Robotera Unveils "End-to-End" Humanoid Solution to Close Logistics’ Final Automation Gap

At CeMAT ASIA 2025, Robotera (Star Action Era) unveiled a new logistics solution that it claims solves one of the most persistent bottlenecks in modern warehousing: the "flexible picking gap." By deploying its full-sized Star-Act L7 humanoid robot powered by a proprietary "end-to-end" Visual-Language-Action (VLA) model, the company is attempting to fully automate the complex process of selecting individual items from storage bins—a task traditionally reserved for human hands.

The announcement marks a significant shift in the logistics robotics sector, moving beyond the now-standard Automated Guided Vehicles (AGVs) that transport shelves, to humanoid systems capable of the dexterity required for item retrieval. Robotera claims this is the "world's first" application of an end-to-end VLA embodied model in a real-world logistics scenario.

The "Goods-to-Person" Bottleneck

The context for this deployment is the massive strain placed on logistics networks by e-commerce peaks, such as China’s recent "Double 11" (Singles' Day) shopping festival. While the industry has successfully automated the transport of goods using "Goods-to-Person" (GTP) systems—where AGVs bring shelves to a workstation—the actual picking of items remains labor-intensive.

Robotera argues that traditional robotic arms (rigid automation) struggle with the high variance of Stock Keeping Units (SKUs) and the rapid fluctuation of orders. When SKU varieties expand from tens of thousands to hundreds of thousands, rigid systems require constant, time-consuming reprogramming and fixture changes. This creates a reliance on temporary human labor during peaks, leading to training deficits and error rates.

The Hardware: Built for the Shelves

The physical component of the solution is the Star-Act L7, a bipedal humanoid designed to fit into existing warehouse infrastructure rather than requiring a facility redesign.

• Range of Motion: The robot features a 3-degree-of-freedom (DOF) waist design that allows it to cover a vertical workspace of up to 2.1 meters. This enables the robot to reach both the lowest and highest shelves typically found in standard storage racks.
• Dexterity: To handle diverse item shapes—from square boxes to irregular pharmaceutical packaging—the L7 is equipped with the "Star-Act XHAND1," a 12-DOF five-fingered dexterous hand.
• Precision: Robotera states the system uses a "cross-shaped wrist" and customized motor optimization to eliminate blind spots inside storage bins and ensure stable grasping without jitter.

The Brain: ERA-42 and VLA

The core differentiator in Robotera's announcement is the software architecture. The L7 runs on "ERA-42," a proprietary embodied AI brain that utilizes a Visual-Language-Action (VLA) model.

Unlike traditional coding, where a robot is explicitly programmed with coordinates for every motion, a VLA model allows the robot to interpret visual data and natural language commands to generate actions dynamically.

• End-to-End Learning: The system reportedly uses an "end-to-end" architecture, meaning it processes raw visual input and directly outputs motor controls. This reduces latency and allows for "high-frequency inference," enabling the robot to correct its movements in real-time if an object shifts position.
• Generalization: By training on a combination of open-source data and real-world machine data, Robotera claims the model can handle new SKUs without requiring specific retraining for every new product that enters the warehouse.

Integration and "Swarm" Potential

For a humanoid to be useful in a warehouse, it must communicate with the building itself. Robotera highlighted that its solution includes a standardized interface that connects the embodied AI directly to the Warehouse Management System (WMS).

The workflow demonstrated involves a seamless handover: a multi-shuttle vehicle brings a tote to the station; the L7 visualizes the target, grabs the item, scans the barcode, and places it in the order bin. If a scan fails, the robot autonomously places the item in a recycle bin.

This "bi-directional" communication allows the WMS to issue tasks and the robot to report status updates (success, failure, anomaly) back to the central system, creating a closed loop of data that helps the model improve over time.

While "world's first" claims are common in the rapidly overheating humanoid sector, Robotera’s focus on a specific, high-value industrial pain point—the inability of rigid robots to handle e-commerce variability—suggests a pragmatic approach to commercialization. The company views this not just as a pilot, but as a scalable "module" that can be replicated across production and circulation logistics.