White paper: Pallet recognition for intelligent logistics

Pallet offset, specification differences, reflective surfaces, and low-reflectivity materials can affect unmanned forklift docking and handling. This article describes the engineering challenges of pallet recognition, the PalletPro recognition and docking workflow, and public application cases. In warehouse logistics, AMRs and unmanned forklifts can perform repetitive handling tasks. Safe operation, localization, and reliable pallet docking still depend on the complete vehicle, perception, control, and site-management system. Risks around mobile equipment must be controlled through vehicle control, obstacle detection, operating procedures, and site management. The UK Health and Safety Executive’s 2023/24 data on fatal collisions with moving vehicles provides useful industry context.

​

Warehouse automation background

Large warehouses use robots for sorting, transport, and picking. Public Amazon Robotics information shows that robotic systems typically combine perception, vehicle control, scheduling, and site-safety mechanisms. For an unmanned forklift, stable pallet and pose recognition is a key input.

​

Engineering challenges in pallet recognition

An unmanned forklift needs the pallet pose relative to the vehicle to adjust its docking path. In practice, lateral offset, heading, structure, and surface reflectivity vary. Ambient light and occlusion can also affect sensor data quality.

​

Recognition approach

PalletPro extracts point-cloud features such as pallet legs and crossbars from 3D ToF depth data, then outputs the pose used by the vehicle for docking. Labels and 2D LiDAR can work in specific environments, but suitability depends on the pallet structure, installation, and site conditions.

​

PalletPro pallet-recognition system

PalletPro combines a 3D camera with pallet-recognition algorithms and supplies pallet position and orientation to the vehicle control system.

• Integrated processing and output: The algorithm processes depth data on the camera and outputs recognition results at 10 FPS.
• Deployment and docking: The camera can be installed between the forklift forks. Automatic calibration establishes the coordinate relationship between the camera, vehicle, and pallet. Standard two-leg and four-leg pallets and stacked pallets can be evaluated for the site; legless or unusual pallets require project assessment.
• Integration: The system provides real-time recognition results. Configure the interface, vehicle installation, and docking strategy according to the project documentation.

• Environmental adaptation: For strong ambient light or low-reflectivity pallets, validate the camera model, mounting distance, exposure, and filter parameters on site.

Watch MRDVS outdoor pallet-recognition field test on Bilibili Low-reflectivity targets such as black pallets can reduce point-cloud quality. Use the field-test video and site data to assess recognition performance. Watch MRDVS black-pallet recognition field test on Bilibili

​

Recognition and docking

An M Series camera can be installed between the forklift forks to capture pallet depth data and output pallet pose through the recognition algorithm. In a typical workflow, the vehicle first captures the pose at approximately 2 m from the pallet and corrects its heading. It then performs precise positioning and lateral correction at approximately 1.5 m. Validate the actual distances and control strategy for the vehicle, speed, and site. The vehicle control system can use either two-stage docking or real-time docking, depending on the task flow.

​

Two-stage docking

1. Remote positioning: Capture recognition results 1,800–2,800 mm from the pallet and correct the heading and lateral offset.
2. Near-field verification: When the fork tips are approximately 200 mm from the pallet front edge, capture the result again and check whether it meets the forking requirements.
3. Forking: Fork the pallet after verification. If verification fails, adjust the vehicle and verify again.

​

Real-time docking

Continuously capture recognition results to adjust the vehicle pose. Make fine corrections and complete forking at approximately 200 mm from the pallet front edge. The control system should use data timestamps to handle latency.

​

Communication

The solution supports interfaces such as TCP, UDP, CAN, RS-485, and API. Check the relevant product and deployment documentation for availability and message formats. Switch between recognition and obstacle-avoidance algorithms according to the project configuration.

​

Case studies

​

Case 1: Pallet and cable-drum-hole recognition for an optical-fiber communications company

Project background: Company F uses unmanned stackers for logistics tasks. The project covers recognition and handling of pallets and cable-drum holes. Solution: PalletPro supplies pallet depth and pose information so the vehicle controller can adjust the stacker pose. 3D vision also recognizes cable-drum holes.

​

Case 2: Non-standard pallet recognition in the fiberglass industry

Project background: China Jushi Group’s Tongxiang base introduced unmanned forklifts for automated production-floor handling. Challenge: Internal logistics uses customized metal pallets with different shapes and placement angles, increasing recognition and forking difficulty. Solution: The unmanned forklift uses PalletPro to capture pallet depth data and correct pallet position and orientation in real time.

​

Sources

1. UK Health and Safety Executive
2. The Robot Report