The Raspberry Pi 5, the small-but-mighty computer that has become quite popular with tech hobbyists and industrial companies, is now also an AI computer. The company just released the AI Kit, a $70 extension kit with a neural network inference accelerator that can be used for local inferencing.
For this new extension module, Raspberry Pi is taking advantage of its HAT+ extension card. HAT stands for “Hardware Attached on Top”. It’s a cute acronym that the company has been using for extension cards that you can fit on top of a regular Raspberry Pi.
Raspberry Pi’s HAT+ extension card adds an M.2 slot, which is a standard extension slot that is commonly used for PC components. (this slot is connected to the Raspberry Pi through a single-lane PCIe 3.0 interface running at 8Gbps for our readers who care about the fine prints).
And the company has partnered with Hailo, an AI chip startup that recently raised $120 millio and wants to challenge Nvidia. Hailo specializes in chips that are designed to run AI workloads on edge devices, such as cars, smart cameras, robotics and now Raspberry Pi devices.
The accelerator module that Raspberry Pi is using for the AI Kit is the Hailo-8L. It’s an entry-level module in an M.2 for factor, which means that it can easily be fitted on the HAT+.
Once everything is installed, what you get is a Raspberry Pi 5 with 13 tera-operations per second (TOPS) of inferencing performance. It’s not much compared to the performance of an Nvidia GPU. But it’s cost-effective and works with the Raspberry Pi’s default 27W power supply.
On the software front, the latest release of the Raspberry Pi OS automatically detects the Hailo module. The neural processing unit becomes available for the OS and applications that take advantage of it.
Raspberry Pi has also updated its suite of camera applications so that they support neural network inferencing as part of the camera pipeline. For instance, it can be used for object detection (“this is a banana”), semantic segmentation (“these three things are moving vehicles”), instance segmentation (“these three moving vehicles are a truck, a red car and a blue car”), pose estimation and facial landmarking.
Those are just examples of what you can do with a Raspberry Pi equipped with the AI kit and a first-party or third-party camera. But the Hailo chip could also be leveraged for non-camera use cases.
It’s going to be interesting to see new use cases emerge from the Raspberry Pi community. This AI extension kit is a tool. Now, it’s up to Raspberry Pi’s customers to figure out what they want to do with it.