Home > Latest > News

Loongson demonstrates full ROS 2 integration for robot navigation, vision, and manipulation

Date: 2026-11-18Source: Beijing E-Town

Loongson Technology, a company based in the Beijing Economic-Technological Development Area (BDA), also known as Beijing E-Town, has completed a full-stack port of ROS 2 Jazzy to its self-developed intelligent mobile robot platform. The integration connects mapping and localization, joystick control, robotic arm operation and depth vision in a single system, demonstrating that domestically developed processors can support the full ROS 2 software stack and provide a locally developed computing foundation for embodied AI applications.

ROS 2 (Robot Operating System 2) is an open-source middleware framework widely used in robotics. It enables sensors, algorithms, and actuators to communicate and coordinate across platforms through distributed messaging and support for multiple programming languages. This provides the software infrastructure for applications requiring close coordination between perception and movement, commonly known as "hand-eye coordination".

Loongson's mobile robot uses the company's 2K3000 processor as its central controller, coordinating operations and issuing commands across the system. The platform supports omnidirectional movement, environmental perception, and object manipulation, enabling it to detect objects, identify their positions, and carry out grasping tasks.

The company said the platform had been tested across four key capabilities: multimodal perception, omnidirectional mobility, motion planning and hand-eye coordination. The resulting system provides a reusable development framework for robotics based on domestically developed hardware and software.

Combining 2D and 3D Vision for Environmental Perception

The robot uses two complementary sensing systems. A LiDAR unit mounted on the side of the chassis provides two-dimensional environmental perception, while a Gemini depth camera attached to the end of the robotic arm captures three-dimensional information.

The LiDAR system builds maps in real time. Loongson has adjusted its parameters for indoor environments to improve map boundary definition and update responsiveness. Combined with odometry data, this helps improve localization stability.

The Gemini camera captures both color and depth images, providing three-dimensional information to guide the robotic arm during object recognition and grasping.

Omnidirectional Movement and Joystick Control

The mobile platform is equipped with Mecanum wheels. By varying the speeds of its four motors, the robot can move forward and backward, travel sideways, rotate and move diagonally without changing its orientation.

A USB joystick connects directly to the main control board. Loongson-developed software converts joystick inputs into movement commands. The controller has six axes: the left stick controls forward, backward, and lateral movement, while the right stick controls rotation. Dead-zone filtering helps prevent unintended movement, and a safety enable button provides an additional control safeguard. Operators can also switch between controlling the mobile platform and the robotic arm with a single button.

Motion Planning Coordinates the Robot's Components

The system uses MoveIt, a widely adopted motion-planning framework in the ROS 2 ecosystem, to coordinate the chassis, LiDAR, camera and robotic arm within a shared coordinate system on the 2K3000 processor.

Once the robot reaches a target location, the planner calculates the movement of each joint, taking into account obstacles, path length and the stability of the arm's orientation. It then generates a trajectory that can be executed directly by the robotic arm.

Hand-Eye Coordination Enables Object Grasping

The robot is fitted with a six-axis robotic arm and an end-mounted gripper for picking up and moving objects. Its Gemini camera is mounted on the wrist of the arm, moving with it to form an "eye-in-hand" vision system.

During grasping, the camera identifies the target's three-dimensional position. The main controller uses this information to plan the arm's trajectory and issue position commands, coordinating the joints to complete the operation.

The arm also supports joystick-based teaching and motion-sequence recording and playback. Operators can demonstrate a task once and save the sequence for repeated execution. When materials or workstations change, recorded actions can be reused without having to reprogram the robot from scratch.

A Reusable Platform for Robotics Development

The integrated system demonstrates that the Loongson 2K3000 processor can meet the computing and real-time control requirements of a full ROS 2 robotics application. The company sees the platform as a ready-to-use, independently developed solution for prototyping mobile robots, industrial automation systems and intelligent warehousing applications.

Loongson plans to add emergency-stop and watchdog functions, integrate recorded motion sequences with higher-level software, and develop three-dimensional object localization by combining object detection with point-cloud data. These improvements are intended to move the platform beyond basic system integration towards more reliable and practical deployment.