The logistics arm of JD.com is expanding its “Wolf Pack” of embodied robots, reflecting the growth of logistics warehouses as the industry’s first major testing ground.
JD Logistics (2618.HK), the logistics arm of e-commerce giant JD.com (9618.HK) (JD.US), used its annual technology conference this week to unveil six new embodied robotics products spanning warehousing, sorting, delivery and cold-chain operations.
The lineup comprises an autonomous “Mother Wolf” pharmacy system designed to operate around the clock, a low-temperature version of the “Smart Wolf” bin-to-person robot for cold-chain logistics, the “Warehouse Wolf” embodied picking robot, a sixth-generation autonomous delivery vehicle, an unmanned logistics drone, and a dexterous robotic arm for automated sorting.
JD Logistics is building what it describes as the world’s biggest deployment of embodied robots, covering the logistics chain from storage and handling to sorting and last-mile delivery.
The launch comes as embodied AI moves into a new phase. After years of demonstrations focused on robots walking, grasping objects and performing household tasks, companies are increasingly being judged on whether their machines can perform useful work reliably in real-world environments.
Warehouses become the proving ground
“Last year, people were watching demonstration videos of embodied intelligence. This year, that’s no longer enough,” one industry insider said at the World Artificial Intelligence Conference in Shanghai in July. “You have to put the robots into real-world scenarios and look at the actual work and specific tasks they can complete.”
Logistics and warehousing have emerged as one of the first places where that test is taking place.
In May, U.S. robotics company Figure publicly livestreamed a 200-hour package-sorting trial involving its Figure 03 humanoid robot. Working beside a conveyor belt, the robot autonomously scanned, picked up, flipped and deposited packages, processing about 249,600 parcels during the trial, nearly 21 a minute on average.
Chinese robotics companies are pursuing similar applications. Beijing-based startup Robotera, also known as Xingdong Era, has deployed an embodied-intelligence solution for small-package feeding with SF Express (002352.SZ) (6936.HK), a delivery services and logistics group, and China Post at more than 10 logistics centers nationwide. Its M7 robot uses a fixed, half-humanoid design rather than legs, reflecting the reality that a sorting station does not require a robot to walk around.
AgiBot has partnered with JD.com on the Genie G2 Max, a wheeled humanoid designed for heavy-duty warehouse work such as palletizing and moving bins between unloading and storage. Warehouse robotics specialist Geek+ (2590.HK) has also established a dedicated embodied-intelligence business and launched Gino 1, a general-purpose robot for picking, box handling, packing and inspection.
The common thread is that companies are often choosing wheeled or fixed-position robots rather than fully humanoid machines. The approach combines the mobility and efficiency of a wheeled platform with the dexterity of a human-like upper body, while avoiding the cost and complexity of bipedal locomotion.
Why logistics is leading the way
The appeal of warehouses is not accidental. Logistics combines several characteristics that make it unusually well suited to the current capabilities of embodied AI.
First, the industry has a large and persistent need for labor. Sorting, palletizing and other warehouse jobs are repetitive and physically demanding, while night shifts and difficult working conditions can make recruitment and retention challenging. For companies, robot adoption is therefore becoming less about a theoretical efficiency gain and more about operational necessity.
Second, logistics offers enormous scale and highly reusable skills.
The logistics robotics market was worth about 70 billion yuan ($10.4 billion) in 2025, according to management consultancy Roland Berger, with most applications concentrated in factory and warehouse internal logistics. Embodied robots still account for only a small share of that market, but growing e-commerce, instant retail and flexible manufacturing are increasing demand for automated handling.
Warehouse robots also encounter millions of different stock-keeping units, exposing them to a huge range of everyday objects. Skills developed through repeated warehouse tasks — such as reliably picking up an object and moving it from one location to another — could eventually become building blocks for applications in other commercial settings and, potentially, homes.
The environment itself is another advantage. Warehouses are relatively standardized and their tasks are clearly defined. JD Logistics says a promising large-scale application should meet four criteria: high frequency, high standardization, high measurability and low failure costs.
A warehouse task such as picking a package is performed thousands of times, its value can be measured precisely, and a mistake is generally recoverable: if a package is dropped, it can simply be picked up again.
That makes warehouses something of a “comfort zone” for current embodied AI — difficult enough to push the technology, but controlled enough to tolerate its limitations.
Three hurdles to mass adoption
The industry is nevertheless some distance from moving from hundreds of robots in pilot projects to thousands or tens of thousands in commercial deployment.
The first challenge is reliability. Companies need to focus not on a robot’s best performance in a demonstration, but on its worst performance in an actual warehouse. JD Logistics says large-scale deployment requires a 99.99% end-to-end task completion rate, rather than a 95% success rate for individual grabs.
At a warehouse handling tens of thousands of tasks a day, even a seemingly impressive 95% single-task success rate could translate into thousands of human interventions.
The second hurdle is cost. Humanoid and wheeled embodied robots remain substantially more expensive than conventional robotic arms and automated guided vehicles, partly because of the cost of dexterous hands, high-precision sensors, advanced joints and computing hardware.
Component durability is also an issue. Geek+ says leading dexterous hands currently have lifespans of roughly 300,000 to 500,000 operations, while a warehouse robot may perform 8,000 to 10,000 grabs a day. The company has opted for a three-finger design in an attempt to balance dexterity with industrial reliability.
The third challenge is engineering. A robot cannot simply be dropped into a warehouse and told to get to work. It must connect with warehouse management and scheduling systems, adapt to different layouts and workflows, and coordinate with human workers.
Without standardized interfaces and operating procedures, much of that integration remains customized — making engineering and deployment costs almost as important as the robot itself.
That is why JD Logistics’ expanding “Wolf Pack” is significant beyond the company itself. The next stage of the embodied-AI race will be determined less by increasingly impressive demonstrations and more by whether robots can become reliable, affordable components of real businesses.
Warehouses, with their scale, repetitive tasks and measurable economics, are where that test is now beginning in earnest.
