English is currently available.
Engineers Gossips DISCUSSION
Boards combining a Linux processor with a microcontroller are getting more attention. Arduino's May 2026 UNO Q article describes this split for AI and real-time I/O.
For a small camera-based sorter, how should the work be divided? The camera classification can take a variable amount of time, but a motor and position sensor still need predictable timing. What would justify the extra processor instead of keeping everything on one device?
Separate the deadlines before selecting hardware. Image acquisition and classification may tolerate tens or hundreds of milliseconds, depending on the conveyor speed. A motor-control loop can have a much shorter deadline. If a 1 kHz loop is required, its period is 1 ms; inference must not block it. An MCU can maintain that loop while Linux handles the camera, model and user interface.
The interface between processors needs more than a class label. Include a sequence number, capture timestamp and confidence or validity information. The MCU should reject stale commands and know what to do when communication stops. If the belt moves at 0.2 m/s and total decision latency is 150 ms, the object moves 30 mm before the decision arrives. That displacement belongs in the mechanical timing budget.
A second processor adds boot time, power consumption, update coordination and a communication failure mode. A small model using a low-rate sensor may fit comfortably on an MCU, so measure model RAM, execution time and energy per inference first. Use the dual-processor option when its workload split solves a measured limitation. Linux alone can also be suitable when the actual timing requirement is soft rather than deterministic.
