Overview
At the JDDiscovery conference on September 9, 2026, JD Logistics presented its full "SuperBrain + Wolf Family" stack: the SuperBrain 3.0 orchestration model system and a fleet of nine product categories (11 robot models) covering the entire logistics chain—warehousing, sorting, transportation, and delivery. Five models were newly released. JD also announced a five-year procurement plan: 3 million robots, 1 million unmanned vehicles, and 100,000 drones—the largest single robot procurement commitment ever announced by a logistics operator. As InfoQ framed it: breaking free from humanoid worship.
Why the industry is stuck
Per Interact Analysis, over 70% of warehouses worldwide rely heavily on manual labor, with automation penetration below 25%. Even automated warehouses still lose over half of operating costs to flexible tasks like picking and replenishment. The killer is non-standardized parcels: roughly 20% of irregular, soft, or odd-shaped items consume nearly 80% of logistics workers' operational effort (BEUMER data). Machines handle standard parcels; the hardest work remains with humans.
SuperBrain 3.0: five models, three layers
| Model | Role | |---|---| | PreX | Demand forecasting (orders, weather, road conditions) | | OptiX | Global decision optimization for billions of parcels | | OmniX | Multimodal perception, cross-device instruction conversion | | GeoX | Spatiotemporal dynamic scheduling network | | EmbodiedX | Embodied model directly driving robot execution |
These map to a three-layer architecture: decision AI → process AI → physical AI. The system simultaneously computes warehouse sequencing, vehicle dispatch, and inter-node timing; on sudden order surges or route changes it re-plans within seconds. Official figures: in-warehouse efficiency up nearly 20%, overall operational efficiency up 15–20%.
The Wolf Family lineup
Existing nine categories: ZhiLang (goods-to-person systems), YiLang (embodied dexterous arm), DuLang (delivery vehicles/drones trucks), FeiLang (drones), TianLang (pallet dense storage), DiLang (transport robots), CangLang (embodied picking robot), MuLang (unmanned pharmacy), and BaoMuLang (automated sorting wall). Newly released five:
| Product | Scenario | Key specs | |---|---|---| | CangLang | Mobile picking in complex warehouses | 2-week deployment, UPPH 80, 99.9% accuracy, >85% SKU coverage | | ZhiLang low-temp | −20°C cold-chain goods-to-person | Storage +100%, efficiency +200%, 10% per-order cost cut | | MuLang Health | Unmanned pharmacy | 7×24, 99.9% dispensing accuracy, 90 orders/hour | | DuLang Gen 6 | L4 delivery vehicle | 200 km range, 1000 kg payload, >8-year lifespan | | FeiLang L05 | Drone | 5 kg payload, 40 km range, 72 km/h cruise |
CangLang's "2-week deployment" is quietly the most important line: traditional warehouse retrofits can halt operations for 6 months, while CangLang works in existing warehouses—opening the retrofit market, not just new builds.
YiLang: 13 million real parcels
JD calls YiLang the world's first at-scale intelligent robotic arm, with nearly 100 interchangeable gripper configurations handling hundreds of irregular items—soft woven bags, inflatable packaging, odd-shaped boxes that conventional industrial arms cannot grasp. Cumulative volume: 13 million real parcels, running 24/7 across multiple logistics centers. Reported figures: per-unit operating cost 20%+ below manual sorting; human-intervention rate below 0.1%.
The UE break-even threshold
The hardest number in the release isn't a performance spec—it's unit economics (UE). JD states mature products like ZhiLang achieve payback within 3 years, while frontier embodied products like YiLang have crossed the UE break-even line, with per-unit costs below human labor. This shifts the evaluation standard from "can the robot do the motion" to "how much cheaper is the motion than a human."
Deployment and five-year plan
The system already runs in 10+ countries and 20+ Chinese provinces. Thousands of sixth-gen DuLang vehicles operate across hundreds of stations, cutting site-level transport costs 21% and improving turnaround efficiency 50%. The five-year plan: 3 million robots, 1 million unmanned vehicles, 100,000 drones.
Heterogeneous vs. humanoid routes
Rather than betting on a single general-purpose humanoid, JD chose heterogeneous collaboration: one unified intelligence hub orchestrating scenario-specific robots. The rationale: in environments with clear task boundaries and quantifiable metrics, customizing form factors per scenario reaches a viable cost model faster than building one universal body. The routes aren't mutually exclusive—heterogeneous systems capture cash flow in closed, measurable settings first, while humanoids bet on cross-scenario generalization.
Unresolved issues
| Challenge | Status | |---|---| | Non-standard parcel limits | YiLang covers hundreds of shapes, but a long tail of SKUs remains | | Cold-chain / pharma compliance | Low-temp ZhiLang and MuLang Health face the strictest regulatory regimes | | Maintenance at scale | 3 million units implies an entirely new O&M system; per-technician device ratios undisclosed |
An external variable: the rollout pace depends on e-commerce growth and last-mile delivery margins, which are under pressure—the very backdrop driving JD's cost-structure overhaul.
The bigger shift
Reading September's embodied-AI news together reveals a common pivot: at the 2026 Wuzhen/Bund Summit, 40+ embodied-AI vendors emphasized robots doing real work over demonstrating backflips; Ant Lingbo launched a robotic smart pharmacy with ~80 cm aisles requiring no store retrofit; a Beijing Shijingshan unmanned postal robot achieved ±0.03 mm positioning; and China's Ministry of Human Resources added "embodied intelligence robot application technician" as a new official occupation.
Three numbers to watch over the next 12 months: YiLang's cumulative parcel count beyond 13 million, actual annual deliveries against the 3-million-unit plan, and CangLang's "2-week deployment" repurchase rate in existing warehouses.
References
1. InfoQ: Breaking humanoid worship—JD rebuilds physical AI with the Wolf Family, 2026-09-11 2. InfoQ: SuperBrain + Wolf Family full lineup, 2026-09-09 3. Beijing Business Today, 2026-09-09 4. PCDaily, 2026-09 5. Xinhua: Robots start doing real work, 2026-09-10 6. Science and Technology Daily: First unmanned postal robot, 2026-09-10