TL;DR
This paper presents an autonomous tracked grain vehicle designed to improve harvesting efficiency by automating the grain transfer process. It combines a novel hybrid A* path planning with fuzzy pure pursuit for navigation and binocular vision with semantic segmentation for precise grain unloading, demonstrating successful field trials.
What Problem Does This Paper Address?
The efficiency of grain harvesting is significantly hampered by the downtime required for combine harvesters to unload their grain tanks. This process often involves manual transport to the edge of the field, consuming valuable harvesting time. Existing solutions often rely on wheeled vehicles, which struggle in the soft, muddy conditions typical of rice fields. The paper addresses the need for a more robust and automated solution for this critical post-harvest step, specifically focusing on a tracked vehicle capable of operating in challenging terrain and performing autonomous grain transfer and unloading.
Methods
The researchers developed a comprehensive unoccupied control system for a tracked grain vehicle built upon a combine harvester chassis. Key methodological components include:
- Mechanical Design: A 4.5 m³ V-shaped grain tank was integrated, along with a hydraulic unloading system.
- Electronic Control System: A BODAS hydraulic controller was used to manage both manual and automatic operation of the vehicle's walking and unloading mechanisms.
- Autonomous Navigation:
- Path Planning: A hybrid A* algorithm was employed to plan the vehicle's travel path.
- Path Tracking: A controller combining fuzzy logic and the pure pursuit algorithm was designed to ensure accurate path following. Fuzzy control was used to dynamically adjust the "look-ahead distance" based on vehicle speed, improving stability.
- Automatic Grain Unloading:
- Grain Tank Recognition: Binocular vision and semantic segmentation (using the DeepLabv3+ network) were utilized to identify and segment the grain unloading region, even under challenging conditions like dust and varying lighting.
- Unloading Regulation: The system precisely controls the unloading tube's position and the stirring auger to ensure uniform grain distribution into the transport vehicle.
Datasets
The paper describes the creation of a dataset for training and validation of the semantic segmentation model. This dataset consists of 900 images of grain unloading scenarios, manually annotated using the Labelme tool. The images were divided into a 7:3 ratio for training and validation sets.
Key Insights and Ideas
- Tracked Vehicle Advantage: The paper strongly advocates for tracked vehicles in wet, muddy agricultural environments, highlighting their superior passability and low ground pressure compared to wheeled alternatives.
- Hybrid Path Planning & Fuzzy Control: Combining hybrid A* for global path planning with a fuzzy-tuned pure pursuit controller for local path tracking offers a robust approach to autonomous navigation in complex agricultural fields. The dynamic adjustment of the look-ahead distance based on speed is crucial for stability.
- Semantic Segmentation for Robust Unloading: Using semantic segmentation (DeepLabv3+) proved effective in overcoming challenges like blurred boundaries and varying lighting conditions, enabling reliable identification of unloading regions for precise control.
- Integrated System Design: The successful integration of navigation and unloading control onto a single Android tablet and BODAS controller simplifies the system architecture and facilitates coordinated operation.
Key Results
- Navigation Accuracy: In straight-line path tracking experiments, the average lateral absolute deviation was found to be 3.14 cm at 0.6 m/s, 4.63 cm at 1.0 m/s, and 6.46 cm at 1.4 m/s. These results met the requirements for unoccupied driving.
- Unloading Precision: The automatic unloading control system achieved a maximum absolute pixel deviation of 92 pixels between the actual and desired grain landing points, demonstrating sufficient accuracy for practical operations.
- System Integration: The system successfully integrated manual and automatic controls for both walking and unloading mechanisms, and the entire system was validated through field experiments.
Limitations and Future Work
- Lateral Deviation with Speed: The study acknowledges that lateral deviation in path tracking increases with vehicle speed. While the system met requirements, further optimization for higher speeds or more complex terrains could be beneficial.
- Forward Speed Synchronization: When operating in conjunction with a harvester, synchronizing the grain vehicle's forward speed with the harvester's is crucial. The paper notes that incorporating a longitudinal tracking control algorithm is necessary for this.
- Environmental Interference: The automatic unloading system can be susceptible to interference from background factors, leading to misidentification or edge fluctuation issues. Future work could focus on refining the semantic segmentation algorithm or exploring alternative methods.
- Overshoot Phenomena: Occasional overshoot in controlling the grain unloading tube's movement was observed. Optimizing the closed-loop control method is suggested to improve accuracy and efficiency.
- Field Terrain: The study primarily focused on speed's impact on control accuracy. Comprehensive evaluation of various field terrain conditions is recommended for further optimization.
Summary updated