Embodied Composite Robot System Design Training Platform RAI-M4
A one-stop embodied AI training platform integrating an omnidirectional chassis, 5-axis robotic arm, depth camera, LiDAR, and voice interaction. Adapts to mainstream large models like Qwen. Features 3 vision solutions and 2 SLAM navigation solutions. Built-in ROS 2. Equipped with comprehensive courses on embodied agents, vision, navigation, and Robot Operating System. Suitable for professional comprehensive training.
Applicable Audience/Scenarios
Ideal for courses on mobile robot control, LLM deployment, robotics, computer vision, ROS, and mobile navigation/localization
Highlights
- Hybrid platform: mecanum chassis plus 4-DOF manipulator
- DeepSeek + Qwen large-model integration for planning and perception
- Course-ready framework for mobile robotics and LLM deployment
Product Features
Deep large-model integration
Qwen-powered ASR and DeepSeek-based LLM understanding equip the robot with natural-language task planning, while Qwen’s multimodal vision removes the need for customised datasets.
Mobile-and-manipulation composite platform
The mecanum chassis and serial arm operate together in confined workspaces, enabling general-purpose tasks when combined with task planning and perception pipelines.
Structured experiment roadmap
Curricula are modularised around machine vision, large-model deployment, robotics, ROS, and navigation so instructors can mix and match modules that fit their teaching needs.
Lab Scenarios
Multi-angle imagery showcases RAI-M4’s chassis, manipulator, and sensor modules to simplify lab setup and scenario design.
Multi-angle gallery
Configuration
Sensor Configuration
A comprehensive sensing suite supports navigation, perception, and interactive feedback.
- HD camera (optional upgrade to depth camera for 3D perception)
- 360° LiDAR (0.12–8 m range)
- IMU / gyroscope for pose estimation and odometry refinement
- Breathing light and status indicators
- Expansion ports for touch displays or additional sensors
Controller Configuration
Dual-controller design: the upper controller manages planning, perception, navigation, and kinematics, while the lower controller delivers motor PID control, manipulator actuation, and interactive I/O, acting as the bridge to the upper layer.
Software Configuration
Ships with Ubuntu + ROS2, MoveIt, YOLO inference, and Qwen / DeepSeek API examples so labs can immediately begin combined mobile robotics and LLM experiments.
Supports Python / C++, ROS2, MoveIt, OpenCV, YOLO, Qwen SDK, DeepSeek API, and other mainstream frameworks for coursework and research extensions.
Experiments
Experiment tracks span machine vision, large-model deployment, robot body control, ROS operations, and mobile navigation, allowing flexible lesson planning.
Machine vision module
Build a full pipeline from classical image processing to deep/multimodal perception.
- OpenCV vision：HSV colour recognition; shape recognition; QR code recognition; barcode recognition; colour-ring detection (integration + filtering)
- AI vision – YOLO：YOLO deployment; dataset annotation; model training and deployment; workpiece data collection and inspection; face detection; face tracking
- AI vision – Qwen multimodal：Qwen multimodal API deployment; object detection and tagging
Large-model deployment & applications
Focus on voice dialogue, multimodal perception, and embodied execution.
- Voice dialogue interaction：ASR deployment with Qwen; LLM semantic understanding with DeepSeek; TTS deployment with Volcano Engine; end-to-end voice dialogue; voice-enabled calculator; voice-triggered music playback
- Multimodal vision detection：Qwen multimodal API deployment; object detection and annotation
- Robot-integrated scenarios：MCP-based grasping task planning; MCP-based navigation task planning
Robot body control module
Practise kinematics and control for both chassis and manipulator.
- Chassis control：Encoder-based motor PID; mecanum kinematics; mecanum odometry with gyroscope fusion
- Manipulator control：Servo position control; manipulator kinematics; motion interpolation control
ROS operations module
Develop ROS topic/service/parameter skills plus MoveIt planning.
- ROS fundamentals：Control turtlesim via topics/services/parameters; spawn additional turtles; port and run packages to drive turtlesim with the keyboard
- MoveIt arm planning：Configure arm URDF; set up MoveIt kinematics; use RViz for motion planning
Mobile navigation & localization module
Cover interfaces, mapping, and navigation with multi-goal support.
- Interface configuration：Keyboard teleop for chassis; keyboard teleop for manipulator; acquire LiDAR data
- Mapping workflow：Configure mapping project files; craft launch files; build new maps
- Navigation workflow：Configure Navigation project; define collision boundaries; point-to-point navigation; autonomous obstacle avoidance; multi-goal navigation
Knowledge Base
Get more technical documentation, tutorials, and FAQs about this product.
Key Questions
Q3: Which products support LLM integration and what can they do?Embodied Composite Robot System Design Training Platform RAI-M4Embodied Robot Task Planning Training Platform RAI-P4Embodied Vision Perception & Decision Training Platform RAI-Q2
Answer: There are three products that support large-model integration:
RAI-P4: integrates Qwen, DeepSeek, and Volcano Engine; supports ASR (Qwen), LLM (DeepSeek), TTS (Volcano Engine), and function calling (such as voice-dialog calculators, music playback, and gimbal / robotic arm task planning), and also supports integrated applications with YOLO, face tracking, and robotic arm control.
RAI-M4: connects to DeepSeek (LLM) and Qwen (ASR + multimodal); supports converting natural language into robot task workflows (voice commands for chassis / robotic arm control) and multimodal object detection (Qwen), combining a mecanum chassis and a 4-axis robotic arm to achieve generalized manipulation.
