, competitions first create exposure opportunities, allowing society to form a deeper understanding of the technology. This deepening of cognition will ultimately transmit to the financing level—affecting investment decisions in primary and secondary markets, and even positively influencing policy directions.

“This time is no different; whether it’s the model competition hosted by Zhiyuan, the robot marathon in Yizhuang, or various robot fighting contests, these activities are all aimed at eliciting attention from the industry and the public, thereby attracting investors and even global attention,” Wang Chao told reporters from the Huaxia Times.

Starting in 2025, robot competitions have become a major avenue for exposure for various manufacturers, including last April’s 2025 Beijing Yizhuang Half Marathon and Humanoid Robot Half Marathon, last May’s CMG World Robot Competition series fighting, and last August’s global first comprehensive competition event for humanoid robots—the World Humanoid Robot Sports Conference. Last month, the inaugural Global Humanoid Robot Free Fighting League initiated by Zhongqing officially kicked off.

The impact of robot competitions on robot manufacturers is immediate: after finishing as the runner-up in the half marathon, Songyan Power quickly secured over 100 million yuan in robot orders; in the 2025 RoboCup Brazil Robot Soccer World Cup humanoid category, both the champion and runner-up used accelerated evolution robots, which also boosted sales of the company’s robots.

Of course, as Wang Chao mentioned, the value brought by robot competitions goes far beyond just competition economics and product sales; it serves as a technical “touchstone,” a capital “shot in the arm,” and an industrial “connector.”

Zhiyuan’s capital layout in the field of embodied intelligence is already quite forward-looking: on one hand, the company has gained control of an existing capital platform through the acquisition of Shangwa New Material; on the other hand, Zhiyuan completed a share reform last year, which the industry interprets as preparation for future listing. In addition, the company itself has completed multiple rounds of financing.

However, Zhiyuan’s ambitions are large, and its steps are also significant. From robot bodies to robot leasing, and to robot data, Zhiyuan has incubated many upstream and downstream “little brothers”—which means that the company also has a huge demand for funds.

From the overall layout regarding embodied intelligence, Zhiyuan’s performance is undoubtedly proactive. However, Yuan Bo, a specially invited expert at Zhican Think Tank and a specialist in the embodied robotics industry, reminds that Zhiyuan is both a complete robot enterprise and a component supplier, which may raise doubts about the competitiveness and reliability of Zhiyuan as a component supplier for other robot companies in the future. “At the same time, efficiently integrating these independent ‘modules’ into complete machines to form a synergistic effect of 1+1>2 also carries risks of adjustment.”

Editor: Huang Xingli Chief Editor: Han Feng