Loongson Unveils Domestic Computing Platform for Intelligent Robot Vehicle with Depth Vision and Object Grasping

IT Home November 11 news, Loongson Zhongke announced yesterday that it completed the full-stack port of the ROS2 Jazzy version based on its self-developed intelligent machine car, connecting the complete links of mapping and positioning, handle remote control, robotic arm control, and depth vision. The entire vehicle is uniformly scheduled by the Loongson 2K3000 processor, possessing environment perception and target grasping capabilities.

ROS2 (IT Home Note: Robot Operating System version 2, Robot Operating System 2, the mainstream open-source middleware framework in the robotics field) leverages cross-platform and distributed communication characteristics to connect sensors, algorithms, and actuators into a reusable network. This full-stack port verified the formation of a reusable domestic robot development solution.

Its core computing power comes from the Loongson 2K3000. The chip uses a 12nm process, integrating 8 LA364E processor cores based on the independent Loong architecture, with a maximum main frequency of 2.5GHz. The cache configuration is 64KB each for primary instruction and data caches, and a shared 6MB secondary cache.

Loongson disclosed that its single-core performance can be benchmarked against ARM Cortex-A76 level processors, with the measured SPEC CPU 2006 Base single-core fixed-point score reaching 30 points.

The perception system is divided into two links. The LiDAR mounted on the side of the vehicle is responsible for planar perception, and the depth camera at the end of the robotic arm is responsible for 3D depth perception. Loongson completed parameter calibration for indoor scenes, complementing the odometer to improve positioning stability and mapping completeness.

In terms of chassis control, the robot is equipped with Mecanum wheels, achieving omnidirectional movement—including forward/backward, lateral, rotation, and diagonal—by combining the rotation speeds of four motors. The main control board connects directly to a USB controller; the self-developed Loongson program translates joystick values into velocity commands and includes deadzone filtering and a safety enable key.

Loongson stated that it will subsequently complete the emergency stop and watchdog functions, integrate action groups into the upper-level program, and advance 3D target localization involving detection and point cloud fusion. The complete link from mapping and positioning to robotic arm grasping verified the feasibility of the Loongson 2K3000 in robot computation processing and real-time control scenarios, providing a reference solution for prototype development in industrial automation, intelligent warehousing, and other fields.