Core Hardware Architecture ("Torso and Senses")
In terms of hardware composition, an AGV forklift consists of a traditional forklift mechanical chassis integrated with four key subsystems:
Onboard Master Control System ("Brain"):
Typically utilizes industrial-grade controllers (such as Linux/RTOS-based industrial PCs or safety PLCs). It is responsible for processing navigation algorithms, receiving backend scheduling commands, and sending millisecond-level control signals to various motors and valves.
Multi-sensor Fusion Perception ("Eyes"):
Navigation LiDAR: 2D/3D LiDAR mounted on the top or front for spatial scanning and map matching.
Pallet Recognition Vision (Cameras): A 3D depth camera (ToF or binocular vision) located in the middle of the forks, specifically designed to identify the position, height, and tilt angle of pallet fork pockets.
Safety LiDAR and Bumpers: Collision-avoidance LiDAR and physical contact strips at the base for close-range obstacle avoidance.
Drive and Steering Motors ("Legs"):
Employs high-precision AC servo motors or permanent magnet synchronous motors (PMSM) paired with encoders for real-time feedback on wheel speed and steering angle, enabling millimeter-level control of movement and steering.
Actuators and Hydraulic Control ("Hands"):
Controls hydraulic cylinders via electro-proportional valve banks, or uses electric cylinders directly to drive mast lifting/tilting and fork sideshifting, ensuring precise control over lifting height and load handling.
Key Control Algorithms ("Operational Logic")
The AGV forklift achieves "unmanned autonomous operation" through several core algorithms running on both the backend and the vehicle itself:
Real-time Localization Algorithms (SLAM / Particle Filter):
Data frames scanned by LiDAR are matched against feature points in a pre-established map. By combining wheel odometry data with Extended Kalman Filter (EKF) or Particle Filter algorithms, the vehicle calculates its high-precision coordinates in 2D or 3D space in real time.
Visual Pose Estimation (Pallet Alignment Algorithm):
After the 3D camera captures pallet point cloud data, the system fits the pallet's 3D pose vector using ICP (Iterative Closest Point algorithm) or deep learning-based point cloud segmentation. Even if a pallet is placed askew, the algorithm calculates the deflection angle and issues real-time correction commands to the drive wheels and mast, ensuring smooth fork insertion.
Multi-vehicle Path Planning and Deadlock Resolution (Scheduling Algorithm):
The backend RCS system utilizes graph theory (such as time-expanded networks) and A*/CBS (Conflict-Based Search) algorithms to plan conflict-free routes for the entire AGV fleet within milliseconds. In scenarios where vehicles meet in narrow passages, a spatiotemporal node reservation mechanism directs lower-priority vehicles to pull over and yield, thereby preventing traffic deadlocks.
Complete Operational Workflow
Communication: The onboard 5G/Wi-Fi module receives task messages (via MQTT/TCP) from the RCS.
Path Navigation: The main controller calculates the trajectory and controls the servo drives for movement, utilizing a closed-loop correction system based on laser and odometry data.
Target Alignment: The vehicle decelerates upon approaching the target storage location; a 3D camera captures the pallet, calculates the positional deviation, and makes fine adjustments to the vehicle body and forks.
Forking and Lifting: The forks enter the pallet slots, the hydraulic valve opens, sensors confirm the load has cleared the ground, and the mast tilts backward and locks.
Safe Transport: Multi-level radar safety zones remain active throughout the journey; upon reaching the destination and unloading, the status is reported back to the WMS/MES.
