On July 17th, a special live demonstration at the World Artificial Intelligence Conference (WAIC) captured significant industry attention. While embodied AI robots are often featured in media for performances like dance or martial arts, there remains a gap in achieving continuous operation, complex collaboration, and stable delivery within actual production environments. Robots that can work in real industrial and commercial settings and generate tangible production value are a key focus for the industry. At this event, Kejie, the smart supply chain brand under DC HOLDINGS (SEHK: 00861), partnered with Daka Robotics to shift the focus from the WAIC exhibition floor to Kejie's logistics warehouse in Wuqing, Tianjin. There, multiple embodied AI robots performed continuous picking operations, using actual production orders to demonstrate a complete workflow. This included receiving tasks, path planning, item identification and grasping, to returning sorting carts and handling composite packaging, vividly showcasing the progress of embodied AI moving from "stage demonstrations" to "real work."

Inside the warehouse, multiple wheeled embodied AI robots navigated Kejie's thousands of square meters of operational space, independently completing the entire warehousing picking process. Multiple robots operated concurrently in the same aisles, using system scheduling and path algorithms to automatically avoid collisions and queue for passage. When battery levels fell below a preset threshold, robots autonomously proceeded to recharging points. While some units charged, others continued working to ensure business continuity.

It's important to note that the robots are not intended to completely replace human workers in the warehouse. Instead, they collaborate with Kejie's frontline staff to fulfill orders. The robots handle the picking process, which is then verified by Kejie personnel. This human-machine synergy leverages the robots' advantage of stable operation while retaining the critical human role in anomaly identification, quality judgment, and order delivery.

Core Distinction: Real Warehouse, Real Operations

Unlike proof-of-concept demonstrations in lab environments, the core of this showcase lies in its "real warehouse" and "actual scenario." As a 5A-level logistics enterprise with nearly 200 warehouses and a peak daily order processing capacity of 5 million, Kejie set a clear requirement from the outset: robots must not perform merely demonstrative picking; they must enter real warehouses and continuously handle real orders.

To achieve this, Kejie provided the Daka Robotics project team access to a several-thousand-square-meter commercial warehouse in Wuqing, Tianjin. This was not a temporarily constructed simulation site but a live operational warehouse handling daily shipments for numerous leading brands. The robots had to contend with challenges difficult to replicate in a lab but crucial for commercial viability, such as positioning drift in long corridors, adapting grasps for shelves of different heights, handling extreme variations from single items to full boxes, and managing diverse packaging across categories like apparel and accessories.

Integrating Operational Expertise into Robot Logic

More fundamentally, Kejie translated over two decades of supply chain operational experience into robot task rules. Which aisles should robots use? Where should waypoints for obstacle avoidance be set? In which areas should sorting carts be placed? Under what conditions can empty carts be dispatched? When should robots recharge to ensure uninterrupted operation? These Standard Operating Procedures (SOPs) for warehouse operations and on-site flow designs were defined by Kejie's team based on their deep understanding of real business processes and embedded into the robots' operational logic.

The system automatically determines the optimal travel path for robots based on order locations, prioritizing tasks that are closer to reduce unnecessary travel. During multi-robot cluster operations, when multiple robots encounter each other, they do not become "deadlocked." Instead, they dynamically adjust their routes based on task locations and aisle status, prioritizing orders with closer paths.

Co-Creation and On-Site Optimization

In the on-site optimization of the robot body and action plans, Kejie and Daka Robotics collaborated closely, ensuring deep integration between the hardware and the scenario's workflow to form a practical, deployable embodied AI warehousing picking solution.

The initial robotic arm design followed a scheme previously validated in smaller settings like pharmacies. However, issues emerged during actual operations in Kejie's logistics warehouse: when some bins contained very few items, the original degrees of freedom, while sufficient for reach, posed risks of failed suction or restricted posture. Based on on-site feedback, the robots customized for this scenario's picking tasks were enhanced with an additional degree of freedom in the arm, upgrading from a 7-degree-of-freedom arm to an 8-degree-of-freedom one, increasing wrist mobility.

Regarding grasping strategies, the robots initially attempted to operate with both "hands" simultaneously to boost efficiency. However, this interfered with the SKU distribution for different clients and the operational flow within the warehouse. Consequently, the final robot picking solution was adjusted: for small SKUs common in sportswear, like wristbands and headbands, the system employs a single-suction nozzle scheme to avoid dual nozzles picking two items at once. For apparel categories with larger packaging, dual suction nozzles are used to ensure stability. One side of the arm's end is equipped with a suction device, while the other side features a dexterous hand for steadily towing the sorting cart.

System Integration and Automated Workflow

The integration of orders and systems also highlights Kejie's system capabilities. Kejie's self-developed DC Treasury supply chain software platform is deeply integrated with the robot scheduling system. After a consumer places an order, it is automatically transmitted via the OMS to the DC Treasury platform. The system then automatically allocates tasks suitable for robot execution. Robots autonomously receive and execute these tasks, requiring no manual intervention for each individual order throughout the process.

Defining Technology Through Real Scenarios

This joint debut at WAIC is a concentrated embodiment of Kejie's philosophy of "scenario-defined technology." It showcases the deeper meaning behind Kejie's positioning as a "technology-driven industry supply chain expert": technology is not an isolated device or algorithm but must be embedded within a collaborative system of orders, warehouses, and people, undergoing continuous validation by real business operations.

Kejie provided more than just a real warehouse. It integrated over 20 years of supply chain operational experience, standardized warehousing processes, real order data, and systematic operational rules into the robot training. It participated in the continuous optimization of everything from warehouse mapping and operational flow to order allocation, multi-robot scheduling, grasping actions, and human-machine collaboration. This helps embodied AI robots progress from being "capable of running" to being "capable of stable operation in real logistics scenarios."