Understanding the Rockchip RV1126B and EASY EAI
Compiling AI models on the Rockchip RV1126B with the EASY EAI framework means turning high-level neural network code into optimized binaries that run locally on this embedded AI chip for fast, low‑power edge AI inference without cloud services. The RV1126B combines general-purpose processing with integrated neural acceleration, making it suitable for vision and inference workloads on resource-constrained devices. EASY EAI adds an online, cross-compilation workflow that links your x86 development machine to the board in real time, so your compiled output matches the libraries and environment on the device. This combination supports embedded AI deployment for applications like smart cameras, sensors, or gateways that must process data locally. You can choose between a cross-compilation workflow for speed and tooling, or local compilation on the board when you want a self-contained development setup.
Setting Up the EASY EAI Cross-Compilation Environment
The recommended path for Rockchip RV1126B compilation is cross-compilation with the EASY EAI framework. EASY EAI is an online compilation environment that connects your PC and the development board, mapping the board’s root filesystem into the PC environment. According to Guangzhou Lingmou Technology, this shields many compatibility problems common in traditional toolchain setups. Start by launching the EASY EAI environment with the provided run.sh script on your x86 machine. Use a serial connection to read the board’s IP address, then verify connectivity with ping from the PC environment. Once the network link works, mount the board’s rootfs to /mnt using an NFS mount command, or the prepared load_rootfs script for convenience. From this point, anything you place under /mnt in EASY EAI is synchronized directly into the board’s root filesystem, forming the basis of your AI deployment workflow.
Compiling and Syncing AI Applications via Cross-Compilation
With the root filesystem mounted, you can compile AI workloads and supporting tools for edge AI inference using the EASY EAI framework. The environment includes examples of common build methods, each with a build.sh script and a simple source file. Running ./build.sh inside one of these example directories generates a hello binary that you can copy into any subdirectory under /mnt, such as /userdata. After that, log into the board via the serial console, change into the matching directory, and confirm the file presence with ls. Executing ./hello proves that synchronization is working and that cross-compilation targets the RV1126B correctly. For more advanced projects, compile C/C++ code using aarch64-linux-gnu-gcc or g++ with --sysroot=/mnt, or integrate include($ENV{HOME}/configs/cross.cmake) in CMakeLists.txt to make AI model compilation repeatable and scriptable.
Using Local Compilation on the RV1126B Board
If you prefer to work directly on the embedded device, you can run local compilation on the Rockchip RV1126B itself. In this mode, the development board acts like a compact computer with an Ubuntu-like environment, suitable for building and testing smaller AI workloads without any extra deployment steps. Access the board’s shell using adb, an onboard serial debugging port, or ssh. Then you can compile code natively with the installed toolchain and qmake located under /mnt/usr/bin/qmake. This approach avoids cross-compilation setup but has trade-offs: the RV1126B CPU offers less performance than an x86 host, so build times are slower, and code editing is limited to the vi editor rather than desktop IDEs. Local compilation is useful for quick patches, configuration tools, or lightweight AI utilities that support your main deployed inference model.
Building a Practical Edge AI Deployment Workflow
To turn compiled binaries into a full embedded AI deployment, combine the strengths of both workflows. Use the EASY EAI cross-compilation environment for regular development of AI models and inference services, compiling with the RV1126B’s libraries via the /mnt sysroot. Keep your core application, model runners, and any pre- or post-processing logic in this pipeline so you benefit from fast x86 builds and IDE support. Then, validate directly on the development board by executing binaries under /userdata or other mapped directories. When you need small adjustments in the field, rely on local compilation via adb, serial, or ssh. This pattern lets you maintain a smooth cycle: develop and optimize centrally, deploy through EASY EAI, then test and refine on-device. The result is reliable, cloud-free edge AI inference on the Rockchip RV1126B, tailored to constrained embedded systems.