Thinking of using Radxa Dragon q6A for a dual-cam AI robotics project – choice or trap? Looking for known issues/gotchas.
Hi everyone,
I’m an Electrical Engineering undergraduate student working on my final year senior capstone project. We are building an autonomous solar farm inspection robot, and I’m heavily considering the Radxa Dragon q6A (12GB RAM, Qualcomm QCS6490) primarily for its 12 TOPS NPU and high memory bandwidth.
Since we have a strict graduation deadline, we need a stable hardware foundation. I’d love to get your insights on whether this board has underlying software/hardware quirks, especially regarding:
Dual-Camera & USB Stack Stability: The robot will process dual streams in real-time—one RGB USB camera and one Uni-T UTi120 Mobile thermal camera (running a custom C++ port of libusb). Does the Qualcomm Linux kernel handle heavy, concurrent USB bulk transfers well without randomly dropping connections?
Qualcomm QNN / SNPE SDK Learning Curve: We need to quantize and deploy custom Object Detection models (YOLOv8-nano variant) to the 12 TOPS NPU. How painful is the current QNN C++ API workflow on Radxa’s official Ubuntu 22.04 headless server? Are there major kernel bugs that could halt our project progress?
M.2 2230 NVMe Thermals: We'll be doing heavy async logging/backups directly to an M.2 2230 SSD while operating outdoors. Does the board overheat quickly under combined NPU + SSD write workloads?
UART / CAN Bus Reliability: For GPS parsing and sending real-time actuation commands to a lower-level motor controller (STM32).
Given that this is a time-sensitive graduation project, would we be safer sticking to the tried-and-tested Rockchip RK3588 ecosystem (like the Rock 5B+/5T) where community support is broader, or is the Dragon q6A ready for a pure C++ production pipeline now?
Appreciate any insights, benchmark experiences, or warnings!