WALL-B Live Demo: 1,816 Items/Hour, Redefining the Value Anchor of Embodied Intelligence
On August 12, 2026, Zhiyuan Robot (自变量机器人) held a livestream of fully autonomous logistics sortation with no human fallback. A wheeled robot with dual mechanical arms and standard grippers sorted 1,816 packages per hour for one continuous hour on a conveyor receiving randomly mixed parcels, with 98% accuracy. On the same metric, Figure AI's Figure 03 averaged 1,248 items/hour across 200 cumulative hours — Zhiyuan set a record using less than 1/200th of the time, widening the efficiency gap to 45%. This was not a muscle-flexing launch event, but a redefinition of the industry's value anchor: "humanoid + dexterous hand" is no longer the default optimal solution — "the cheapest structure that can work reliably" is.
Across-the-Board Lead on Three Dimensions
| Dimension | Figure 03 | Zhiyuan approach | Gap | |---|---|---|---| | Efficiency | 1,248 items/hour (200-hour average) | 1,816 items/hour (1-hour live peak) | +45% | | Hardware cost | ~$85,000 BOM | 70%+ lower than Figure | -70% | | Scenario complexity | Highly uniform parcels | Random mix of boxes, foam containers, cylinders, soft bags | Order-of-magnitude jump |
The gap stems not from arm count but from the brain. Figure 03 uses the Helix VLA architecture, stitching together separate vision, language, and action modules — data loses information at each module boundary, leaving the robot's physical understanding at the "recognize → assemble commands → execute trajectory" level. Zhiyuan uses WALL-B (released April 2026), built on the World Unified Model (WUM) architecture, jointly training vision, hearing, language, touch, and action in a single network.
Notable live-demo details:
- Facing two stacked parcels, the robot did not grab directly; it first assessed occlusion, separated the parcels, then handled each individually
- Soft bags were nudged, unfolded, and flattened before surface-position calibration
- Heavy boxes triggered dual-arm cooperative handling, pushing from the side to reduce damage risk
- Natively multimodal: vision, language, and action jointly trained in one network; module boundaries disappear
- Physical-law prediction: can predict causal effects of actions on objects and the environment
- Zero-shot generalization: for unseen boxes, bags, and irregular items, it composes strategies on the fly from existing physical knowledge
- Dec 2023: Zhiyuan Robot founded in Shenzhen
- 2024: Released WALL-A
- Apr 2026: Released WALL-B, the first WUM-architecture model; announced B-round led by Xiaomi strategic investment
- Apr–Jun 2026: Completed B+, B++, and C rounds consecutively; post-money valuation exceeded RMB 20 billion
- Investors include Meituan (A lead), Alibaba (A+ lead), ByteDance (A++ lead), Xiaomi (B lead) — reportedly the only Chinese embodied-AI company with all four internet giants leading separate rounds — plus China Mobile, the National AI Industry Investment Fund, Sequoia China, IDG, Source Code, Fortune Capital, CICC, and 30+ others
- Founders: CEO Wang Qian (Tsinghua EE, among the earliest researchers to introduce attention mechanisms into neural networks; PhD in Robot Learning at USC); CTO Wang Hao (Peking University computational physics PhD, former head of IDEA Institute's Fengshenbang LLM team)
- Home: robotics cleaning service with 58.com at RMB 149 per session; a "robots entering households" program has put WALL-B-powered robots in real homes
- Industrial: entering a BMW Brilliance seat-and-interior supplier's production line; partnering with SF Express on embodied foundation models in industrial logistics
- When the embodied robotics industry drops its "look human" obsession - 163.com
- 1,816 items/hour, 45% more efficient than Figure AI - new.qq.com
- Chinese embodied AI sets global record at 30% of Figure AI's cost - news.qq.com
- Zhiyuan Robot Technology (Beijing) Co., Ltd. - Baidu Baike
- Valuation over RMB 20 billion, four funding rounds in two months - leaderobot.com
- Zhiyuan: end-to-end technical breakthrough - finance.sina.com.cn
- RMB 149 cleaning: robots get taught by ordinary households - sports.sohu.com
This on-the-spot judgment was not pre-scripted trajectories, but policies generated in real time by WALL-B after perceiving physical regularities.
Why This Route Wins
This is not a "cheaper than Americans" story, but one of the ceiling trade-off between hardware complexity and software capability.
A five-finger dexterous hand has 16 degrees of freedom, each requiring precision motors, sensors, and control — dexterous hands alone account for 14%–18% of a humanoid robot's BOM, the first bottleneck for mass production. Logistics sortation only requires "pick up and place correctly" — no walking, no fine 16-DOF manipulation. Two arms with standard grippers, grippers swapped in minutes ("like an F1 pit stop"), and a whole production line adapts in 30 minutes.
But simplifying hardware requires a strong enough brain — WALL-B's real battlefield:
A May 2026 Goldman Sachs survey of 14 Chinese robotics firms reached a consistent conclusion: most prioritize wheeled chassis with two-to-three-finger grippers, covering 70%–90% of industrial applications.
Company Background
Commercialization is advancing on two tracks:
Industry Implications: From Concept-Chasing to Deployment
When the embodied-AI industry starts seriously calculating the economics of cost, efficiency, and stability, the signal is clear: blind pursuit of "looking human" is over. Similar pragmatic shifts are spreading across the industry — four-finger hands replacing rarely-used pinkies to cut costs, wheeled robots regaining favor in flat warehouses and inspection scenarios, and less insistence on fall-prone bipedal forms.
Competition has crossed the "can it work" demo stage into the industrialization deep water of "who can finish the job at lower cost and higher efficiency." For Chinese manufacturing this is a structural amplifier: complete supply chains lower hardware costs, while massive real-world workloads feed back into model iteration. Cheapest hardware + smartest brain is the engineering paradigm to watch in Chinese embodied intelligence over the next 12 months.
---
Sources