Back to feed
News Story
APriority78
机器之心
1 sources

Zhiyuan Robots Wins Double First in Debut at Humanoid Robot Games

Zhiyuan Robotics, in its first appearance at the 2026 World Humanoid Robot Games, topped both the gold medal and total medal counts with 18 gold, 16 silver, and 12 bronze medals. Its mass-produced models excelled in dexterous hand, scenario-based tasks, and high-dynamic sports, demonstrating comprehensive capabilities.

SynthePulse Insight · AI deep reading

Zhiyuan Tops Both Leaderboards: The Comprehensive Capability Contest Behind Humanoid Robot Competitions

Version 1 · 1 source

At the 2026 World Humanoid Robot Games, Zhiyuan topped both the gold medal and overall medal tables with 18 gold, 16 silver, and 12 bronze medals, totaling 46. This achievement not only showcases individual strengths but also highlights deep accumulation in comprehensive capabilities.

  • Zhiyuan, in its debut, won 18 gold, 16 silver, and 12 bronze, ranking first in both gold medals and total medals.
  • All competing models were production versions, not customized for the competition, making results closer to real-world capabilities.
  • OmniHand won 7 golds in 8 dexterous hand events, demonstrating fine manipulation skills.
  • Elf G2 won 6 golds in 12 scenario events, covering real work scenarios like hotels and libraries.
  • Expedition A3 and Lingxi X2 won 5 golds and 1 gold, 1 silver, 1 bronze respectively in athletic events, showcasing whole-body movement capabilities.
  • Zhiyuan's technical architecture 'Three Intelligences in One' and core models (GO, GE, RW-RL) underpin its generalization ability.
Open section navigationEvent Overview and Results

Event Overview and Results

The 2026 World Humanoid Robot Games featured 51 events, divided into three categories: athletic competitions, dexterous hand challenges, and scenario events. Zhiyuan, in its debut, finished with 18 gold, 16 silver, and 12 bronze, totaling 46 medals, ranking first in both gold medals and total medals.

All models Zhiyuan fielded were production versions, with no custom bodies built specifically for the competition. This means the results are closer to the products' capability boundaries under real-world conditions, rather than specialized prototypes optimized for the rules.

The event organizing committee's published leaderboard shows Zhiyuan ranked first in both gold medals and total medals. This achievement spans multiple capability dimensions and holds comprehensive reference value.

Model Performance and Capability Dimensions

OmniHand won 7 golds in 8 dexterous hand events, including block building, powder weighing, and tweezers picking beans, with 3 of those golds achieved fully autonomously, demonstrating the adaptability of the same hardware across various tasks.

Elf G2 secured 6 golds in 12 operational scenario events, including hotel, indoor emergency, and library settings. Scenario events require completing multiple operational steps in sequence within 20 to 30 minutes, where task chains are long and errors accumulate easily.

Expedition A3 won 5 golds in athletic events such as dance, martial arts, and table tennis, while Lingxi X2 earned a gold in the 100m hurdles, and a silver and bronze in the 400m hurdles, showcasing whole-body coordination and high-speed movement capabilities.

Technical Architecture and Generalization Ability

Zhiyuan's technical architecture is 'Three Intelligences in One,' meaning a single body carries motion intelligence, interaction intelligence, and operational intelligence. Among these, operational intelligence receives the heaviest investment, comprising three core models: GO for task understanding, GE for reasoning about action consequences, and RW-RL for real-robot reinforcement learning.

This system brings generalization: skills learned in a hotel can transfer to a library, and Elf G2's skills can transfer to other bodies. This explains why Zhiyuan dared to compete with production models.

In motion intelligence, Expedition A3 employs a Sim2Real strategy, learning in simulation before transferring to the real robot; Lingxi X2 uses the self-developed AGILE engine, combining vision to generate motion strategies in real time.

Comprehensive Capability and Industry Insights

UC Berkeley professor Sergey Levine points out that robotics is a discipline of 'building ultra-complex integrated systems,' requiring completeness from underlying hardware to high-level decision-making. Zhiyuan's medal distribution is broad with no obvious weaknesses, reflecting such comprehensive capability.

Humanoid robot competition is shifting from single-event showmanship to system completeness. Zhiyuan's top ranking on both leaderboards is more like a comprehensive capability report card; winning a single event is easy, but being comprehensively strong without weaknesses is the truly difficult championship in the industry.

Credibility boundary

This article is based on a report from Machine Intelligence, with event results published by the organizing committee. However, specific technical details (such as model names and architecture) come from Zhiyuan's official statements or the report's retelling and have not been independently verified.

Insight takeaway

Zhiyuan's top ranking on both leaderboards is not just a competition victory but a comprehensive reflection of its 'Three Intelligences in One' technology stack and production capabilities, indicating that humanoid robot competition has entered a systematic phase.

Primary report

机器之心

Primary source