> Zhichai AI · Embodied Intelligence · August 27, 2026
Source: China.com interview with Zhishen Robotics co-founder Liu Yulong, WRC 2026
One-Line Warning
While the rest of the industry shows off demos, Zhishen Robotics co-founder Liu Yulong brought a set of numbers every peer can read: 15,000+ units cumulatively mass-produced, 5,000+ units of monthly capacity, thousands of security deployments, 100+ power inspection projects delivered, and coverage of 100+ fire brigades.
At WRC 2026, Liu Yulong played a completely different card: "Build running legs first, then install working hands." The company focused first on quadruped robots because two pieces of the technology puzzle—all-terrain mobility and spatial navigation—are already complete. Once these modules run reliably in real scenarios like security, emergency response, and power inspection, the company will return to fill in manipulation intelligence for humanoids.
This路线 mirrors the WHRG 2nd World Humanoid Robot Games: WHRG is embodied intelligence on stage (records, autonomous tasks, assembly), while Zhishen is embodied intelligence in the real world (24-hour community patrols, fire brigade solutions, routine substation inspection).
Symptoms of the "Demo Bubble": 100,000 Units on Slides vs. 100 Units in Reality
At WRC 2026, the industry showcased over 400 robot models with claimed annual output targeting 100,000 units. Unitree chairman Wang Xingxing was blunt: "In fixed scenarios, task success rates approach 100%; change the object or environment and success rates collapse." This is the "demo bubble": prototype explosion, mass-production slide decks, and choreographed performances that fail the moment they leave the showroom.
Zhishen's booth was not a demo hall. It showed:
- Quadrupeds climbing open-frame steel staircases;
- Robotic arms performing precise operations in a simulated work zone;
- Joint modules running continuously 1 meter underwater.
- Power: P85MAX-S joint module drives 90 kg load at ~1 kg weight; Gangben L2 achieves a 7:1 payload-to-weight ratio; humanoid YinYi NE01 reaches 180 N·m peak torque per joint.
- Environmental adaptation: Second-gen Qilin alloy joints with IP66 industrial outdoor rating; continuous operation 1 m underwater; autonomous climbing of open-frame industrial ladders.
- Intelligence: A "brain + cerebellum" architecture—importing the smart-vehicle "rules stack + end-to-end" approach into robot navigation, plus a perception-control unified model for autonomous climbing.
- Upstream: In-house CHAMP high-power-density joint modules, annual capacity over 1 million units, supplying core actuation components to the industry.
- Downstream: 300+ integrators, universities, and industry partners co-building scenarios.
- Security: Thousands of quadrupeds in routine patrol at residential communities, business districts, industrial parks, and coastlines; leading security integrators are all partners. One robot replacing a night guard saves roughly ¥5,000/month; at scale, deployments imply tens of millions of yuan in monthly labor savings.
- Emergency: 100+ fire brigades/detachments in Beijing, Tianjin, Gansu, etc., using a "robot dog + drone / off-road vehicle" multi-robot solution across 20+ product models covering explosion, toxic, smoke, and heat scenarios.
- Power: 100+ inspection projects delivered across substations and energy storage stations—storage being one of the fastest-growing segments, with heat, humidity, and toxic gas leakage risks that suit robots well.
Liu: "We are not 'planning to enter'—we are already running in real scenarios."
The "Lay Eggs Along the Way" Logic: 15,000 Units Is Not a Slogan
Most peers bet everything on a perfect future humanoid, with no intermediate revenue and constant burning of cash. By 2026, investors demand real self-sustaining revenue. Zhishen decomposes embodied intelligence into modules—all-terrain mobility, spatial cognition + autonomous navigation, and manipulation intelligence. The first two already enable quadrupeds to handle tedious, repetitive, dangerous work; each time a technology matures, it goes into the real world to generate revenue and fund further iteration.
Hard data as of June 2026:
| Dimension | Data | |---|---| | Cumulative production | 15,000+ units | | Monthly capacity | 5,000+ units | | In-house joint modules | 1 million units / year | | Security deployments | Thousands (property, business districts, scenic areas, coastlines) | | Power inspection projects | 100+ | | Emergency product matrix | 20+ models | | Fire brigade coverage | 100+ | | Ecosystem partners | 300+ | | Provinces/cities covered | 100+ |
The "Three Gates": From "Can Build" to "Can Use Well"
Gate 1 — Technology: From "Works Once" to "Stably Repeatable"
Customers care whether the same task can be done repeatedly in similar sites. Zhishen tracks IP66 protection, endurance (Gangben L2: 150 kg static load, 16 kg sustained carry), load capacity (P85MAX-S drives 90 kg at ~1 kg module weight), joint reliability, thermal rise, fall recovery, and long-term maintenance.
Gate 2 — Scenario: From "Getting There" to "Finishing the Task"
Arriving is only half; real deployment requires defining what to inspect, how to collect data, how to handle anomalies, and how results enter customer workflows—organizing body, navigation, payload, communication, algorithms, and business rules into a task a customer can define, verify, and accept. This maps to Zhishen's "chain leader + module output" model: 1 million joint modules/year, with 300+ integrators handling scenarios.
Gate 3 — Sustained Delivery: From "One Project" to "Reusable Product Capability"
Recurring requirements are folded back into products via standard interfaces, SDKs, modular extensions, and version iteration.
Strength in Three Dimensions: Power + Environment + Intelligence
Business Model: Chain Leader + Module Output
This separates building robots from using robots: the embodied-AI company does the hard engineering of making them; integrators do the hard work of deploying them.
Three Verticals with Verifiable Numbers
"Lay Eggs Along the Way" vs. "Wait for Perfect"
| Dimension | Wait for Perfect | Lay Eggs Along the Way | |---|---|---| | Revenue start | After general humanoid matures | Now, quadrupeds first | | Unit cost | High (R&D on few units) | Medium (mass production) | | Customer validation | Closed R&D → one-shot | Continuous feedback → iteration | | Cash flow | Constant fundraising | Revenue funds R&D | | 6 months later | Lab prototypes | Tens of thousands running in the field |
Liu on why robots deserve to exist: protect lives by letting machines go in first; fill gaps in tasks humans cannot cover long-term and consistently; raise efficiency by turning experience-dependent work into standardized, traceable tasks; and give sensors and algorithms a mobile carrier into the real world.
The Real Gap Is Not Technology but Standards
Liu's overlooked judgment: "The industry still lacks a mature standard—performance metrics, functional metrics, evaluation systems, certification." No standards means no scale; no scale means no cost advantage; no cost advantage means no real deployment. Whether China's embodied AI industry moves from private standards to industry-wide standards in the next 12–18 months is the key variable in moving from "PPT valuations" to "revenue valuations."
Key Takeaways
| Dimension | Zhishen | Demo-route peers | |---|---|---| | Product form | Quadruped first, humanoid later | All-in on humanoid | | Production | 15,000+ / 5,000 per month | 100K on slides, hundreds real | | Scenarios | Security + emergency + power | Showroom performances | | Customers | Integrators + fire brigades | Investors + media | | Revenue | Rolling | Fundraising | | Moat | 1M joint modules/year | Concepts + stories |
References
1. China.com, "Dialogue with Zhishen Robotics co-founder Liu Yulong," 2026-08-27 — https://news.china.com/socialgd/10000169/20260827/49701024.html 2. Sina Finance, "Zhishen Robotics empowers 'robots that really work,'" 2026-08-26 — https://finance.sina.cn/2026-08-26/detail-iniprqcu2379993.d.html 3. Tencent, "Zhishen's Liu Yulong: embodied AI mass production needs industry standards," 2026-08-26 — https://new.qq.com/rain/a/20260826A05XJU00 4. NetEase, "15,000 units produced: why Liu Yulong's 'robots that work first' resonates," 2026-08-26 — https://www.163.com/dy/article/L59Q6FQN0556C2XD.html