Artificial intelligence is changing the way logistics is automated and will bring new benefits to labor-intensive industries.

Automation uses technology to enable humans to perform more tasks. In the logistics sector, the potential for automation is huge and the benefits are obvious, especially when there are major changes in operations or increasing demand. Scaling up operations often requires adding additional staff, who are often not available immediately, especially when other industries have similar needs. How to respond quickly to market fluctuations requires rapid action and other additional capabilities throughout the operation.

Logistics automation can quickly increase capacity based on changes in demand. By elevating logistics automation to a strategic position, it can not only improve productivity, but also reduce human errors, thereby improving work efficiency. With the right logistics automation software, hardware, and platform resources, the impact on operating costs is relatively small even in periods of low demand, far less than the cost of maintaining a large number of manpower. As demand increases, the capacity of operations is ready and can be quickly started. Although these methods can bring the flexibility needed for logistics companies to respond quickly to changes in demand, there is still an opportunity to do more.

Artificial intelligence will amplify the impact of logistics automation

Introducing AI into logistics automation will greatly enhance the impact of AI. AI can reduce errors in common semi-skilled tasks such as sorting and sorting products. Using autonomous mobile robots (AMRs) can improve the efficiency of package delivery, including the most expensive last-mile delivery. AI helps autonomous mobile robots (AMRs) plan routes and identify features such as people, obstacles, delivery portals, and doorways.

When integrating logistics automation into any environment, it presents certain challenges. It can be as simple as replacing a repetitive process with a powered conveyor belt or as complex as introducing autonomous robots with collaborative capabilities into the workplace. When artificial intelligence is added to the automation and integration process, the challenges become more complex, but so do the benefits.

As solutions become more connected and gain better insight into other stages of the process, the efficiency of each automated element increases. Placing AI close to the devices that generate data and take action is called edge AI. The adoption of edge AI is redefining logistics automation.

Edge AI is developing extremely rapidly and its uses are not limited to logistics automation. The benefits of placing AI at the edge of the network must be balanced with the availability of resources, such as power, environmental operating conditions, logistics location, and available space.

Inference at the Edge

Edge computing brings computing and data closer together. In traditional IoT applications, most data is sent to (cloud) servers over the network, where it is processed and the results are returned to the edge of the network (such as physical devices at the edge). Only cloud computing introduces latency considerations, which is unacceptable for time-sensitive systems. Here is an example of edge computing at work. During the sorting process, capturing and processing image data of packages locally allows logistics automation systems to respond in just 0.2 seconds. Network delays in this part of the system will slow down the sorting process, but edge computing can eliminate this potential bottleneck.

While edge computing brings computing closer to data, introducing artificial intelligence to the edge can make the process more flexible and less prone to errors. Similarly, the last mile of logistics relies heavily on manual labor, but autonomous mobile robots (AMRs) using edge artificial intelligence can improve this situation.

The introduction of artificial intelligence will have a significant impact on the hardware and software used in logistics automation, and there are more and more potential solutions. Typically, the solutions used to train AI models are not suitable for deploying models on the edge of the network. The processing resources used for training are designed for servers, and their demand for resources such as energy consumption and memory is almost unlimited. At the edge, energy consumption and memory are limited.

Heterogeneous Trend

On the hardware side, large multi-core processors are not well suited for edge AI applications. Instead, developers are deploying heterogeneous hardware solutions optimized specifically for edge AI. This approach includes CPUs and GPUs, but of course, it can also extend to ASICs, MCUs, and FPGAs. Certain architectures, such as GPUs, excel at parallel processing, while other architectures, such as CPUs, are better at sequential processing. Today, no single architecture can truly provide the best solution for AI applications. The general trend is to configure the entire system with hardware that can provide the best solution, rather than using multiple instances of the same architecture.

This trend points to heterogeneity, where there are many hardware processing solutions of different architectures configured to work together, rather than using the same architecture across multiple devices (all based on the same processor). The ability to bring in the right solution for any given task, or consolidate multiple tasks on a specific device, can provide greater scalability, as well as optimized performance per watt and/or per dollar.

"Moving from homogeneous systems to heterogeneous processing requires a large ecosystem of solutions and mature capabilities to configure these solutions at the hardware and software level. That's why it's important to work with a vendor that has the ability to build partnerships with all the silicon vendors, to deliver solutions for edge computing, and to work with customers to develop systems that are scalable and flexible."

In addition, these solutions use common open source technologies such as Linux, as well as specialized technologies such as the Robot Operating System ROS2. In fact, more and more open source resources are being developed to support logistics and edge AI. From this perspective, there is no single "correct" software solution, and the same is true for the hardware platform that runs the software.

Building edge computing with a modular approach

To increase flexibility and reduce vendor lock-in, ADLINK has developed a modular approach at the hardware level that allows for more flexible hardware configurations in any solution. In fact, hardware-level modularity allows engineers to change any part of the system hardware, such as the processor, without causing system-wide disruption.

The ability to “upgrade” the underlying platform (whether software, processors, etc.) is particularly important when deploying new technologies such as edge AI. Each new generation of processor and module technology generally provides a better power/performance balance for the inference engine at the edge of the network, so being able to quickly take advantage of these performance and power gains, minimizing disruption to the entire logistics automation system, and edge AI system design is also a clear advantage.

By using microservices architecture and Docker container technology, modularity in hardware is extended to software. If a more optimized processor solution is available, even if it comes from a different manufacturer, the software takes advantage of the fact that the processor is modular and can replace the previous processor without changing the rest of the system. Software containers also provide a simple and powerful way to add new functionality to run in edge AI.

Software within containers can also be modular. ADLINK's Edge Vision Analytics (EVA) SDK (software development kit) for AI vision products is a typical example. The platform is based on Gstreamer and focuses on the basic functions needed to build an AI vision pipeline. Each stage of the AI ​​vision pipeline uses ready-made open source plug-ins (which themselves contain modules) to simplify the development of the pipeline. These plug-ins include image capture and processing, AI reasoning, post-processing, and analysis.

The modularization and container approach of hardware and software minimizes the risk of vendor lock-in, which means that the solution is not dependent on any specific platform. It also increases the abstraction between platform and application, making it easier for end users to develop their own applications that are not dependent on any platform.

We simplify the upgrade process through a database that characterizes components as they become available. Using this database, engineers can select products that provide the perfect balance between inference performance and system resources.

One of the most important requirements for logistics automation is real-time response. Therefore, it is very important to work with a supplier who has extensive experience in developing systems with a combination of software and hardware that can meet the application requirements. ADLINK's approach is to use modules that can be integrated with professional third-party technologies such as LiDAR sensors.

in conclusion

Deploying edge AI in logistics automation does not require replacing the entire system. First, you need to assess the workspace and identify the stages that can truly benefit from AI. The main goal is to reduce operating expenses while improving efficiency, especially in times of labor shortages to cope with increased demand.

More and more technology companies are working on developing AI solutions, but most companies usually only target cloud computing, not edge computing. On the edge side, the operating conditions are different, resources may be limited, and even a dedicated network may be required.