Empirical Trials on Unmanned Agriculture in Open-Field Farming: Ridge Forming

Abstract

The decreasing rural population and migration to urban areas for high-tech opportunities have weakened the agricultural labor force. While data technology has been adopted in protected agriculture, numerous challenges remain in field agriculture. In this study, we focus on one of the fundamental steps of field operations, i.e., ridge forming, specifically in unmanned agriculture. We compared the performance of a conventional tractor with an implement to that of a ridge-forming robot. The operation data were collected using an acquisition system, and a comparison between the results of both methods was conducted. Additionally, we analyzed the linearity of autonomous driving and the expenses associated with the selected operation. Our findings indicate that the developed robot for ridge forming caused less torque damage and achieved a more accurate target soil depth, with a linearity performance showing a distance error of only 0.267 m. Furthermore, it eliminated the need for hiring an operator and significantly reduced fuel consumption, which accounts for 50.81% of the operational expenses. These results suggest that field operations can be effectively replaced by autonomous systems, and further research on unmanned agriculture is warranted.

1. Introduction

The agricultural industry has recently undergone significant reconstructions in both field and protected agriculture [1]. Alternative energy sources are increasingly replacing fossil fuels in machinery [2] due to the effects on the environment from agricultural emissions. Additionally, migration from rural to urban areas is accelerating [3,4], indicating that effective labor is decreasing, and the climate crisis is driving the development of greenhouse technology [5]. The aging farming population [6] further highlights the importance of robotic technology in agriculture. Given that farming poses physical [7,8] and mental risks [9], the trend of declining rural populations [10] is likely to continue. Consequently, experts are increasingly focusing on automated and digital agriculture [11].

Protected agriculture offers significant advantages, primarily because it is less influenced by external factors, such as the climate crisis, compared to field agriculture [12,13]. Advanced technologies allow for flexible responses and adaptations. For example, strategies for optimizing indoor environments [14] and wireless data acquisition systems using IoT and sensors enable remote monitoring services for farmers [15]. Despite these advancements, the accessibility of developed technology is not appropriate for all countries, only the advanced nations [16,17]. Also, the development of fully automated systems remains an ongoing challenge with the integration of advanced sensing technologies, the quality of intelligent decision-making algorithms, and the precision of automation [16].

While advanced technology in protected agriculture is successfully reducing the need for human labor, field agriculture still lags behind due to precision challenges. The irregular terrains in farmland increase the complexity of determining the recognition and manipulation of crops [18], and the poor knowledge of and lack of interest in current technologies of aging farmers also hinder their advancement of agriculture [19]. These challenges affect every step of field agricultural operations: (1) fertilizing, (2) rotary tilling and soil disinfection, (3) ridge forming, (4) planting, (5) management, (6) harvesting, and (7) transport. In detail, variable-rate fertilizer application shows fluctuating coefficients of variation, indicating a need for machinery redesign [20]. Constructing datasets with qualified data requires long periods of time, and the uncertainty of achieving it always exists [21]. Task allocation systems for robotic spraying have demonstrated reliable performance but still face collision risks, necessitating safety measures [18]. Robots in field agriculture also play crucial roles in vision and monitoring. They can oversee the location and density of planted seeds during the initial stages [22]. Vision systems often require Lidar sensors to detect unspecified obstacles [23]. However, these systems face endurance and precision challenges, particularly in detecting numerous obstacles and maintaining performance when working outdoors, where thermal issues can arise. For the harvest, the use of costly articulated arm and multi-joint robots for picking crops, coupled with computers and cameras for autonomous driving and AI technology for crop identification, has demonstrated the potential for automation. However, improvements in speed and quality are still needed [24,25].

Despite these challenges, agricultural automation is expected to be achieved through multi-robot systems, where robots collaborate and share data on soil, environment, and field conditions. This collaboration will enable a more flexible and scalable field [26]. Consequently, time-consuming tasks, such as data collection and processing, and predictable physical works can be replaced by achieving agricultural automation [27]. Then, fully automated agriculture can be accomplished in the areas of preparation, cultivation, and management [28].

Previous studies have identified autonomous robot systems as key to reducing labor needs in agriculture and achieving automated agriculture. Therefore, the effectiveness of robots at each stage of the field must be evaluated. As a primary step, this study focuses on the ridge-forming stage, comparing the performances of conventional methods with a developed robotic system. The results of this comparison are used to discuss the feasibility of replacing human labor with robots for this specific operation.

2. Materials and Methods

2.1. Robot Setup

For the field test, we utilized a developed robot and an attached implement. The robot features a continuous track with a length and width of 1200 and 180 mm, respectively. The robot’s overall dimensions (length, width, and height) are 3977, 1720, and 1460 mm, respectively, with a total weight of 1350 kg. It is powered by four 12 V, 63 Ah batteries. The attached implement (BG600, Bulls Co., Ltd., Daegu, Republic of Korea) is a standard ridge-forming machine with a length, width, and height of 1400, 1200, and 720 mm, respectively. Further details are provided in

Table 1. Specifications of the robot and implement used for a field test.

Machine	Specification		Value
Robot	Company		Sungboo Co., Ltd. (Chilgok-gun, Gyeongsangbuk-do, Republic of Korea)
	Model		SB-2000
	Frame	Dimension, mm (L × W × H)	3977 × 1720 × 1460
		Weight, kg	1350
	Velocity, km/h		1.5~3.0
	Type of brake		Electronic
	Maximum distance of point search, mm		2000
	Distance between identified point, mm		200~1000
	Battery	Voltage, V	12
		Amplifier Hour, Ah	63
		Quantity, ea	4
	Track	Dimension, mm (L × W)	1200 × 180
Implement	Company		Bulls Co., Ltd. (Seongju-gun, Gyeongsangbuk-do, Republic of Korea)
	Model		BG600
	Dimension (L × W × H)		1400 × 1200 × 720
	Weight, kg		350
	PTO revolution speed, RPM		440~800
	Number of furrows, ea		6
	Target structure of ridge	Width, mm	1200
		Depth, mm	150~200

.

Figure 1 illustrates the pre-setup for data acquisition from the robot and implement. To ensure accurate robot driving linearity, GPS data were collected using an RTK sensor (UM982, Unicore Communications, Inc., Beijing, China), and motion data were gathered using a motion tracker (MTi-1, Xsens, Enschede, The Netherlands). These data were integrated via an embedded system consisting of a Raspberry Pi 4 and stored on a laptop at the field site. The acquired implement data comprised three major parts: (1) torque, (2) revolution speed of the PTO shaft, and (3) GPS location.

2.2. Autonomous Driving Setup

The robot was operated using waypoint data generated from the GPS coordinates of the test field. The setup process for autonomous driving is outlined in Figure 2 and described sequentially. (A) Setup: The robot loaded the waypoint data and located the start point. The PTO shaft began to rotate in idle mode. (B) Initialization: After loading the data, the robot moved to the start point and leveled itself both horizontally and vertically for the operation. (C) Set off: The robot performed ridge forming along a straight line as per the imported data. Upon reaching the end point, the implement was raised, and the RPM of the PTO shaft and the number of ridges formed were calculated to ensure zero errors. After completing these calculations, the system shut down.

For autonomous driving, maintaining linearity in driving was a crucial parameter. The vehicle’s steering needed to respond promptly to real-time data, and precise control would enhance linearity. This concept can be described using the kinematic model below [29,30]:

$$\dot{c}={\displaystyle \frac{v·\left(cos\theta \right)}{1-p\left(c\right)y}},$$

(1)

$$\dot{y}=v·sin\theta ,$$

(2)

$$\dot{\theta }=v·\left[{\displaystyle \frac{tan\delta }{L}}-{\displaystyle \frac{p\left(c\right)·cos\theta }{1-p\left(c\right)·y}}\right],$$

(3)

where c is the coordinate on the path, v is the robot’s velocity, y and $\vartheta$ are the lateral and angular deviations of the robot along the path, respectively, and L is the wheelbase of the robot.

Based on this kinematic model, the equation for controlling the robot, without incorporating additional factors except for the attachment of electro-hydraulic cylinders for adjusting the robot’s behavior, is given as follows [31,32]:

$$\begin{gathered} \delta (y,\vartheta )={tan}^{-1}\left(L·\right.{\displaystyle \frac{{cos}^3\theta }{{\left(1-y·p\left(c\right)\right)}^2}}·\left(-K_d\left(1-y·p\left(c\right)\right)·tan\theta \right)-K_p·y \\ +p\left(c\right)·\left(1-y·p\left(c\right)\right)·{tan}^2\theta )+{\displaystyle \frac{p\left(c\right)·cos\theta }{1-y·p\left(c\right)}}, \end{gathered}$$

(4)

where K_d and K_p are the constants for tuning the robot’s driving behavior with a controller.

2.3. Cost Analysis Framework

To evaluate the replaceability of the developed robot, the investment cost advantage was analyzed from an agronomic perspective. Using a conventional agricultural machine as a comparative method, we first analyzed the total cost by determining the machine’s burden area using Equation (5). Next, we examined both the fixed and variable costs associated with field operations, including employment costs. Given that the developed robot is not yet a commercial product, we assumed a purchasing cost of KRW 200,000 for both the conventional machinery and the robot developed in this study.

$$A_c=A_e·T={\displaystyle \frac{SWUD}{10}}·{\epsilon }_f·{\epsilon }_u·{\epsilon }_d,$$

(5)

where $A_c$ is the workable burden area for a machine; $A_e$ is the effective workable ratio; $T$ is the available time during field operating season; ${\epsilon }_f$ is the operation efficiency; ${\epsilon }_u$ and ${\epsilon }_d$ are the actual operating time and days, respectively; $S$ is the travel speed of the vehicle; $W$ is the width of the implement; $U$ is the daily operating time; and $D$ is the number of days in the field operating season.

2.4. Field Test

The field test of the ridge-forming robot was conducted at 555-9, Deokpyeong-ri, Yongam-myeon, Seongju-gun, Gyeongsangbuk-do, South Korea (latitude 35°50′35.9″ N, longitude 128°18′11.8″ E) on 18 June 2024 (Figure 3). The field comprised sandy loam topsoil and sandy subsoil. The ridge-forming operation was performed for a 10 m length and repeated three times. The same operation was conducted using a conventional method for comparison. For the conventional method, an agricultural cultivator (KM-2000, TYM Co., Ltd., Seoul, Republic of Korea) with a length, width, and height of 2650, 1400, and 1900 mm, respectively, and a rated power of 14 kW, was used, equipped with the same implement as the robot.

3. Results and Discussion

3.1. Operation Data Analysis

During the operation, the PTO revolution speed, PTO torque, and ridge-forming depth were analyzed using both the conventional method and the developed robot. The PTO revolution speed was consistently maintained at 450 RPM, and the vehicle’s velocity was 1.5 kmh⁻¹. However, the torque data fluctuated more with the conventional method compared to the developed robot. The average torque for the ridge-forming operation was 5.89 ± 1.40 and 5.40 ± 2.19 kg_f·m for the conventional method and robot, respectively. The conventional method showed higher deviation due to the driver’s intuitive adjustments to the implement depth, which resulted in a deeper ridge than the target depth. Although this might seem like better performance, it leads to unnecessary wear and tear on the machinery, reducing its maintenance efficiency. In contrast, the robot initially formed a shallower ridge but soon achieved a consistent ridge depth of 17.87 ± 0.88 cm between 2 and 9 m of travel. The ridge depth at the end of the travel was 15.6 cm (Figure 4). This demonstrates the robot’s ability to maintain a more consistent depth, minimizing unnecessary damage to the equipment.

3.2. Linearity Performance

The linearity performance of the developed robot during ridge forming is illustrated in Figure 5. To assess this objective, the acquired GPS-RTK data were compared with the baseline, which is an ideal straight line between the start and end points. From the data collected during repeated operations, the location gap between the baseline and the robot’s autonomous driving was 4.84 × 10⁻⁴ ± 1.83 × 10⁻⁴ m. This represents the vertical distance between the actual and ideal locations. When converted to standard SI units, the deviation is ~0.267 m. However, these results do not account for external factors, such as field slope, sensor performance errors, friction coefficients, and tire deflection. By excluding these factors, the linearity could potentially be improved further. Despite these considerations, the developed robot demonstrates the potential to replace conventional methods for ridge forming.

3.3. Cost Analysis

Assuming a travel speed of 1.5 km/h for both the conventional method and the developed robot, the time required for ridge forming on 1 ha was 416.66 h. The employment cost for operating agricultural machinery was KRW 15,000/day, with a workable time of 8 h/day. Therefore, the employment cost for ridge forming on 1 ha was KRW 781,240. The total expense of using the conventional method for ridge forming includes the analyzed labor cost and the sum of fixed and variable costs (

Table 2. Results of cost analysis.

		Conventional Method	Developed Robot
Fixed cost (A)		7413.33	7413.33
	Depreciation	1583.33	1583.33
	Repair	1200.00	1200.00
	Interest	4410.00	4410.00
	Tax	-	-
	Insurance	20.00	20.00
	Housing	200.00	200.00
Variable cost (B)		10,228.75	8675.00
	Annual operating time, h	100	100
	Fuel, L	13.51	-
	Lubrication	2.03	-
	Labor cost	18.75	18.75
	Usage cost per hour	68.00	68.00
Sum of A and B		17,642.08	16,088.33
Workable burden area, ha		17.94	17.94
Annual usage cost, thousand KRW/ha		983.39	896.78
Assistant employ cost		781.24	-
Total, thousand KRW/ha		1764.63	896.78

). The fixed costs for both methods were assumed to be the same, but the variable costs were higher for the conventional method due to fuel consumption. Additionally, the conventional method incurred employment costs, while the developed robot did not. Consequently, the required cost for ridge forming was 1,764,630 and 896,780 KRW/ha for the conventional method and the developed robot, respectively. Comparing these costs, using the developed robot for ridge forming can reduce expenses by 50.81% per hectare.

4. Conclusions

While data technology has revolutionized unmanned agriculture in protected systems, field agriculture faces challenges posed by external factors, necessitating the consideration of replacing manual labor in every operational step. This study focused on analyzing the replaceability of a fundamental step, ridge forming, by comparing operation data and expenses between conventional and autonomous driving robots and evaluating the linearity of autonomous driving.

In terms of operation data, ridge forming was executed with a maintained PTO revolution speed, revealing less torque damage when using the developed robot compared to the conventional method. The conventional method exhibited higher torque due to deeper-than-targeted soil depth during operation, highlighting limitations in precision when performed manually. The linearity of autonomous driving demonstrated reasonable performance, with a distance error of 0.267 m relative to the baseline. Further enhancements in autonomous driving performance could be achieved by addressing the discussed external factors. Moreover, using the developed robot for ridge forming resulted in a significant 50.81% reduction in expenses, amounting to 896,780 KRW/ha in field agriculture.

Based on the findings of reduced costs, improved operational precision, and demonstrated linearity in autonomous driving, the replaceability of ridge forming in field agriculture has been substantiated. Future endeavors should focus on advancing unmanned agriculture across all operational facets.