I’m sure you’re familiar with the inexpensive Raspberry Pi single board computer. I have ten or so of them doing all sorts of specialized tasks around the workshop and Fab Lab. As of February, Raspberry Pi has sold 61 million of the devices over a 12 year period since launch.
These devices aren’t just used the way I use them in hobbyist experiments or to control 3D printers and smart home features. Clusters of hundreds of Pis have been used to test supercomputer software. They’re widely used for industrial automation control. They’re embedded inside commercial products. Back in 2020, Raspberry Pis even powered ventilators keeping COVID-19 patients alive.
Also: Connect to your Raspberry Pi from a web browser, anywhere in the world – here’s how
Mixing AI with the Raspberry Pi architecture is nothing new. But now, Raspberry Pi is shipping its own $70 AI Kit, based on hardware from edge chipmaker Hailo. Adrian Kingsley-Hughes discussed the tech in this kit in his article on the topic. He’s intrigued by the camera and pose analysis capabilities, with an eye towards upgrading his security system.
Just before the launch, I had the opportunity to sit down with Eben Upton, CEO and co-founder of Raspberry Pi and Orr Danon, CEO and co-founder of Hailo to talk about the product, the launch, AI, and what it all means.
And with that, let’s get started.
ZDNET: What inspired the partnership between Hailo and Raspberry Pi?
Eben Upton (Raspberry Pi): Raspberry Pi and Hailo are on very similar missions: to bring the power of general-purpose computing and of high-performance high-efficiency AI acceleration to a broad community of enthusiasts, educators, and industrial and embedded customers. So it seemed like a natural collaboration, and we’ve been working for the last six months to make it a reality.
ZDNET: How does the Raspberry Pi AI Kit differ from previous Raspberry Pi products?
Eben Upton (Raspberry Pi): Our customers have been building AI applications on our products almost since we launched Raspberry Pi 1 back in 2012. These either ran on our increasingly capable Arm CPU cores, or on older medium-performance accelerators, of which Google Coral is the most obvious example.
The AI Kit breaks new ground, both as our first Raspberry Pi-branded AI product, and by pushing performance up substantially in comparison with those legacy accelerators. We think it’s going to enable a new wave of AI applications both in our hobbyist community and in the industrial and embedded space.
ZDNET: What are some key features of the Hailo-8L M.2 AI acceleration module?
Orr Danon (Hailo): The Hailo-8L provides 13 tera operations per second of compute power, enabling the Raspberry Pi 5 to run advanced AI modules, mainly vision-based applications for video analytics. This is done in a very low power consumption which does not require active cooling of the device, empowering Raspberry Pi users with state-of-the-art AI capabilities.
ZDNET: Can you share some examples of innovative projects that could be developed with the Raspberry Pi AI Kit?
Orr Danon (Hailo): There are endless options to embed video analytics and AI algorithms into Raspberry Pi applications, and we are anxious to see what users are going to develop with these new capabilities.
There are already many Raspberry Pi users who embedded Hailo AI accelerators into their projects before we launched this product. One of them is Velo.ai who is using a Raspberry Pi Compute Module 4 empowered by a Hailo-8 AI accelerator to provide the Copilot, a camera-based, high-performance, low-power rider assistance system to improve bicycle riders’ road safety.
Eben Upton (Raspberry Pi): Based on past experience, I would expect to see our hobbyists finding new ways to apply the usual range of “standard” models, including object detection and pose extraction. This represents a continuation of an existing trend, but with the Hailo accelerator supporting higher resolutions and frame rates, and more modern, higher-accuracy models.
Also: The best Raspberry Pi alternatives
In the industrial space, there are numerous possibilities, including autonomous control of industrial robotics, and automated product inspection.
But inevitably the most exciting applications are the ones we can’t predict. Just as Raspberry Pi drove down the cost of high-performance embedded compute, the AI Kit drives down the cost of edge AI inference acceleration. This means that many novel applications, which today would be cost-prohibitive, will have a positive return on investment. It’s going to be exciting to discover what those applications are.
ZDNET: Can you explain how the AI capabilities of the Raspberry Pi AI Kit can be utilized in home automation projects?
Orr Danon (Hailo): As mentioned earlier, we can only imagine how users would leverage the advanced AI capabilities of their Raspberry Pi. This is basically empowering the house with vision and intelligence; that could be applied into a whole range of applications, from feeding pets to watering plants, creating new games or empowering the home assistant with eyes and understanding (“suggest a recipe based on the ingredients on my table”).
ZDNET: How do you envision the Raspberry Pi AI Kit being used in security applications?
Orr Danon (Hailo): The AI Kit allows Raspberry Pi users to embed advanced video analytics to identify people, animals, objects and vehicles. Users can develop access control and perimeter protection applications based on these capabilities.
Eben Upton (Raspberry Pi): I think this is one of the use cases where the AI Kit will really shine. We already see many examples of security or presence-detection applications from both enthusiasts and industrial customers: in fact, the “Santa Detector” was one of the earliest educational projects featured on our website. Adding a layer of semantic understanding of the scene will allow the user to experiment with more sophisticated responses to specific visual cues.
ZDNET: What advantages does the Hailo-8L provide for robotics projects compared to other processors?
Orr Danon (Hailo): In comparison to other processors, the Hailo-8L provides high compute power with a double-digit TOPS [Theoretical Operations Per Second] capacity, at the industry’s best performance-to-cost and performance-to-power consumption ratios.
This enables the AI kit to provide very attractive computing capabilities while maintaining cost efficiency.
ZDNET: How does the integration of AI with Raspberry Pi enhance its capabilities for educational purposes?
Eben Upton (Raspberry Pi): We’ve always believed that it is important to teach children about the state-of-the-art in computer science and technology. We do them a disservice by teaching them a simplified or obsolete view of the world.
Also: How to install Linux on your Raspberry Pi
The AI Kit allows us to open their eyes to the new tools that neural network models add to the toolbox that they’ll take with them into their careers.
ZDNET: What kind of AI models can be deployed on the Raspberry Pi AI Kit, and how can users create or obtain them?
Orr Danon (Hailo): [The AI Kit] is fully integrated with Raspberry Pi’s camera software stack and supports numerous out-of-the-box AI applications through Hailo’s robust software suite and model zoo which covers all the popular machine vision tasks, such as classification, object detection, semantic segmentation, pose estimation, depth estimation super resolution, image denoising, etc.
ZDNET: How does the Raspberry Pi AI Kit address the growing demand for edge AI solutions?
Eben Upton (Raspberry Pi): The AI Kit provides a ready-to-use platform for developing and deploying “edge AI” applications. With our Compute Module range of system-on-module products, we support seamless transition from low-to-medium volume deployments using our Raspberry Pi single-board computer products and AI Kit, to higher volume developments using Compute Module accompanied by Hailo 8 or 8L silicon on a carrier board.
ZDNET: What security measures are in place to ensure the safety and privacy of AI applications developed with the Raspberry Pi AI Kit?
Orr Danon (Hailo): Since the AI is running on the edge device and not in a cloud server, this solution provides the highest level of privacy and data safety. Personally identifiable information (PII) can be processed on the edge, and anonymized information can be uploaded to the cloud for further analysis and storage.
Eben Upton (Raspberry Pi): It is the responsibility of the developer to ensure that their application ensures the safety and respects the privacy of its users. But, as Orr says, locating inference operations at the edge of the network reduces the scope for privacy violations, and, by enabling the edge device to autonomously make AI-supported decisions in the absence of network connectivity, eliminates certain opportunities for safety issues to arise.
ZDNET: How does the Raspberry Pi AI Kit contribute to the democratization of AI technology?
Eben Upton (Raspberry Pi): We believe very strongly in the social value of giving startups and small businesses access to cutting-edge technology. In support of this mission, and as with all Raspberry Pi products, the AI Kit is available to all comers at a low, fixed, price point [US$70].
Orr Danon (Hailo): With an attractive price tag and easy-to-use software package this product allows developers around the world, whether professionals or hobbyists, to apply intelligence into their projects, combining the power of the Raspberry Pi ecosystem with the new world of applications enabled by AI.
ZDNET: What advice would you give to new users starting out with the Raspberry Pi AI Kit?
Eben Upton (Raspberry Pi): Explore the wide range of curated, pre-trained, pre-converted models in the Hailo model zoo for inspiration. Many common use cases can be supported using these models, or alternatively they can serve as inspiration and a jumping off point for your own experimentation.
Orr Danon (Hailo): Let your imagination run wild…almost everything is possible.
ZDNET: Any final thoughts or insights you’d like to share with our readers?
Orr Danon (Hailo): Hailo is very excited about this cooperation. For Hailo, this is an opportunity to reach new audiences and truly enable AI at the edge for both professional engineers and creative enthusiasts. As generative AI is slowly expanding from the cloud to the edge, we see a lot of potential for future collaboration with Raspberry Pi and new ways to empower developers.
Eben Upton (Raspberry Pi): This is a watershed moment for us: the first time we’ve deployed a first-party AI product, and an opportunity to work with a partner whom we’ve admired for a long time. We can’t wait to see what people do with the AI Kit, and where our collaboration with Hailo takes us.
Final thoughts
ZDNET’s editors and I would like to give a huge shoutout to Eben Upton and Orr Danon for taking the time to engage in this in-depth interview. There’s a lot of food for thought here. Thank you, Eben and Orr!
Also: Don’t buy a Raspberry Pi 5 without also buying this amazing accessory
The Raspberry Pi 5 and the AI Kit are available from Raspberry Pi’s list of approved resellers.
What do you think? Did their recommendations give you any ideas about how to engage with AI in your creator journey, or for your company or organization? Let us know in the comments below.
