I can’t believe how fast things are moving in artificial intelligence (AI) space (where no one can hear you scream). As one example, I just saw a video of a high school student called Benjamin Choi who built a prosthetic arm that he can control with his mind using AI.
As reported by Smithsonian Magazine, Benjamin built this low-cost manipulator using his sister’s 3D printer. He employs only two non-intrusive sensors (one on his ear and the other on his forehead) to monitor brain activity. This data is transmitted to the prosthetic arm via Bluetooth, where it’s processed by an AI model that’s running in a chip in the arm.
The thing that totally blows me away is that Benjamin did all this himself, including developing the AI model. Think about this for a moment. I mean… seriously? This sort of technology would have been the stuff of science fiction fantasy only a couple of decades ago. Even today, it’s the sort of thing you’d expect would require a team of mad scientists squirreled away in secret underground bunkers. By comparison, Benjiman worked his magic on the ping-pong table in his basement. Suffice it to say that if you ran across this plotline in a movie, you’d roll your eyes in disbelief.
Returning to the part where I said, “This data is transmitted to the prosthetic arm via Bluetooth, where it’s processed by an AI model that’s running in a chip in the arm.” The article doesn’t mention what sort of chip. I wonder if it’s a Hailo-8 from the chaps and chapesses at Hailo Technologies. If you aren’t familiar with the Hailo-8, may I point you at my previous column where I posed the question: Are These the Top-Performing Edge AI Processors?
The answer to this question then (and the answer now) is “You Betcha!” Well, it is if you are talking to D.C. Smalley, who is the General Manager, Americas, at Hailo. D.C. says the Hailo-8 provides 10X the performance at 1/10X the power of competing products.
Meet the Hailo-8 AI/ML accelerator (Source: Hailo)
The Hailo-8 offers 26 tera operations per second (TOPS) running at about 3 TOPS per watt (all the benchmarks comparing Hailo-8 to competitive offerings are available on Hailo’s website).
The reason the Hailo-8 is currently at the forefront of (what I laughingly call) my mind is that I was just chatting with D.C., who kindly brought me up to date with the latest and greatest offerings from Hailo, and with Brandon Hjelstrom from Limelight Vision, whose latest and greatest smart camera features a Hailo-8.
Starting with Hailo, in addition to the Hailo-8 AI Accelerator itself, we also have Hailo-10H Generative AI (GenAI) accelerator-based modules. With an M.2 form factor, these modules, which deliver up to 40 TOPS, can be plugged into existing PCs and edge devices with an M.2 socket to execute deep neural network inferencing in real-time utilizing low power for a broad range of applications and market segments.
As the folks at Hailo will tell you (if you don’t manage to get out of the way fast enough), “To unlock the full potential of generative AI and seamlessly integrate it into our daily lives, it needs to be fully immersive, readily accessible, and an integral part of edge devices.” Well, who amongst us could argue with logic like that?
Also, we have the Hailo-15, which is a family of AI Vision Processor Units (VPUs) for smart cameras. The Hailo-15 VPU System-on-a-Chip (SoC) devices combine Hailo’s patented and field-proven AI inferencing capabilities with advanced computer vision engines, generating premium image quality and advanced video analytics. The unprecedented AI capacity of these VPUs can be used for both AI-powered image enhancement and the processing of multiple complex deep learning AI applications at full scale and at superior efficiency.
Furthermore, there’s a cornucopia of cards, such as the Hailo-8 Century High Performance PCIe Card, which supports up to 8x Hailo-8 chips offering up to 208 TOPS of AI performance with low power consumption.
Actually, while I’m thinking about it, Avi Baum, CTO at Hailo wrote a very interesting blog about The Evolution of AI on the Edge from Perception to Creation.
The Evolution of AI on the Edge from Perception to Creation.
Avi’s blog discusses the rise of perceptive AI, the emergence of enhancive AI, the promise of generative AI, and the case of generative AI at the edge.
Sad to relate, none of the above is what I was wanting to talk about (sorry). The reason for this column (yes, of course there’s a reason) is that I just heard about an exciting step forward for STEM education and the development of vision-based robots.
Are you familiar with the guys and gals at Limelight Vision? These little scamps specialize in developing high-performance, zero-code smart cameras designed to simplify robotic vision and perception. Designed to be user-friendly, Limelight products cater to both beginners and professionals. Their flagship product line enables users to integrate advanced vision capabilities into robots without extensive programming knowledge.
The reason for my waffling is that Hailo and Limelight Vision have collaborated to launch the Limelight 4 Smart Camera. This zero-code solution is designed to be easy enough to use for individuals with no or limited coding experience, yet powerful enough for even the most sophisticated robotic projects and products.
Powered by Hailo’s Hailo-8 AI accelerator with its 26 TOPS compute power, the Limelight 4 equips any robot with advanced AI capabilities, enabling a range of real-time perception tasks such as object and people detection, 3D robot localization, image segmentation and more. Additionally, the Limelight platform supports various vision pipelines, such as color blobs, AprilTags, neural networks, and more, all configurable through a built-in web interface. It’s also powered by the Raspberry Pi CM5 and includes 2GB of RAM and 16GB of storage.
The platform also supports multiple programming languages and protocols, including C++, Java, Python, .NET, Rust, WPILib, ROS, ModbusTCP, WebSocket, and REST, ensuring versatility and ease of integration.
Limelight 4 Key Specifications are as follows:
- OV9281 Global Shutter Monochrome Camera (1280×800 at 120FPS, 640×400 at 240FPS)
- Hailo-8 Support for accelerated robot localization and object detection
- Raspberry Pi Compute Module 5
- Full 3D Robot Localization with IMU fusion, up to 240 FPS
- REST, Websocket, NetworkTables, ModbusTCP APIs
- Gigabit Ethernet
- Plug-and-Play USB-Ethernet functionality
- 3.5V-35V Buck-Boost power stage
Founded in 1989, FIRST (For Inspiration and Recognition of Science and Technology) is a global nonprofit organization that prepares young people for the future through a series of team-based robotics competitions held across the US and globally.
Designed for high school students, the FIRST Robotics Competition FRC) is the highest-level robotics competition within FIRST. Teams of students, mentors, and sponsors design, build, and program industrial-sized robots to compete in an annual game with strict rules, limited time, and a standard set of parts. FRC emphasizes teamwork, problem-solving, and STEM education.
Do you want to see some FRC teams explaining their use of LimeLight cameras? I know I do! Here are some YouTube videos from 3015 Ranger Robotics, 4414 HighTide, 1796 RoboTigers, and 118 Robonauts.
The exciting news is that Limelight 4 cameras will be made available to participants in the 2025 FIRST Robotics Competition. (The 2025 season, themed “REEFSCAPE presented by Haas,” invites teams to use their engineering skills to strengthen ocean habitats.)
The folks at Hailo and Limelight say that their Limelight 4 collaboration marks a significant milestone in democratizing AI-powered technologies, making advanced robotics and AI capabilities more accessible to users across skill levels. I totally agree. What say you?
