The Hailo-8 AI Accelerator is a cutting-edge AI processor integrated in TTCONTROL’s computing platform FusionAI. It supports the integration of AI-enhanced applications by delivering real-time, low-latency, and highly efficient AI inferencing. The Hailo-8 AI Accelerator enables advanced image and sensor data processing in FusionAI to support sophisticated assistance systems and highly automated functions in off-highway vehicles.
Hailo.ai and TTCONTROL have put together a comprehensive tutorial for developers using FusionAI. The webcast takes less than 30 minutes to complete. It guides users through practical examples step-by-step, from the download of the tooling to the execution of a locally run AI model on FusionAI.
Prerequisites and access
The starting point is the Hailo.ai Developer Zone. For the FusionAI preparation chapter, you will also need access to the TTCONTROL Customer Portal. The linked resources below are the most direct entry points described in this tutorial:
- Hailo.ai Developer Zone - Tools, documentation, and developer resources.
- Hailo Getting Started Guide - Overview of the referenced Hailo Software Suite documentation.
- Hailo Community - Community support and discussions.
- TTCONTROL Customer Portal - FusionAI product downloads and platform resources.
Getting access and orienting yourself
The Hailo-8 AI Accelerator is integrated in FusionAI to support efficient AI inference for image and sensor-data processing. This opening chapter introduces the developer resources used throughout the series and frames the journey from tooling access to executing a model locally on FusionAI.
Watch: Getting started with Hailo-8 in FusionAI
Useful links
- Hailo.ai Developer Zone - Start here for tools and documentation.
- Hailo Getting Started Guide - Use this as the documentation companion.
- Hailo Community - Use for community questions and shared knowledge.
Explore computer-vision models in FusionAI
This chapter opens the Hailo Model Explorer Vision and uses the YOLO family as example. It then inspects a pre-trained YOLOv8 model, including its architecture in Netron, and introduces the Hailo Model Zoo as part of the model-selection and preparation context. This chapter is dedicated to model exploration, not deployment.
Watch: Explore YOLOv8 and Hailo model resources
Useful links
- Hailo Model Explorer Vision - Compare available computer-vision models.
- Netron model viewer - Inspect neural-network model architecture.
Install the tooling and optimize a model
Quantization can reduce model size, increase inference speed, and lower power consumption. This chapter shows installation and use of the Hailo Software Suite in a Docker environment, followed by the compilation and deployment flow for an optimized, quantized model on FusionAI.
Watch: Boost AI performance on FusionAI
Useful links
- Download the Hailo software package - Referenced software download.
- Docker installation steps - Referenced installation instructions.
Prepare FusionAI with Potara
Potara is FusionAI’s OS (Operating System). This chapter covers access to the TTCONTROL Customer Portal, development-environment setup, installation of the SDK and Universal Update Utility, flashing FusionAI with the current software release used in the video, hardware connections, power-up, and verification through SSH.
Watch: Prepare FusionAI with Potara
Useful links
- TTCONTROL Customer Portal - FusionAI product resources.
Useful commands
# Change to the Downloads directory cd ~/Downloads # Unzip the release unzip Fusion_Release_V1.4.0.1 -d fusion-release cd fusion-release/Fusion_Release_V1.4.0.1 ls # Change to the Development Environment directory cd development_environment # Allow installer execution chmod +x potara-glibc-x86_64-ttc-image-qt6-cortexa72-cortexa53-crypto-fusion-toolchain.sh # Install SDK (select the installation directory when promted) ./potara-glibc-x86_64-ttc-image-qt6-cortexa72-cortexa53-crypto-fusion-toolchain.sh
# Install NXP’s Universal Update Utilitysudo snap install universal-update-utility# Make sure you are in the right directorycd ~/Downloadscd fusion-release/Fusion_Release_V1.4.0.1# Add write permission to the directory ownerchmod u+w -R .# Change to the Update directorycd update/uuu
# Unzip the FusionAI imagegunzip ttc-image-qt6-fusion.rootfs.wic.gz# Check if TTCONTROL FusionAI is connectedsudo uuu -lsusb# “Flash” (write the bootloader and rootfs image) the devicesudo uuu -v -b emmc_all ./imx-boot-qt6 ./ ttc-image-qt6-fusion.rootfs.wic
# Copy the SSH key to device to avoid typing the passwordssh-copy-id root@10.100.30.160 # Password: Kamehameha# Connect to FusionAIssh root@10.100.30.160 # Password should not be required anymore
Run real-time object detection on FusionAI
This chapter demonstrates compilation and deployment of YOLOv8 model on FusionAI. It runs real-time inference, detects and annotates objects under edge constraints, and uses Hailo-8 acceleration.
Watch: Run real-time object detection on FusionAI
Useful links
- TTCONTROL Customer Portal - FusionAI software and application resources.
- Hailo Apps repository - Reference applications and examples for Hailo-based AI pipelines.
- Hailo Model Zoo - Browse Hailo-optimized models, configuration files, and compilation resources.
Useful commands
# Change to the Downloads directorycd ~/Downloads# Clone Hailo Apps repositorygit clone --recurse-submodules https://github.com/hailo-ai/hailo-apps.git -b 26.03.1 # Change to the Hailo Apps Cpp Object Detection directorycd hailo-apps/hailo-apps/cpp/obeject-detection# Create a “build” directorymkdir build# Change to the build directorycd build# Source (setup) the Potara SDK Environment. /opt/porata/1.4.0/environment-setup-cortexa72-cortexa53-crypto-ttcontrol-linux# Use cmake to configure the Projectcmake ..# Build using makeMake# Download YOLOv8 Obejct Detection Hailo Executable Filewget https://hailo-model-zoo.s3.eu-west-2.amazonaws.com/ModelZoo/Compiled/v2.18.0/hailo8/yolov8m.hef# Copy Object Detection application binary and HEF to FusionAIscp object_detection yolov8m.hef root@10.100.30.160:/app/# Connect to FusionAIssh root@10.100.30.160 # Password should not be required# Execute the Object Detection applicationcd /app./object_detection --net yolov8m.hef --input usb
Inside the Inference App – run a reference application on FusionAI
The reference application uses OpenCV with the HailoRT Async API and a three-thread pre-process, infer, and post-process design connected by queues. The chapter then replaces YOLOv8 with SC-Depth v3, changes the outputs from 16-bit unsigned integer to dequantized 32-bit floating point, updates post-processing, and visualizes depth maps on FusionAI.
Watch: Explore the HailoRT asynchronous inference pipeline
Useful commands
# Download SCDepth v3 Hailo Executable Filewget https://hailo-model-zoo.s3.eu-west-2.amazonaws.com/ModelZoo/Compiled/v2.18.0/hailo8/scdepthv3.hef# Copy modified application binary and HEF to FusionAIscp object_detection scdepthv3.hef root@10.100.30.160:/app/# Connect to FusionAIssh root@10.100.30.160 # Password should not be required# Execute the modified applicationcd /app./object_detection --net scdepthv3.hef --input usb
