-
Products
-
Solutions
-
Markets
-
-
Technology
-
Resources
-
About
Development partners
System Electronics (industrial OEM) & Hipert (software company)
System Electronics, part of Coesia Group, is a company dedicated to the innovation of industrial electronics
Application
Automated order checking for pickers working with AS/RS
Accurately detect hand movements and identify objects moved from tray
Challenge:
Accurately detect hand movements and identify objects moved from tray
The goal is to reduce picking errors by more than 90% and enhance the overall productivity of the warehouse with real-time order checking.
Hardware used
Axelera Metis M.2 accelerator module
ARM computer module
4 HD input cameras
Warehouse AI
for warehouse automation
Across the globe, logistics operations incorporate warehouse automation to achieve operational excellence, speed, and efficiency. For instance, they use AS/RS, warehouse AI, augmented reality and warehouse robotics to streamline logistics processes, reduce the required space, effectively manage inventory, simplify warehouse management, and minimize the possibility of human error.
The demo showcase
Applying computer vision software for Automated Warehouses
Vittoria Cavicchioli - Perception Engineer at Hipert – talks about applying the computer vision software that they developed for System Electronics' warehouse automation system. Which is originally meant for real-time verification of orders picked in AS/RS warehouses.
How System Electronics developed a warehouse AI system for real-time order checking
During Embedded World 2024, Davide Bianchi (system architect at System Electronics) explains how System Electronics developed the system for an automated order verification system and why they included Axelera AI's m.2 module.
Developing Warehouse AI for automated order verification
During Embedded World 2024, Vittoria explains how Hipert developed the software for an automated order verification system. The system for this Warehouse AI application is designed by System Electronics and includes Axelera AI's m.2 module.
Our partnership with Axelera AI ensures that we not only meet customers’ current needs but also helps future-proof their deployments by providing a best-in-class solution.
~Andrea Gozzi, General Manager of
System Electronics
Automated order checking for warehouse pickers using AS/RS
Systems Electronics, part of the Coesia Group, aimed to develop warehouse AI hardware and software to automate order verification for people utilizing AS/RS, such as a vertical lift module. This innovative computer vision technology is intended to improve the accuracy of picking while increasing throughput. The goal is to reduce picking errors by more than 90% and enhance the overall productivity of the warehouse with real-time order checking. The company also wanted to ensure that the use of warehouse AI on the factory floor:
The challenge
Accurately detect hands and items for order verification
One of the major challenges was to develop a cost-effective, highly accurate vision AI system that could analyze an AS/RS scene in real-time, automatically track the movements of warehouse pickers, and verify if an article was picked up. This included addressing corner cases, such as workers using different types of gloves or long sleeves, changing lighting conditions, and instances where none or more than one item was picked.
As with most AI vision applications, it’s important to evaluate neural network models and training data sets needed. An easy-to-use, end-to-end software development environment will help automate data quantization, model evaluation, accuracy tuning, and hardware runtime integration.
Our group is committed to providing cutting-edge solutions to the industrial automation space. Better performance and accuracy at a similar price point compared to alternative AI accelerators brought us to select Axelera AI as our partner.
~Andrea Gozzi, General Manager of
System Electronics
Processing power for real-time video inference
Real-time analysis of hand movements and object detection requires more processing power than what a high-performance industrial PC can provide. So, System Electronics needed to find appropriate hardware with enough headroom for the intended warehouse AI.
The application needs to have enough AI performance to run an automated analysis of hand movements in four partly overlapping 1080p high-definition video feeds of the workspace in the AS/RS. It should also deliver low latency and robust accuracy. Since this is an industrial application, the embedded hardware needs to be ruggedized for 24/7 use, support industrial temperature range, and be in stock. The required combination of high AI performance, high accuracy, compact size, low power, and industrial requirements made traditional GPU solutions unsuitable for this automated warehouse application. That’s why System Electronics selected the AI accelerator hardware of Axelera AI.
We are able to level up in terms of power and computational efficiency.
We can offer our clients good-quality, rugged warehouse equipment with lower energy dissipation and space occupation.
Davide Bianchi, System Architect of System Electronics
The solution:
Real-time video analytics enabled by Axelera AI hardware
System Electronics built their order checking system by connecting four high-definition camera feeds to an embedded industry-grade ARM-based system. This system runs on a high-performance, low-power Axelera Metis M.2 AI accelerator card. The Axelera Metis M.2 delivers ample performance to support the application with headroom for future improvements, at a fraction of the cost and power consumption of traditional GPU solutions.
Functional warehouse AI for automated order checking
The picking system is integrated with the warehouse logistics management system which shows warehouse pickers which item to pick. If this warehouse AI for order checking detects that the wrong goods have been removed, or if more than one item was removed, or even if no goods were removed within a specific time, the warehouse pickers receive an alert and get feedback on how to proceed.
The Axelera Voyager SDK supported end-to-end development including steps of neural model quantization, compilation, metrics evaluation, pipeline profiling, and runtime hardware integration. In this way, System Electronics and Hipert could develop functional warehouse AI for real-time order verification Over 30 neural network models and options were evaluated and a pipeline using MMPose for palm keypoint detection and tracking was selected. A vision geometry block identifies the removal of an object from the vertical lift module and compares it with the data provided by the logistics management system.
The result:
Order checking in an automated warehouse
With the AI-enhanced order flow assist system installed and optimized, users of AS/RS can expect a significant reduction in false pickings and an increase in throughput efficiency within the automated warehouse. Achieving a 90% decline in false picks is a realistic goal, which will lead to reduced stress for workers and easier onboarding of new workers. The order pickers further appreciate the assistance of the warehouse AI, resulting in higher job satisfaction, and the seamless introduction of new goods in the process.
