On August 18, according to Zhiyuan Robotics' official WeChat account, the company's full product series went on sale on both Zhiyuan Mall and JD.com, including the OmniHand dexterous hand, Expedition A2 humanoid robot, D1 EDU robot dog, and Juechen C5 commercial robotic vacuum cleaner.

On August 18, related concept stocks surged again with daily limit gains. Xiangming Intelligence, Qiangrui Technology, and Zhongke Haixun hit 20% daily limits. Additionally, over 40 stocks including Kersen Technology, Kaizhong Precision, Lingyi iTech, Guoji Jinggong, Jintuo Holdings, and Zhiwei Intelligence either hit daily limits or gained over 10%.

**Priced at 198,000 Yuan, Currently Out of Stock**

Today, when a reporter from 21st Century Business Herald inquired with customer service at Zhiyuan's official JD.com store, they learned that the newly launched Expedition A2 Youth Edition humanoid robot is "currently priced at 198,000 yuan, primarily used for entertainment and commercial performances for display purposes. However, there's currently no stock available - it's on pre-order with an estimated delivery time of 60 days after payment."

Regarding specific parameters, the A2 Youth Edition stands 169cm tall, weighs 63kg, has a maximum torque of 270N.m, maximum arm payload of about 2kg, walking speed of 0.5-0.8m/s, charging time of 2 hours, and standing endurance of approximately 4.5 hours.

After purchase, customers receive five complimentary movements including "waving," "handshaking," and "Tai Chi." Additionally, customers can purchase seven other movements including Nezha dance, pom-pom dance, and drumming.

**Should Humanoid Robots Work First or Perform First?**

Actually, many manufacturers are currently conducting humanoid robot factory training. On August 11, Zhiyuan Robotics disclosed that it recently reached a multi-million yuan project cooperation with Fulin Precision Engineering, with nearly 100 Expedition A2-W units to be deployed in Fulin's factory for tasks such as container handling.

For robots, performances may be a pathway to quickly gain public recognition. For instance, Unitree Robotics gained widespread popularity through its Spring Festival Gala dance performance.

A representative from Fourier also publicly stated to media that "technological innovation is inseparable from application scenarios. Dancing is also a way of presenting technology, and it's a mode that the public can quickly 'understand.'"

Comparing prices, Unitree's JD.com official store currently sells two humanoid robot models - G1 and G1 EDU, priced at 99,000 yuan and 169,000 yuan respectively.

Customer service at Unitree's JD.com store told reporters that "G1 currently only supports simple walking, forward and backward movement, rotation and translation, handshaking and waving movements. Only EDU and higher versions can be developed. Other versions like PRO and AIR do not have open secondary development interfaces. The EDU version currently offers optional 40tops or 100tops computing expansion docks for development."

Meanwhile, reporters learned that Unitree's G1 humanoid robot can be used for commercial performances but doesn't include complimentary movements - secondary development is required.

In fact, the "brain" technology roadmap for humanoid robots has not yet converged. Due to the lack of unified task decomposition frameworks, manufacturers typically need to independently train specialized models for single actions (such as grasping or walking) during R&D.

Notably, on August 14, Zhiyuan Robotics released the industry's first open-source robot world model platform - Genie Envisioner (GE), which has attracted capital market attention.

Unlike traditional fragmented "data-training-evaluation" pipeline models, GE integrates future frame prediction, policy learning, and simulation evaluation for the first time into a closed-loop architecture centered on video generation, enabling robots to complete end-to-end reasoning and execution from "seeing" to "thinking" to "acting" within the same world model.

Based on 3,000 hours of real robot data, GE-Act significantly outperforms existing state-of-the-art methods in cross-platform generalization and long-sequence task execution.

Additionally, this model differs from mainstream VLA (Vision-Language-Action) methods that rely on vision-language models to map visual inputs to language space for indirect modeling. GE directly models robot-environment interaction dynamics in visual space.

Benefiting from this approach, GE-Act requires only minimal data to achieve cross-platform migration. On entirely new robot platforms such as Agilex Cobot Magic and Dual Franka, GE-Act achieved high-quality task execution using just 1 hour (approximately 250 demonstrations) of teleoperation data.