JD.com is accelerating its shift toward fully automated logistics by planning to deploy 3 million robots, 1 million unstaffed vehicles, and 100,000 delivery drones over the next five years. The Chinese e-commerce giant unveiled its industrial Wolf Robot series at an event in Beijing on September 9, 2026, marking one of the most ambitious physical AI deployments in the logistics sector to date. Rather than simply replacing workers, the company is committing to retrain its vast workforce of couriers and warehouse staff for technical roles in robot maintenance and servicing.

What Are Wolf Robots Designed to Do?

The Wolf Robot system represents a comprehensive approach to automating repetitive, labor-intensive logistics tasks. According to Liu Lige, head of embodied intelligence robots at JD Logistics, the robots are engineered to pick, sort, transport, and deliver goods with minimal human intervention. The lineup includes specialized units capable of operating in extreme cold environments down to minus 20 degrees Celsius, automated pharmacy dispatchers, and delivery drones. These robots are designed to handle the kinds of physical tasks that have traditionally required human workers in warehouses and distribution centers.

The scale of JD.com's commitment is striking. The company currently employs approximately 700,000 delivery and logistics personnel across its ecosystem. Rather than viewing automation as a threat to employment, JD.com founder Liu Qiangdong introduced the "Nirvana Plan" earlier in 2026, an initiative specifically designed to retrain couriers and warehouse staff for technical roles like robot servicing and maintenance. This approach acknowledges the reality of job displacement while attempting to retain workers within the company's operations as automation spreads.

How Is Physical AI Reshaping Industrial Automation?

Beyond JD.com's logistics push, the broader physical AI sector is experiencing rapid innovation. On September 8, 2026, Palladyne AI and FANUC America announced a strategic collaboration pairing FANUC's industrial robot portfolio with Palladyne AI's physical AI software platform. The partnership aims to simplify robotic deployment, increase adaptability, and expand the range of manufacturing and logistics applications that can be automated. The companies plan to develop intelligent robotic capabilities that reduce the complexity traditionally associated with robotic automation while enabling faster deployment and more intelligent operation.

Palladyne IQ, the software foundation of this collaboration, uses artificial intelligence and machine learning to provide reasoning capabilities for industrial robots and collaborative robots. The platform operates through a four-stage cycle: observing the environment through multi-modal sensor fusion, learning new tasks from as few as one to five demonstrations, reasoning to generate real-time motion plans at the edge, and acting to control the robot and its end effector. The software is designed to run autonomously at the edge without requiring a cloud connection, making it practical for real-world deployment.

• Optimization and Training: The collaboration includes optimization of Palladyne IQ on FANUC robotic platforms, along with joint work on AI-driven motion planning and adaptive robot behavior.
• Simulation and Deployment: The companies plan simulation and AI model training to accelerate deployment, plus joint validation of customer use cases across manufacturing, warehousing, and logistics applications.
• Standardized Workflows: Development of standardized deployment workflows for system integrators and end users aims to make robotic automation more accessible and repeatable across industries.

The collaboration frames its effort around persistent challenges facing manufacturers: skilled labor shortages, increasing product variability, and pressure to improve productivity and operational efficiency. Target industries span industrial manufacturing, defense, automotive, aerospace and aviation, construction, infrastructure maintenance and repair, and energy.

What Makes Embodied AI Models Robust in Real-World Conditions?

Meanwhile, HiDream.ai launched HiDream-O1-Embodied, an embodied world model designed to advance physical interaction for embodied intelligence. The model ranked first on RoboColiseum's Robustness leaderboard, a standardized simulation benchmark for embodied intelligence models, with a score of 0.692. RoboColiseum evaluates models across four major dimensions: instruction following, spatial understanding, robustness, and general-purpose manipulation through 78 high-fidelity simulation tasks.

HiDream-O1-Embodied's strength lies in its ability to handle real-world imperfection. The model moves beyond traditional keyword matching in language understanding, covering an equivalent instruction space encompassing diverse verbs, sentence structures, and expressions. Rather than failing when instructions are rephrased, the model focuses on underlying intent. For example, it can understand "Bring me the cup," "Get me a cup," and "Hand me the cup" as equivalent requests.

"We believe a complete world model foundation requires three core capabilities: omni-modal representation, causal reasoning, and physical-world modeling, all centered on the ability to express, understand, and generate within the real world," explained Ting Yao, CTO of HiDream.ai.

Ting Yao, CTO of HiDream.ai

The model also integrates information from multiple viewpoints, allowing different visual channels to complement one another rather than relying on a single fixed perspective. When part of the visual information becomes inaccurate or temporarily unavailable, the model can leverage other viewpoints to understand the scene and continue execution. This transforms the system from one where a single failure causes total system failure into one where local limitations do not prevent overall operation.

HiDream.ai's approach to building robust embodied AI involves proactively introducing non-ideal conditions during training. By repeatedly exposing the model to incomplete, noisy, and unstable information, the system learns to make reliable decisions based on limited visual cues. Changes in lighting, image degradation, occlusion, signal fluctuations, and scene variation are all common challenges robots face during real-world operation, and HiDream-O1-Embodied is designed to maintain stable task execution under these complex, dynamic conditions.

How Are Companies Building Better Training Data for Physical AI?

A critical challenge in embodied intelligence is that the cognitive boundaries of a model are largely shaped by the data it can access. High-quality embodied data remains one of the scarcest and most decisive resources in the field. HiDream.ai employs a dual-driven "model plus data" strategy, making data production an integral part of model iteration. The company has developed a "real-world foundation plus generative augmentation" data production paradigm that enables the model to actively participate in creating and refining the data it needs to improve.

This approach is exemplified through HiDream.ai's collaboration with Noitom, which provides high-precision human motion-capture data as the real-world foundation. HiDream.ai leverages its native omni-modal capabilities to achieve 100-fold scale data augmentation and refinement, transforming limited real-world data into a much larger training dataset. This strategy addresses one of the fundamental bottlenecks in embodied AI development: the scarcity of high-quality training data that reflects real-world conditions.

The convergence of these developments, from JD.com's massive logistics automation initiative to advances in physical AI software and embodied world models, signals a fundamental shift in how companies approach automation. Rather than deploying rigid, single-purpose robots in controlled environments, the industry is moving toward flexible robotic systems capable of handling real-world complexity, variability, and imperfection. The challenge now is not whether robots can perform tasks in ideal conditions, but whether they can do so reliably when conditions are far from ideal.