Development of autonomous driving transfer trolley on field roads and its visual navigation system for hilly areas

Abstract: In hilly areas, it is difficult to realize mechanized transportation with high safety for agricultural products and materials due to constraints of natural conditions. With the gradually decreasing of rural labor force, farmers in hilly areas urgently need highly automatic field road transfer trolley to reduce the amount of labor required to transport agricultural products and to increase productivity. In this paper, an autonomous driving transfer trolley with visual navigation system for hilly areas were developed and studied. The transfer trolley mainly consisted of drive and brake system, control system, autonomous navigation system, ultrasonic radar obstacle detection system and automatic steering system. The autonomous navigation system included a RTK-GNSS (real-time kinematic-global navigation satellite system) and a machine vision module. The RTK-GNSS functions as road coordinate information collecting, real-time positioning and path planning, the machine vision module functions as field road identifying and path tracking line extracting. To avoid the effect of incorrect positioning resulted from occasional GNSS signal outages due to obstacles such as trees and crops along both sides of the field road, autonomous guidance was implemented by the machine vision module at the non-intersection segments of the road, while it was implemented by the RTK-GNSS at the intersection segments of the road. According to the features of field road with large curvature and fluctuation, in the global path planning, an improved A* algorithm was presented through adjusting the evaluation function by introducing the curvature at intersection nodes and fluctuation information of the road into cost function. In the field road image processing, in order to better distinguish the road area from its surroundings, V component of HSV (hue-saturation-value) color space was used for image segmentation, and S component and V component, after point operating, were fused by weighted method to identify the shadows on the road. Then the shadow regions were combined with the segmented road region. After obtaining the accurate road region in the image, the region was divided into 12 blocks and the centroid points of each block were extracted and smoothed to form a virtual line which was taken as the autonomous navigation line on the road. According to driving speed of the transfer trolley, points on the navigation line were selected as targets for preview tracking control, and a fusion method of front wheel steering angle was used to realize the transition between 2 sequential images. A global path planning simulation test based on actual field road network information was performed to compare the results between the improved A* algorithm and the Dijkstra algorithm. The simulation results showed that: compared with the Dijkstra algorithm, the accumulated altitude change of the path planned by the improved A* algorithm reduced 29.87%, and the total energy consumption of the transfer trolley through the path reduced 29.4% accordingly. Therefore, the improved A* algorithm was more suitable for field roads with large curvature and fluctuation and the corresponding planed path was more reasonable. An actual driving test on a 1.2 m wide field road was carried out. The transfer trolley was set to automatic driving with a constant speed of 2 m/s. To survey the deviation between the autonomous travel trajectory and the actual midline of the road under various conditions in hilly areas, 3 types of field roads, namely straight, complex multi-curvature and fluctuating roads, were selected as test roads. The autonomous driving test showed that: the mean deviations between the actual midline of the road and the automatic travel trajectory on straight roads, multi-curvature complex roads and undulating roads were 0.031, 0.069 and 0.092 m, and the maximum deviations were 0.133, 0.195 and 0.212 m, respectively. Taking the distance from road edge to road centerline as the calculating basis, the average relative errors of the transfer trolley, automatic traveling along the road centerline of these 3 roads, were 5.16%, 11.5% and 15.3%, respectively, the autonomous visual navigation system meeted the safety requirements of autonomous driving transfer trolley on field roads in hilly areas.