Conference Agenda

In mountainous regions, which account for 40% of Japan's farmland, weed mowing on steep slopes is hard and dangerous work. This research aims to automate mowing on steep slopes with an incline of over 30°. The mower under test consists of a crawler vehicle and a two-rotor mower driven by a battery and electric motors. The mower travels back and forth along contour lines on steep slopes. To prevent the mower from falling downhill, it has a mechanism that changes the height of the left and right crawlers to change the mechanism and vehicle width, allowing for stable travel even on slopes. The mower which is equipped with RTK-GNSS and IMU can travel automatically using target path following control.

Cable yarders are a key technology for timber extraction in steep terrain, yet studies from the Northwestern Italian Alps remain limited. This study evaluated the time consumption, productivity, and operating costs of a truck-mounted cable yarder equipped with a self-propelled carriage operating in a Norway spruce stand at 1760–1860 m a.s.l. A time and motion analysis was conducted on 120 yarding cycles, and regression models were developed to identify the main operational factors affecting performance. The machine utilization rate was 85.4%. Carriage inhaul (26%), outhaul (22%), and hooking (21%) were the most time-consuming work elements, while delays accounted for 14% of total cycle time. DBH, tree height, and yarding distance significantly influenced cycle time (R²=0.75) and productivity (R²=0.94), with DBH and yarding distance as dominant predictors. The average productivity was 14.8 m³SMH^-1, fuel consumption 10.59 l SMH^-1, and yarding cost 7.78 € m^-3, which is competitive compared to alternative steep-terrain systems. These results confirm that truck-mounted cable yarders, operated by experienced crews and integrated with processor-based landing equipment, represent a productive and cost-effective solution for final felling operations in mountainous conditions.

Mechanically harvested fresh tea leaves often show poor uniformity and cannot directly meet the raw material consistency required for famous tea processing. To address this problem, a compact fresh leaf sorting system suitable for hilly and mountainous tea gardens was developed for integration into a cross-ridge continuous harvesting machine. The system combined an electromagnetic vibrating screen and a two-stage differential-speed conveying unit to reduce leaf adhesion and accumulation. Tea-leaf targets were segmented using color-index-guided region extraction, and a lightweight classification network was designed to identify three grades: broken or single leaves, one bud with one leaf, and one bud with multiple leaves. A dynamic region-of-interest strategy was further introduced to improve the recognition of boundary leaves across adjacent frames. Coupled with multithreaded ONNX Runtime inference and an FPGA-based delayed air-jet execution unit, the system achieved closed-loop sorting from image recognition to actuator control. The proposed network contained about 3.7 M parameters and 0.04 G FLOPs, achieved 96.36% classification accuracy, and enabled 50-70 fps under CPU-only deployment, with an average sorting accuracy of about 82% and a single-channel capacity of 26.7 kg/h. These results demonstrate the feasibility of integrated mechanical harvesting and fresh leaf sorting for famous tea production.

An increasing number of manufacturers are entering the market with diverse precision weeder suspension systems, yet the actual performance benefits regarding soil adaptation remain scientifically unverified. This study develops a methodology to compare these systems by analyzing the force equilibrium under varying micro-topography. A critical challenge in precision weeding is maintaining a consistent work effect. As the tine deflects to follow the soil profile, the engagement geometry changes, shifting the ratio of vertical to horizontal force components. This study integrates quasi-static bench test data into a theoretical framework to investigate actual vertical pressure as a function of displacement and system hysteresis. Results identify significant differences in force-displacement characteristics among mechanical, pneumatic, and hydraulic systems. Theoretical modeling suggests that static force uniformity cannot guarantee consistent soil pressure if geometric angle shifts are not compensated during deflection. Consequently, a dynamic testing methodology is proposed using a decoupled dual-axis force-sensing frame to measure horizontal and vertical force vectors simultaneously via load cells. This research provides a fundamental basis for evaluating current implements and the necessary data for developing active tine-pressure control systems capable of responding to changing soil conditions in sustainable agriculture.

Abstract. Over 525,000 hectares of Germany’s forest require urgent reforestation due to drought, storms, heat events, and pests, with climate change and skilled labor shortages in an aging population accelerating the problem, particularly in mountainous regions, where tree planting is currently only manually possible.

The DraAuf project addresses this challenge by developing an autonomous system consisting of a 30kg planting robot carrying tree saplings, tethered to a heavy-lift drone that carries the robot to pre-mapped planting sites.

Due to the drone’s 8.2 kW hover power consumption, total planting cycle time is crucial for system efficiency. The drilling mechanism must balance minimal mass with maximum drilling speed and reliability in varying soil conditions. A dedicated outdoor test stand equipped with multi-axis force sensing and closed-loop feed control enables systematic comparison of rotary impact drilling against conventional drilling. Impact drilling achieves up to 4.5x reduction in continuous reaction moment while enabling equivalent or improved drilling speed and increasing system robustness against jamming. Peck drilling and active debris removal improve consistency of hole depth and diameter.

The complete DraAuf system combines LiDAR data, AI-based site selection, and autonomous planting. Terrain and environmental data are integrated into interoperable digital databases for further forestry operations.

Balancing agronomic sustainability with resource optimization requires advanced automation systems based on high-performance computer vision and deep learning. This study evaluates and compares state-of-the-art object detection models for on-canopy olive identification under realistic orchard conditions, with the aim of supporting reliable yield estimation. The analysis was conducted on an in-field RGB image dataset acquired under variable viewing angles and natural illumination. A major challenge was the selected phenological stage, in which green drupes must be detected against green foliage. This condition creates severe chromatic camouflage and low target-background contrast, providing a demanding benchmark for model robustness. The comparison included Faster R-CNN, YOLO-based models, the anchor-free FCOS architecture, and the transformer-based RT-DETR model. Performance was assessed in terms of Precision, Recall, mean Average Precision (mAP), and inference speed. RT-DETR achieved the highest mAP, indicating that its global attention mechanism is effective in handling chromatic ambiguity. Among convolutional detectors, FCOS reduced detection redundancy in dense clusters and outperformed Faster R-CNN. YOLO models provided the highest inference speed, confirming their suitability for real-time edge applications. Overall, the results highlight a trade-off between robustness and efficiency: transformer-based models perform best in visually complex conditions, whereas lightweight detectors remain preferable for latency-sensitive tasks.