Edge AI as a topic was something Advantech began seriously thinking about “one year ago,” Thomas Kaminski, Director, Product Sales & Marketing Management at Advantech Europe told IoT Insider at electronica 2024.
Advantech, who was present at electronica with their partner with Hailo, are focusing on expanding its energy-efficient Edge AI portfolio, with the aim for Advantech to use Hailo’s AI accelerators to develop Edge AI systems and AI acceleration modules.
“Hailo is, I would say, one of the strongest companies … besides the chips [they offer], they’re strong in the software environment,” said Kaminski. “50% of the AI workload is software effort. If we have the right software environment, you can develop your model, train it and get it executed on the right hardware [easily].”
In short, Advantech’s partnership with Hailo strengthens the software side of AI which is especially important for AI and Edge AI processing. Additionally, although AI is in its infancy, the established nature of Hailo as a company, alongside the longevity of their solutions, was an important point for Advantech.
“If you’re looking all over the market, you may find a lot of startups or young companies who might have a good product … If you’re beginning to spend a lot of money in an industrial environment, you need to be sure that the same [solution] is still available and maintained for five years, possibly even longer,” added Kaminski. “If you select a younger company, how can you be sure that company will survive in this new and aggressive market?”
Edge AI
Edge AI, which refers to processing data on the Edge, on the device, and is markedly different from, for example, storing data in the Cloud.
“Today, if you’re using Siri, your phone is capturing your voice, transmitting your voice to a server in the Cloud, executing and translating it in the Cloud and sending back the recommendation,” explained Kaminski. “If you ask [Siri] for recommendations for the nearest restaurant, it’s translated and decided in the Cloud.
“On the Edge, your voice is captured at the phone, translated on the phone, and the recommendation is done on the phone.”
There is still a place for both Cloud and Edge processing, however the final application will determine what is best to use.
“[Edge AI is best] where you can’t accept any latency on capturing information, like watching a patient in hospital; controlling cameras in public; transportation,” said Kaminski.
Delving into these real-world applications where Edge AI is deployed further, examples include production environments, “like watching the workers, watching the quality, watching the products … If it’s a production area and there are safety aspects with a forbidden area nobody can enter, it could be shut down without latency.”
Keeping public order is another key aspect, where Edge AI can be deployed to monitor violence or even avoid overcrowding in popular areas; and autonomous mobile robots (AMRs), “if they need to cooperate, sync to one another and make decisions in real time,” shared Kaminski.
2025 talking points
In looking ahead to 2025 talking points, Kaminski said that they would be moving more functionalities to the Edge, as well as the impact the Cyber Resilience Act is anticipated to be a major topic for hardware platforms and operating systems, “starting from development [all the way to] selecting the right operating system or software,” according to Kaminski. Energy grid management is hoped to be transformed using artificial intelligence, which could be used to identify at what times consumption is taking place and patterns of behaviour.
“The AI in the grid will learn how to balance [a peak],” said Kaminski, “this is when grids are going to improve.”
“We have the intelligence, we need only use it,” he concluded, in a poignant point that will echo across industries.
We’ve just begun with Edge AI and will move functionalities to the Edge, which requires new computing performance, more memory, more storage.”
Thomas Kaminski, Director, Product Sales & Marketing Management at Advantech Europe
There’s plenty of other editorial on our sister site, Electronic Specifier! Or you can always join in the conversation by commenting below or visiting our LinkedIn page.
