CES 2025: The Year Edge AI Took Over

In June 2024, we announced our collaboration with Raspberry Pi, which selected Hailo as its AI provider with the introduction of the AI HAT+ for Raspberry Pi 5. This integration highlights the maturity of our software, empowering tens of thousands of engineers, developers, hobbyists, and makers worldwide to create innovative AI applications. At CES, we demonstrated pose estimation on the Pi, a fun application used in gaming and entertainment but also applicable to security and wellness use cases, such as detecting abnormal or dangerous behaviors.

CES 2025 provided the perfect platform to unveil the Limelight 4, a zero-code smart camera and robot controller powered by Raspberry Pi and empowered by Hailo-8. This product, exclusively selected for the prestigious FIRST Robotics program, enables school robotics teams to integrate AI into their robots without writing a single line of code. Our collaboration with Limelight Vision exemplifies our commitment to making AI more accessible and impactful across industries.

Starting with the smallest form-factor, visitors saw a drone carrying the Astrial board, powered by Hailo-8 with 26 TOPS. This demo showcased people counting and face detection applications. The low power consumption of the Hailo-8 processor enables the platform to work with active cooling, maintaining the lightweight design and high performance necessary for surveillance in challenging areas, such as large fields, dense urban environments, or disaster zones.

This demo featured an ITS camera empowered by Hailo to capture vehicle license plates for monitoring, billing, and access control. This system is deployed by Flash Parking in multiple parking lots across the U.S.

Dell integrated Hailo’s AI capabilities into its industrial, ruggedized gateway, leveraging the M.2 Hailo-8 module to enable advanced people detection and segmentation. This demo showcased the gateway’s ability to deliver real-time AI-powered insights, enhancing performance in demanding environments

Velasea’s all-in-one gateway supports two Hailo-8 M.2 devices, delivering up to 52 TOPS and enabling CVEDIA analytics and Milestone VMS on up to 24 streams.

Running on an AIC 2U server, which can accommodate up to 16 Unigen E1.S cards with 2 x Hailo-8 each, this demo utilizes just 25% of its available compute power to process AI video analytics on 100 channels, demonstrating exceptional scalability.

Hailo-15 vision processors are starting to hit the market through our early access partners. At CES, we showcased security cameras from Korean-based Truen and American manufacturer Vicon, which featured its newly launched NEXT camera.

Additionally, we demonstrated an innovative integration of the CLIP zero-shot video language model, enabling free-text search on the camera, or across multiple video streams without pre-training. This technology could help locate missing persons or objects in crowded areas or identify suspects based on physical descriptions.

This ECU for driver assistance integrates the Renesas V4H sensor and Hailo-8 co-processor. Set for production this year by a top Chinese automaker, it enables “Feet Off, Eyes On” functionality through the processing of seven cameras, creating a 3D detection system and Bird’s Eye View for partially automated highway driving.

Utilizing low-cost radar sensors for 3D perception, this system detects free space and objects around a vehicle, functioning even in complete darkness or severe weather.

A camera and LiDAR fusion system for highly accurate long-range pedestrian detection, ensuring privacy through on-device anonymization.

A ruggedized ECU for off-road vehicles and heavy machinery, using six camera-based surround views to detect and protect workers near industrial and agricultural machines.

At CES 2025, we debuted the Hailo-10 AI accelerator, specifically designed to run large language models (LLMs) on edge devices. We demonstrated the Hailo-10 operating offline on a compact edge platform, running Stable Diffusion 1.5 to generate high-quality images from text prompts. Processing generative AI directly on personal devices ensures low-latency performance, maximum privacy, and seamless operation and integration of different apps without the need to program, write code, or rely on available features.