AI Robotic Vacuum Brain based on RK3308

Client

A consumer robotics manufacturer developing a mid-range robotic vacuum with premium navigation features at a competitive price point.

Challenge

Running simultaneous localization and mapping (SLAM), obstacle detection, and path planning on a sub-$50 compute module while maintaining 5-hour battery life.

Product Concept & Architecture

We architected the system around the Rockchip RK3308B quad-core Cortex-A35 processor. A 360° lidar provides ranging data, while an optional depth camera enables object classification. An STM32G0 MCU handles motor control as a safety co-processor.

Engineering Solution

The system runs a custom Yocto Linux with ROS 2 Humble. The RK3308B runs Cartographer SLAM for mapping, while objects are classified using a quantized MobileNet SSD on the CPU. We implemented a tiered navigation system: lidar for global path planning, IR bumper for contact detection, and camera for object avoidance. The STM32G0 independently monitors cliff sensors and shuts down motors if the main SoC hangs.

Result

Sub-2cm mapping accuracy. Obstacle avoidance success rate of 99.2% against common household items (cords, shoes, pet waste). BOM cost of the compute module: $38.