In a real-world dark store, customers place orders through a mini program, and the robot autonomously manages the entire fulfillment process, from understanding order details and planning routes to identifying and picking products, opening bags, packing items, and placing completed orders.

Powered by Galaxea’s G0.5 embodied foundation model, the system requires no pre-programming for individual products, routes, or workflows. It performs end-to-end tasks autonomously in real-world environments that are open, dynamic, and constantly changing.

A Real Dark Store, Not Redesigned for Robots

Products do not need to be carefully arranged. The robot operates directly in a real dark store environment.
Faced with a wide variety of products, changing positions, and irregular arrangements, the robot does not rely on fixed shelf locations or SKU-specific grasping rules. Based on order information and visual features, it can autonomously identify, locate, and pick products.

Generalization Across 10,000+ SKUs

Powered by the Galaxea G0.5 embodied foundation model, the robot can autonomously identify, locate, and pick different products based on order information and product images.
Across different packaging, sizes, materials, and placement conditions, the robot can generalize across more than 10,000 SKUs without requiring individual programming for each item.

From Single Actions to Long-Horizon Tasks

Dark store operations are not defined by a single pick, but by the autonomous coordination of a sequence of tasks.
The robot autonomously completes order understanding, route planning, navigation, product picking, bag opening, bagging, and item placement.
When tasks are interrupted, the environment changes, or execution errors occur, the robot can adjust its strategy, recover the workflow, and continue the task—enabling reliable execution of complex long-horizon operations.

Precise Manipulation of Flexible Objects

Plastic bags are soft, highly deformable, and difficult for robots to manipulate. Opening a bag and placing products inside are among the most challenging tasks in automated dark store operations.
Driven by the foundation model, Galaxea robots can autonomously open plastic bags, maintain the bag opening, place products inside, and complete the bagging process, enabling continuous and precise manipulation of flexible objects in real-world environments.

Real-World Productivity Metrics

From Demo to Real-World Productivity

The world’s first end-to-end robotic dark store operation demonstrates more than a single grasping or navigation capability. It validates a complete embodied AI workflow in a complex real-world environment.
Powered by Galaxea’s self-developed G Series embodied foundation models, the robot integrates product generalization, route generalization, flexible-object manipulation, and long-horizon task orchestration without requiring dedicated programming for individual SKUs, routes, or workflows.
This marks a shift for embodied AI from isolated capability demonstrations toward real-world productivity that can operate continuously, be replicated across scenarios, and scale to broader deployment.