Unmanned Agricultural Machine Operation System in Farmland Based on Improved Fuzzy Adaptive Priority-Driven Control Algorithm

Abstract

Autonomous driving technology for agricultural machinery can maximise crop yield, reduce labour costs, and alleviate labour intensity. In response to the current low degree of automation and low tracking accuracy of driving paths in agricultural equipment, this research proposes an unmanned agricultural machinery operating system based on an improved fuzzy adaptive PD control algorithm. Firstly, mechanical kinematic models and fuzzy adaptive control algorithms are introduced to achieve autonomous driving, and parameter settings and speed adjustments are made while considering errors. Secondly, in the autonomous driving operation system, taking a certain rice machine as an example, perception information, trajectory design, dynamic control, operation supervision, and remote control design are carried out. The experimental results show that the improved fuzzy algorithm exhibits smaller deviation results in driving path tracking, with an average error between the actual path and the expected path of less than 0.001 m. In different testing scenarios, compared with the actual control results, the maximum deviation of the control system platform in straight sections is less than 2.8 m, which is more stable. More than 95% of the lateral deviation results in the road sections are within 0.11 m. And the tracking distance error of the proposed method in the straight and curved segments is relatively small, far smaller than other comparative algorithms. The unmanned agricultural machinery operation system proposed in this study can significantly improve the efficiency and accuracy of agricultural machinery work, promote the development of intelligent and modern agricultural machinery, and provide reference value and important contributions to social and economic development as well as the progress and promotion of related technologies.

1. Introduction

With the advancement of industrialisation and urbanisation, a large number of labourers are gradually transferred from the countryside to the city, and the shortage of rural labourers makes the search for diversified means of agricultural production mechanisation gradually become an important element in promoting agricultural development. Agricultural mechanisation is the key to agricultural modernisation and sustainable development, which not only has a profound impact on agricultural production but also promotes the development and upgrading of related industries [1]. The traditional agricultural equipment has a low degree of automation, needing professional drivers to operate, with high labour costs, and the quality of operation is easily affected by the level and fatigue of drivers, which makes it difficult to ensure the operational efficiency and intelligence level of agricultural production [2]. Therefore, the design of an automatic driving and operation system of agricultural machinery is the key to alleviate the shortage of rural labour. Agricultural mechanisation, as an important means to promote the improvement in agricultural production efficiency, is gradually moving towards intelligence and automation, especially the research and application of unmanned agricultural machinery, which has become a hotspot and trend in the field of agricultural mechanisation. At present, the automatic driving technology of agricultural machinery mainly focuses on path planning, obstacle avoidance, operation control, etc. Although significant progress has been made in recent years, there are still problems, such as low precision and poor adaptability to complex environments [3,4]. He et al. [5] established a kinematic model for agricultural machinery based on posture correction to address the issues of sideslip and slippage in muddy, hard-bottomed, and uneven paddy fields. This model is based on an attitude prediction control algorithm for linear model design and path tracking control research of paddy field agricultural machinery. The results show that the average root mean square error of this method in three-line straight path tracking is 0.043 m, which can effectively suppress lateral position mutations and improve control accuracy. Ji et al. [6] proposed a new adaptive second-order sliding mode control method to solve the path tracking control problem during unmanned tractor operation. By using an improved increased power integrator and adaptive mechanism, an ASOSM controller was designed. The results showed that this method can effectively improve the chattering problem in traditional sliding mode control, and the control method has good applicability in vehicle models. Ge et al. [7] proposed a novel adaptive sliding mode control for path tracking of unmanned agricultural vehicles, which uses a sliding mode observer to achieve system state estimation and adaptively adjust the uncertainty limit of lateral stiffness. This method exhibits good robustness, and its simulation results also indicate good performance. Previous methods have shown that traditional control algorithms often struggle to cope with changing operating conditions and environmental disturbances, which can affect the efficiency of agricultural machinery applications, especially in complex agricultural environments. However, in the past, there has been relatively little discussion on the agricultural machinery autonomous driving problem in the agricultural working environment, and the control algorithms involved mostly rely on model parameters. The actual agricultural machinery operation is more complex, and overly complex algorithm models may reduce operational efficiency. Therefore, research is conducted on the overall system architecture design based on traditional agricultural machinery, and in the autonomous driving part, fuzzy control algorithms and kinematic models are introduced to optimise control parameters through adaptive mechanisms in order to improve adaptability to complex environments and control accuracy. Fuzzy control algorithms, with their strong ability to handle uncertainty and fuzzy information, can effectively plan and design path results in agricultural machinery autonomous driving. The innovation of this research lies in two aspects. One is the combination of fuzzy logic and adaptive control, which enables the controller to flexibly adjust control parameters in the face of changing environments, improving computational efficiency while ensuring system adaptability and responsiveness. The second is to study the overall performance and integrated design of unmanned agricultural machinery operation systems, which can provide reference and guidance for the development of intelligent agricultural machinery and the enrichment of mechanical control theory. This research contribution lies in the fact that the proposed method can better consider the uncertainty of agricultural machinery parameters and changes in road interference, and the proposed nonlinear control method based on multi-deviation feedback can ensure overshoot during autonomous driving of unmanned agricultural machinery. The entire design of the driving and operation control system can achieve full automation of operations. This research is divided into four parts in total: the first part is a content review of the current technical literature related to unmanned agricultural machines, the second part is a methodological discussion of both system architecture and algorithm design, the third part is an experimental validation of the improved fuzzy adaptive control algorithm proposed by this research, and the last part is a summary of the whole unmanned agricultural machine system in the farmland.

2. Literature Review

In the 21st century, with the development of global positioning and navigation technology, the research on unmanned agricultural machines based on satellite positioning and navigation has also made some progress. For example, Yin et al. [8] proposed a path planning method for unmanned agricultural machines, which creates a straight-line reference path by giving two points and determines the target path based on the initial state of the vehicle and the work task. Beloev et al. [9] introduced a small agricultural robot based on artificial intelligence algorithms, which was designed as an end-user autonomous mobile system that is capable of self-localisation and can map or inspect specific agricultural areas. In order to reduce the cost, manpower, and other resources in field agriculture, Ma et al. [10] proposed a low-cost and energy-efficient local motion planner based on a depth vision camera, low-range hardware, and a two-layer control algorithm for autonomous navigation implementations in vineyards. Zhang et al. [11] proposed a novel fully autonomous aerial reconnaissance method based on reinforcement learning and convolutional neural networks, whereby a portion of a field is sampled to sense and predict the crop health of the whole field, thus extending the battery life. Kanagasingham et al. [12] developed a new crop row detection algorithm by integrating GNSS, compass, and machine vision into a rice field weeding robot to achieve fully autonomous navigation for weeding operations. The experimental results proved that the system has good performance at low weed concentrations, with a heading compensation accuracy of less than 2.5° and an average deviation from the ideal path of 45.9 mm. Faryadi et al. [13] proposed an algorithm for unmanned agricultural machines applied in agriculture based on reinforcement learning. The system can deploy an unmanned farm machine to map plant rows, find obstacles, and define areas of interest in the field (e.g., areas of high water stress).

Conker et al. [14] used the method of geometric modelling to derive the parameter motion error, slip coefficient, and tipping angle that affect the accuracy and stability of robot motion. Combining the advantages of proportional–integral–derivative control and fuzzy control, an eight-wheeled omni-directional mobile unmanned agricultural machine based on a fuzzy adaptive proportional–integral–differential (PID) control algorithm was proposed, and trajectory tracking control experiments were carried out to validate the superiority of the fuzzy adaptive PID control method. Mahmoodabadi et al. [15] proposed a ¼ vehicle model with an active suspension system for an optimal fuzzy adaptive PID controller. Song et al. [16] designed a variable fertiliser application system based on conventional PID control, fuzzy control algorithms, and genetic algorithms and used prescription maps to configure the speed and cutting width of the fertiliser application equipment in order to accurately control the amount of granular fertiliser to be applied to the field. The experimental results show that the system can well simulate the precise control and uniform application of cart granular fertiliser variables. Gao et al. [17] proposed a variable-theory domain fuzzy controller control algorithm in order to achieve fast and accurate adjustment of tractor ploughing depth control to adapt to complex and variable agricultural environments. Cheng [18] based their study on the fuzzy proportional–integral differentiation of the temperature control in an agricultural greenhouse, which was the object of the study, through mathematical expressions to construct the greenhouse temperature model. The results of the study found that fuzzy temperature control has the advantages of a short response time and stable temperature control effect, and it is the optimal intelligent control mode for agricultural greenhouses. Xiao et al. [19] proposed a method of applying the fuzzy PID control method to the tillage depth adjustment system of a tillage machine to achieve automatic control. The experimental validation results of this system show that the fuzzy PID control system can meet the requirement of ploughing depth consistency during operation by increasing the soil fragmentation rate by 3% on the basis of reducing the ploughing depth stability change by 24%. Kherkhar et al. [20] proposed using an Interval Type-2 Fuzzy Proportional Differential Controller to achieve position and attitude tracking control of quadcopter unmanned aerial vehicles (UAVs) in order to ensure their dynamic control stability. The results show that the control structure can adjust the parameter evaluation gain well and has a good control effect. To improve the stability of the path tracking system of the intelligent weed killer, Liu et al. [21] determined the control strategy of the fuzzy adaptive proportional–integral–derivative (PID) algorithm and constructed the mathematical model and control system model for path tracking. The results show that this method has a better response time and path tracking control effect compared to the PID controller and can effectively improve the steering control accuracy of the weeding machine. Ulu et al. [22] proposed a neural network-based method to adjust the optimal PID control parameters for the trajectory control limitations of unmanned micro aerial vehicles, solving the problem of PID control being limited by environmental characteristics. The results show that this method has a low trajectory error, can adapt well to different flight environments, and has good application performance. The PID controller is one of the commonly used controllers in unmanned aerial vehicle systems. Amertet et al. [23] added a hybrid fuzzy PID control to control the state of a quadcopter. The results show that this method has better control performance than other traditional methods and can effectively improve the altitude and airspeed of the aircraft (

Table 1. Differences in application performance of different methods.

Differences in Application Performance of Different Methods
Literature	Advantage	Shortcoming
Yin et al. [8]	The planning process of a straight reference path is intuitive and can quickly adjust the target path in the initial state	There are limitations to the path in complex scenarios; not considering the impact of terrain or environmental changes
Beloev et al. [9]	The autonomy of exercise is guaranteed; a mobile system with strong adaptability can effectively respond to environmental changes	May rely on accurate algorithms and data support; the training requirements for the algorithm are too high
Ma et al. [10]	Economy; can effectively integrate deep visual information and action planning	May be limited by the performance of low distance hardware; real-time performance is easily affected
Zhang et al. [11]	Reinforcement learning and convolutional neural networks for automation and predictive analysis	Network training is difficult and requires a lot of data and resources
Kanagasingham et al. [12]	Capable of completely autonomous navigation	The complexity of system integration and its dependence on multiple sensors
Faryadi et al. [13]	Strong adaptability in complex environments; can define areas of interest	Processing delays may affect real-time response; poor winning ability
Conker et al. [14]	Enhanced control accuracy and stability	May be overly dependent on specific parameters; difficulty in debugging
Mahmoodabadi et al. [15]	Can provide theoretical support for dynamic control in complex environments	Dependence on the suspension system may result in additional costs.
Song et al. [16]	Good control performance	Variables are difficult to accurately calibrate
Gao et al. [17]	Capable of adapting to complex and ever-changing agricultural environments	Timely control feedback requires high system requirements
Cheng [18]	Stable control effect	Not universally applicable; the construction process is quite complex
Xiao et al. [19]	Automatic control enhances work efficiency	The experimental results may be limited by specific conditions
Kherkhar A et al. [20]	Strengthen dynamic control stability; enhanced the adaptability and flexibility of the system	Requires high computing resources; dependent on multiple environmental parameters
Liu J et al. [21]	Improved the stability of the path tracking system	Poor universality
Ulu B et al. [22]	Improved the performance of PID control, which helps to improve the accuracy of trajectory control	High sensitivity to changes in environmental characteristics; more data are required for network training
Amertet S et al. [23]	Improved attitude control performance	System complexity

).

In summary, for unmanned agricultural machinery algorithms, most scholars use positioning algorithms and modelling control algorithms for research, which may perform well in specific environments or tasks, but less consideration is given to the stability of control algorithms applied to agricultural equipment. Therefore, this study proposes an improved parameter-adjustable, multi-deviation feedback fuzzy control method and presents a complete design scheme based on the agricultural machinery drive and operation control system to ensure its performance accuracy and application effect. The method proposed by this research institute has a simple design, is more adaptive and flexible, and can better meet the requirements of modern agricultural automation operations.

3. Methods

3.1. Design of Unmanned Agricultural Machine Operation under Improved Fuzzy Adaptive Algorithm

The prerequisite for unmanned agricultural machinery operation is to achieve autonomous driving, where the automatic steering and speed conditions of the agricultural machinery are key to ensuring driving quality and effectiveness. A stable and accurate driving direction can include the machine driving along the planned operation path. The design of an autonomous driving simulation model using agricultural machinery kinematic models is researched and a control algorithm based on fuzzy adaptive control that considers multi-deviation feedback is proposed. A framework for an unmanned driving operation system to better reduce its error situation in the control process is built.

3.1.1. Kinematic Model and Improved Adaptive Algorithm Design

The automatic driving control algorithm of unmanned agricultural aircraft can be divided into two parts: path tracking and travelling speed regulation, in which the precise path tracking algorithm can ensure that the vehicle moves along the planned path accurately and correctly, avoiding the safety hazards caused by deviation from the route and adapting to the complex changes in the environment and terrain. The adjustment of the travelling speed can improve the work efficiency under the condition of mastering various dynamic information, effectively reduce fuel consumption, and guarantee the safety of travelling while taking into account the operating speed [24]. This study ignores the role of suspension and does not take into account the existence of the side-slip and lateral inclination of the vehicle and the influence of the steering system. Considering all practical factors can make the model very complex, thereby increasing the computational complexity and algorithm design difficulty. Therefore, in the design process, this study ignores the suspension effect and vehicle tilt influence to facilitate the calculation and analysis of the model, reduce uncertainty, and focus on factors related to motion control. The inclination of vehicles does indeed affect their driving conditions on different terrains, but the research mainly analyses the application of agricultural machinery, and the overall height difference of farmland terrain has a relatively small impact on vehicle inclination. At the same time, the auto drive system researched and designed can adjust the changes caused by the terrain through damping and other adjustment systems to ensure the stability of the operation. With the help of the two-wheeled vehicle model to achieve the establishment of the kinematic model of unmanned agricultural machinery, the direction of the coordinate system in the kinematics model is due east and due north, in which the counterclockwise direction of the heading angle is positive. The lateral deviations to the right and left of the path are distinguished as negative–positive. When the front wheel steering of agricultural machinery adopts the Ackermann steering mechanism, the rotation angles of the left and right front wheels are different during steering. Therefore, when studying the kinematic model of agricultural machinery, the kinematic model of agricultural machinery is simplified as a two-wheeled vehicle model. The front wheel angle sensor is fixed to the fixed rod of the steering mechanism, and the sensor rotation link is installed in the middle position of the steering mechanism rotation rod. The feedback of the front wheel angle information from this position is more accurate. The simplified kinematic equations of the model are easy to calculate and analyse, and without considering other factors, the model can reduce parameters, making system analysis and algorithm control more efficient, and can describe the planar motion state of agricultural machinery well [25]. The model is shown schematically in Figure 1.

In Figure 1, $\zeta$ represents the front wheel angle, $\gamma$ represents the steering angle deviation, and $\theta$ represents the heading angle. The heading angular velocity in this model can be expressed as Equation (1).

$$\dot{\theta }=\left(v\tan \delta \right)/L$$

(1)

In Equation (1), $v$ represents the vehicle speed, $\delta$ represents the front wheel angle of the control variable, and $L$ represents the wheelbase of the farm machine. The front wheel steering angle is usually measured relative to the longitudinal centreline of the vehicle. The front wheel steering angle represents the angle between the front wheel and the longitudinal centreline of the vehicle body, or the steering angle of the front wheel. In the process of constructing the kinematic model, it is assumed that both the vehicle body and wheels are rigid, without considering factors such as elastic deformation. The speed of the vehicle remains relatively stable, while only the change in steering angle is considered during the turning process, and the position of the centre of mass of the vehicle does not change with the action of external forces during the motion process. The kinematic model of the farm machine can be expressed as Equation (2).

$$\left[\begin{array}{l}\dot{x}\\ \dot{y}\\ \dot{\theta }\\ \dot{\delta }\end{array}\right]=\left[\begin{array}{l}v\sin \theta \\ v\cos \theta \\ \left(v\tan \theta \right)/L\\ 0\end{array}\right]+\left[\begin{array}{l}0\\ 0\\ 0\\ 1\end{array}\right]\varpi$$

(2)

In Equation (2), $\dot{x}$ denotes the component of velocity $v$ at the rear axle of the vehicle due east, $\dot{y}$ is the component of velocity due north, $\dot{\delta }$ is the front wheel angle vector, and $\varpi$ is the steering speed. The lateral deviation of this farm machine can be expressed as Equation (3).

$$\left|d\right|=\sqrt{{\left(x_m-x_t\right)}^2+{\left(y_m-y_t\right)}^2}$$

(3)

In Equation (3), $x_t$ denotes the horizontal coordinates of the point $t$ on the desired path, and $y_t$ denotes the vertical coordinates of the point $t$ on the desired path. $x_m$ is the horizontal coordinates of the position $m$ where the vehicle is located, and $y_m$ is the vertical coordinates of the position $m$ where the vehicle is located. Subsequently, based on the kinematic model, this study builds a path tracking steering simulation model based on MATLAB, see Figure 2. MATLAB models can quickly develop prototypes and test them, and their built-in mathematical and engineering functions are suitable for testing agricultural machinery models proposed in research. Compared to other simulation software such as Python, LabVIEW, etc., it provides a more efficient, intuitive, and systematic solution, which is particularly effective for developing complex control systems and conducting detailed data analysis. Moreover, MATLAB’s interactive features and powerful visualisation capabilities make it particularly outstanding for monitoring and evaluating the performance of control systems, making the verification and display of simulation results intuitive and easy to understand, such as path tracking effects, error curves, etc.

In this model, the lateral deviation and heading angle deviation are taken as the physical quantity feedback inputs, and the output result is the desired front wheel angle, and the deviation in the vehicle trajectory is displayed according to the tracking effect on the desired path. This model has flexibility, intuitiveness, and strong scalability in design and simulation and can be optimised and debugged using rich tools and resources. The key to path tracking steering is to calculate the motion and attitude information from the sensors. Then, the steering wheel control system executes the corresponding motion based on the calculated cornering results, and the accuracy is evaluated by the lateral deviation results. However, the vehicle steering needs to take into account the influence of the lateral deviation and heading angle deviation, but the deviation results are more easily affected by subjective and objective factors. So, this study considers the fuzzy adaptive scheduling (priority-driven scheduling, PD) control algorithm to achieve the adjustment of the deviation results, which ensures the stability and rapidity of the path tracking algorithm. PD (proportional–differential) control and PID (proportional–integral–differential) control are two common feedback control algorithms. The difference lies in whether they contain an integral term (I). The integral term in the PID controller can respond by integral deviation, which is affected by cumulative error interference and affects the efficiency of algorithm processing. There is no obvious steady-state error in the path tracking control of actual agricultural machinery, and the reduction in integral terms is not necessary. Its reduction can also simplify control algorithms and reduce computational burden. And in the research using PD control, nonlinear control algorithms are also introduced to eliminate errors, which can ensure algorithm accuracy. By using simulation models and setting the initial heading angle deviation and lateral deviation, agricultural machinery movement speed, and vehicle wheelbase, it was found that when the differential adjustment coefficient is small, the online speed of agricultural machinery is faster, but the stability is poor. According to PD control theory, increasing the proportional adjustment coefficient can make the system response faster, while increasing the differential adjustment coefficient can reduce system overshoot and enhance stability. Therefore, research is conducted on setting fuzzy rules to ensure that the control algorithm has good path tracking accuracy.

If the lateral deviation is “small” and the heading angle deviation is “small”, then increase the proportional gain and differential gain. If the lateral deviation is “medium” and the heading angle deviation is “medium”, then slightly increase the proportional gain and differential gain. If the lateral deviation is “large” or the heading angle deviation is “large”, then decrease the proportional gain and differential gain. The adjustment of the proportional gain and differential gain is to improve the dynamic response of the system and reduce overshoot. The specific fuzzy control rules are shown in

Table 2. Fuzzy control rules.

Control Parameters		Lateral Deviation
		−5	−2.5	0	2.5	5
Heading angle deviation/°	−5	1	1.7	2.3	1.7	1
	−2.5	1.7	2.3	2.9	2.3	1.7
	0	2.3	2.9	3.5	2.9	2.3
	2.5	1.7	2.3	2.9	2.3	1.7
	5	1	1.7	2.3	1.7	1

.

The domain for setting the control parameters is [1, 3.5], the quantisation factor is 1, and the quantisation level can be divided into {1,1.7, 2.3, 2.9, 3.5}. The domain for lateral deviation and heading angle deviation is [−0.5 m, 0.5 m] and [−30°, 30 °], and the quantisation level is divided into {−5, −2.5, 0.5}. The membership functions for the control parameters, lateral deviation, and heading angle deviation are all Gaussian functions, and the membership functions for each variable are shown in Figure 3.

This study uses lateral deviation and heading angle deviation as the inputs for fuzzy controllers. The parameters of the PD controller are taken as the output of the fuzzy controller and input the control rules in the fuzzy logic design tool. The sensors collect the information about the current state and operating environment of the agricultural machinery, and the fuzzy controller receives input signals and processes them using fuzzy logic. The signals are fuzzified based on a preset fuzzy set and rule library. After fuzzy reasoning, the generated fuzzy output is converted into an accurate control signal, and its output value is converted into a control command or control signal.

The front wheel angle can be transformed into a mathematical expression under fuzzy control, see Equation (4).

$$\delta =K_{p}d+K_{d}v\phi$$

(4)

In Equation (4), $K_d$ denotes the differential adjustment coefficient of the PD control algorithm, and $K_p$ is the proportional adjustment coefficient. The range of the front wheel angle of the agricultural crop machine is limited to plus or minus 90°, and the output results are nonlinearly processed with an inverse tangent function to obtain the tracking algorithm under the controller, see Equation (5).

$$\delta \prime =\arctan (K_{p}d+K_{d}v\phi )$$

(5)

In Equation (5), $\arctan$ is the arctangent function, and $\delta \prime$ is the front wheel angle under control processing. Based on the above, the tracking model for the control simulation can be obtained, see Figure 4.

When the current and subsequent path tracking deviations of the crop machine are opposite, it will lead to positive and negative fluctuations in the lateral deviation results. In order to simplify the control algorithm process and improve the performance of the automatic driving of the farming machine, this study is based on the relationship between the two input deviations and proposes a nonlinear control algorithm combining the lateral deviation and heading angle deviation since the steering angle error is eliminated. Therefore, the desired front wheel steering angle control can be expressed as Equation (6).

$$\delta ″=\phi +\arctan {\displaystyle \frac{k^{\prime }d}{v}}$$

(6)

In Equation (6), $k^{\prime }$ is an adjustable control parameter. The result of the lateral deviation of the tracking path of the agricultural machine is related to the amount of control of the turning angle. So, this study considers the reduction in the desired path deviation and the oscillatory situation during tracking by adjusting the turning angle of the vehicle, so the steering deviation of the agricultural machine is replaced by the deviation at the front wheels, and Equation (7) is obtained.

$$d_{front}=d-L\sin \theta$$

(7)

In Equation (7), $d_{front}$ is the lateral deviation at the front wheel. And the partial integration link is introduced to reduce the Vitae error to obtain the corner equation under the deviation combination control, see Equation (8).

$$\delta ‴=\phi +k_1\arctan {\displaystyle \frac{k_2(d-L\sin \theta )}{v}}+k_i{I}_{\mathrm{int}}$$

(8)

In Equation (8), $k_1$ denotes the weight of the position deviation feedback gain on the steering angle, $k_2$ denotes the weight of the motion state on the steering angle, $k_i$ denotes the weight of the integral feedback gain on the steering angle, and $I_{\mathrm{int}}$ is the partial integral. A farming machine does not operate in a completely parallel section; it needs to perform turnaround steering to complete the whole section of the path. The focus of attention is on its speed and stability when turning and switching sections, and the automatic turning lies in identifying whether the farm machine has reached the boundary of the ground headland area, so the distance between the position of the farm machine and the boundary can be calculated and the distance result can be compared with the threshold value to carry out the command judgement.

3.1.2. Design of Automatic Driving Operation Control System of Agricultural Machine

When the agricultural machine carries out an unmanned operation, it needs to be equipped with corresponding sensors, main controllers, execution parts, and other systems, in which the sensors and main controllers mainly analyse the vehicle’s operation information and control instructions, including the positioning information, heading and attitude, steering, acceleration and deceleration, etc., and the operation execution parts and terminal system mainly carry out the execution of the received instruction information and interactive control [26]. In the automatic navigation and operation monitoring system of agricultural machinery, the five key components include the perception information system, the trajectory design system, the dynamic control system, the operation monitoring system, and the remote control system. This research selected the Shanghai Shidar 2BDXZ-12CP (20) rice hole live broadcast machine as the research object. Its sowing method is pit eye seeding and natural drop seeding, and the sowing part is hydraulic. It can choose different rows and bottom plate structures according to the needs of the rice varieties, and its combination with different power units (Y-type or I-type) can adjust the hole distance. Combined with a synchronous side fertilisation device and an unmanned driving system, it can further improve the mechanisation of agricultural sowing. The architectural idea of its overall scheme is shown in Figure 5.

In Figure 5, the perception information system is mainly a sensor system, and the sensor device is responsible for providing necessary information about the vehicle’s position, attitude, and front wheel steering angle for the control system as reference feedback. The sensor system includes positioning sensors (differential positioning (RTK-GNSS) devices), vehicle inclination sensors (inclination sensors based on accelerometer and gyroscope integration), and angle sensors (DWQT-RS485-G-360-38 model sensors). The responsibility of the core control unit is to process the data from the sensor devices and the feedback from the power actuator units and to calculate the required amount of action of each power actuator unit according to the control algorithm. The calculated action quantity is transmitted to the subsystem of each power actuator unit via a bus or serial bus. The power actuator unit mainly consists of three parts: steering control, throttle control, and operation mode switching, which are responsible for manipulating the steering wheel, throttle, and operation mode selector, respectively. Each power actuator unit is equipped with an exclusive underlying controller, which drives the corresponding mechanical components according to the received control signals to achieve the functions of steering adjustment, throttle position adjustment, and operation mode switching [27]. The user operation management interface mainly realises the functions of remote monitoring of the state of the agricultural machine, setting operation parameters, and sending core control commands, etc., and it can exchange data with the main controller in the automatic planning of parameters. In the positioning sensing part, the traditional position estimation method based on the global positioning system is easily affected by external environment obstruction and leads to the generation of error results. Although magnetic induction positioning can avoid being affected by obstacles, light, and weather and other disturbances, it relies heavily on the deployment of the magnetic nails, which is only limited to specific places; inertial navigation, in the process of positioning calculations, has an accumulation of errors, which need to be calibrated with the help of the global positioning [28,29]. Considering the fact that the environment faced by farming machines during operation is mostly open farmland, this study collects the vehicle’s coordinates, speed, and other information with differential global positioning. In order to ensure the smoothness of signal reception, the antenna is mounted on the roof of the vehicle and then the differential positioning coordinates output from the mobile station can be obtained with the help of antenna point coordinates [30,31]. When the operating environment of the agricultural machine is not flat, the positioning coordinates output by the system will have a large error with the actual coordinate position, so the vehicle position coordinates need to be corrected, and its calculation formula is shown in Equation (9).

$$\left\{\begin{array}{l}{x}_m=x_m\prime +H\sin \phi \cos \theta -H\sin \psi \sin \theta \\ y_m=y_m\prime +H\sin \phi \cos \theta -H\sin \psi \sin \theta \end{array}\right.$$

(9)

In Equation (9), $(x_m,y_m)$ is the coordinates of the corrected point, $(x_m\prime ,y_m\prime )$ is the coordinate result of the differential positioning measurement, $H$ is the antenna mounting height, $\phi$ is the lateral inclination angle, $\psi$ is the pitch angle, and $\psi$ is the vehicle heading angle. The output positioning results can be processed with the help of the equation to reduce the tilt error when different tilt angles are considered. The main controller drives the steering wheel to rotate when calculating the desired turning angle of the farming machine, and in the kinematic modelling, the conversion of the turning angle is mostly carried out with a two-wheeled vehicle. Equation (10) is the equivalent converted expression for the turning angle of the front wheels.

$$\delta ={\displaystyle \frac{{\delta }_L+{\delta }_R}{2}}$$

(10)

In Equation (10), $\delta$ is the virtual front wheel angle of the model, ${\delta }_L$ is the left front wheel, and ${\delta }_R$ is the right front wheel. The angle of rotation of the differential lever position can be measured with the help of the angle sensor so as to effectively obtain the feedback information of the front wheel steering of the vehicle.

In order to facilitate the realisation of automatic steering of the vehicle, this study provides a motor power source for the steering wheel and the wheel turning angle joint main controller in the controller cycle by changing the message situation to achieve speed control. The throttle pedal in the electric actuator contracts and extends during the movement, so this study focuses on the controller’s conversion module to achieve, based on the feedback actuator information, the formation of closed-loop control of the throttle. The speed of agricultural machinery is adjusted by the throttle position, and the more the throttle pedal is pressed, the faster the vehicle’s speed. To achieve automatic driving of agricultural machinery, a speed regulation and control system has been researched and designed. In this system, the electric push rod controls the depression of the accelerator pedal through a steel cable connected to the accelerator pedal. When the electric push rod contracts, the accelerator pedal is pressed down; when extended, the accelerator pedal returns upward due to the action of the spring. The push rod is equipped with a sliding rheostat inside, and the slider changes with the movement of the push rod. Therefore, the voltage of the sliding rheostat can be measured through the AD conversion module of the controller to obtain the position information of the push rod, which is used for closed-loop control of the throttle position. The speed controller receives throttle position instructions from the main controller, uses the push rod position feedback read by the AD module, and adjusts the push rod to the desired position through PID control algorithm to achieve automatic speed adjustment. The actual position of the throttle will also be transmitted back to the main controller through the CAN bus. At the same time, this study focuses on the selection of agricultural machines for console switching toggles through the main controller to achieve the control of the toggle and automatic switching. Figure 6 shows the schematic diagram of the construction of the system parts of the agricultural machine.

During the operation of agricultural machinery, the lifting of the rear operating mechanism is adjusted by the operation switch lever on the control panel. The farmland operation mechanism has two fixed positions, and only when it is in the lower position can the drive shaft mesh with it to ensure its normal operation. The switching of the lever can control the operating status of the agricultural machinery. For this purpose, a lifting mechanism controlled by a hydraulic valve is studied and designed. Due to the fact that the lever has only two positions, the operation start–stop controller controls the forward and reverse movement of the electric push rod to achieve the lifting and lowering of the mechanism. The limit switch equipped on the push rod can help it switch between two extreme positions, thereby precisely controlling the lifting and lowering of the working mechanism. In the terminal design part, the relevant parameters of agricultural machines and the operation parameters are considered, and the parameter results are synchronised to the main controller to achieve the management and analysis of the information data. The terminal part is mainly an interactive platform that can be used for operation management, and its contents include several parts, such as operation parameter management, operation environment information management, operation path planning, monitoring and track reality, remote control, historical data, and communication management connection. The serial communication in this system is based on Visual Studio 2019 and developed using a 64-bit serial communication module. The baud rate for serial communication is set to 115,200 bps. The upper computer software system reads the information of the angle sensor and the positioning information of the satellite navigation receiver through RS232 communication, completes real-time communication with the main controller, and uses the NMEA0183 protocol standard to obtain the required positioning data. When obtaining program information, the Cserial Port class calls the thread creation function in the Windows API to create a thread for the serial port. Before closing the serial port, existing serial port threads should be deleted to avoid memory overflow. The controller is implemented using the Controller Area Network (CAN) bus, which reads information from the tilt sensor and sends control messages to the lower computer. Based on the received information, the controller can control the vehicle. When the communication is interrupted, the agricultural machine should stop the current operation, in which the control commands such as the emergency stop, start–stop, key parameter setting, etc., are realised with the help of the transmission control protocol, and the remote control is realised by uploading or downloading the telegrams. Operation spacing is the main parameter. This study focuses on the agricultural machine’s driving and operation work, first providing the user with the necessary data on the terminal. And then, based on the operation between the section of the turnaround and the distance calculation formula, the driving path is planned. Equation (11) is the formula for calculating the spacing between operation rows.

$$D={\displaystyle \frac{\left|b_i+1-b_i\right|}{\sqrt{q^2+1}}}$$

(11)

In Equation (11), $b_i+1$ denotes the intercept of the linear equation at the point $(i+1)$, $b_i$ denotes the intercept of the linear equation at the point and $i$, and $q$ is the slope of the line.

4. Results and Discussion

Driverless Agricultural Machine System Testing and Application Analysis

This research aims to analyse the trajectory tracking effect of the designed agricultural machinery system in order to better verify its application effect in the operation process. At the same time, this study chooses the paddy field environment as the experimental scenario because factors such as muddy soil conditions, uneven ground, and the presence of water in the paddy field environment require the operation, stability, and traction of agricultural machinery. Moreover, the common rice working environment is mostly the paddy field environment, and choosing this environment, one can better analyse the performance and application adaptability of agricultural machinery. The experimental site for the paddy field environment was selected as an ordinary paddy field in Yexie Town, Songjiang District, Shanghai, with uneven ground. This farmland often suffers from the issue of traditional rice direct seeders getting stuck in the mud during operation, and there is a certain degree of error in the row spacing during sowing. Especially in the turning section, there is a significant deviation between the planning results and the actual results when the live broadcast machine achieves lifting and lowering while turning the steering wheel. This study builds an autopilot and operating system based on a rice planter, with an experimental Android user system as the management terminal system, and uses radio and Bluetooth to build a remote communication link. The technical parameters of the main controller in the system platform are a dual-core Cortex-A8@800MHz processor, Linux operating system, and 128MBDDR2 and 256MBSLCNAND memory and storage and the Ethernet interface is 10/100 M adaptive. And the parameters of the navigation and sensor devices used in the experiment are shown in

Table 3. Parameters of navigation and angle sensors.

Navigation Device Parameters
RTK plane accuracy	2.5 cm
Heading angle accuracy	0.1°
Speed measurement accuracy	0.01 m/s
Data output method	RS232 (Four routes)
Data refresh rate	10 Hz
Serial baud rate	115,200 bps
Parameters of angle sensor equipment
Range	0°~360°
Accuracy	0.1°
Resolving power	0.023°
Output method and frequency	RS485 (Two routes), 200 Hz

.

The proposed method and system of this study are examined in two scenarios, a concrete floor and paddy field environment, and the lateral deviation and heading angle deviation of the starting position of the agricultural crop machine from the AB line are set to 0.5 m and 0°, respectively, and the travelling speed is set to 0.8 m/s. The schematic diagram of the experimental site for agricultural machinery operation is shown in Figure 7.

Among them, the tracking control algorithm refers to path tracking control based on an unmanned navigation positioning system, and adaptive control is the PD algorithm. The proposed algorithm is a fuzzy adaptive PD algorithm that combines the multi-deviation feedback nonlinear control method with the PD algorithm. The direction deviation weight of the proposed adaptive fuzzy control algorithm is 1, the curvature deviation weight is 3, and the speed deviation weight is 0.05; the proportional adjustment coefficient in the PD control algorithm is 1.2, and the differential adjustment coefficient is 2.0. The preview distance of the pure tracking algorithm is set to 2 m. This study utilises the kinematic model of agricultural machinery and MATLAB tools for the simulation analysis. The initial lateral deviation is set to 0.5 m, the vehicle speed is set to 1 m/s, and the wheelbase of the agricultural machinery is set to 1.06 m. Figure 8 shows the results of the driving path tracking.

The results in Figure 8 show that the lateral deviation curves of the PD control algorithm under different travelling distances vary with large ups and downs, and their maximum and minimum values reach 0.046 and −0.032, respectively, with obvious fluctuations. The fuzzy adaptive PD algorithm, on the other hand, has a smaller deviation value in path tracking, and shows better stability overall, with its average absolute error reaching 0.0425. Considering the absolute value of deviation of the multi-deviation feedback control algorithm, the absolute value of deviation maxes out at 0.0243 m, and the average absolute error turns out to be less than 0.005 m, which is of high tracking accuracy. The result in Figure 8 indicates that traditional PD control algorithms use proportional and differential terms to respond to errors, making it difficult to handle the dynamic characteristics of nonlinear systems of agricultural machinery in operating environments. When there are disturbances (such as terrain undulations, soft soil, wind interference, etc.), it may lead to a decrease in tracking performance. The performance of the PD algorithm highly depends on the selection of control parameters. If the parameters are not properly adjusted to adapt to the specific conditions of agricultural machinery operation, it can also lead to poor control effectiveness. The control algorithm proposed in this study is a combination algorithm that can effectively handle uncertainty and adapt to complex environments. Then, the performance of the control algorithm proposed in this study is further compared with PD control [32] and adaptive control [33], and the results are shown in Figure 9.

In the results of Figure 9, under the concrete pavement (Figure 9a), the fuzzy adaptive algorithm proposed in this study has a faster response speed compared to the PD algorithm with fixed parameters. When the travelling distance is less than 10 m, the PD algorithm and the path algorithm have higher overshooting and their degree of undulation is higher. They then tend to stabilise with the increase in the travelling distance, but their average absolute deviation and maximum absolute deviation are significantly larger than that of the research-proposed algorithm. Under the paddy field environment (Figure 9b), when the system reaches the steady state, the maximum absolute deviation of the fuzzy adaptive control algorithm among them reaches 0.2016, and the corresponding differences with the PD parameter and tracking control algorithm are all above 0.13. The changing trend of this curve is basically floating at ±0.15 m deviation, and the result of the deviation between the maximum and the minimum is smaller than that of the other two algorithms. The average absolute deviation of the PD parameter and tracking control algorithms have mean absolute deviation results of 0.0715 and 0.0689, which are much larger than the algorithm proposed in this study (0.0216). The research-proposed algorithm has a smaller regulation coefficient, better overall response and adaptability, and higher robustness in complex operating environments. We compare the proposed control algorithm with traditional control algorithms, and the results are shown in Figure 10.

The results in Figure 10 indicate that when the starting point of the controller is at the origin, the traditional PID control system exhibits significant lateral deviation, and there is a significant discrepancy between its path planning results and the reference path. The lateral deviation under the adaptive fuzzy control proposed in this study is smaller than the fixed-time domain result, and the error value between it and the reference path is smaller. The above results indicate that adaptive fuzzy control has better adaptability than the traditional control methods. The performance of the system under different test scenarios is analysed to obtain Figure 11, which shows the driving of the farming machine under two scenarios: concrete and paddy field.

The results of the travelling trajectory (Figure 11) show a high degree of overlap between the actual path and the desired path, with an overall average error of less than 0.001. The bias data analysis of the results of the farming machine under path tracking was carried out and the results are shown in Figure 12.

In Figure 12, when in the concrete field, the maximum deviation of the straight-line section of the control system platform is 2.75 m, and the average value of the deviation of the turning section reaches 2.05 m. The large oscillation situation is mainly manifested in the turning section, and the stability of the overall control process is better. In the paddy field deviation results, the average value of the lateral deviation in the straight section is 2.03 m, and the average value of the deviation in the turning section is basically 2.54 m. According to the operation quality standard of the agricultural machine, it can be seen that the row spacing of the operation should be ≤5 cm, and the statistics of the experimental results show that the actual path deviation points account for less than 5 per cent of the actual number, which indicates that the design system can better meet the operational requirements and the qualification rate is high. The planning results of agricultural machinery operating in paddy fields are subject to certain deviations due to factors such as soil moisture and road surface smoothness. The difference between the two operating environments of agricultural machinery is due to the special nature of paddy fields. The soil in paddy fields usually has a higher moisture content, making the ground softer and smoother. The traction of agricultural machinery tyres is affected, which affects the accuracy of steering and the stability of driving. The water level in paddy fields may change due to irrigation, rainwater accumulation, or evaporation, directly affecting the difficulty of agricultural machinery operations and their adaptability to the ground. The driving of agricultural machinery in paddy fields requires sufficient power to overcome the resistance of water and the friction of muddy soil, and its power system will inevitably affect the path tracking results.

To further verify the adaptability of the fuzzy control method proposed by this research institute, obstacles were set up in farmland, which were, respectively, arranged in the first and second half of the farmland planning route. Subsequently, the adjacent method was used as a comparison algorithm to analyse the turning effect of the agricultural machinery operations, and the adjacent method [34] operated according to adjacent strips. The results are shown in Figure 13.

The red part in the picture represents obstacles. The results of the graph indicate that the fuzzy control method used in this study performs better than the adjacent method in terms of work area and work distance, with an average improvement of 9.5% and 8.4%. In addition, this method can effectively avoid obstacles, reduce the area of turning areas, and has strong adaptability. A farmland operation plan route can reduce sharp turns and drastic flight changes, thereby reducing the risk of collisions. However, the operation plan route of the comparative method has bends and partially dense situations. Curved operation routes may be limited by the terrain of farmland, surface obstacles, etc., which may cause the machinery to be unable to move in a straight line during operation, resulting in a high risk of collision. Frequent turns may cause the equipment to be unable to avoid ground obstacles due to the need to adjust the angle. The deviation situation of the two methods during farmland operation was analysed, and the results are shown in Figure 14.

The results in Figure 14 indicate that the proposed fuzzy control method has a smaller lateral deviation during operation than the adjacent method, and its deviation values on different bands are all at (−0.05,0), with an average lateral deviation value of −0.034. The lateral deviation results under the adaptation method vary significantly, especially near obstacles, with an overall average lateral deviation value of 0.042. The above results indicate that the proposed fuzzy control method exhibits good deviation adaptability and stable processing performance in job planning. The path tracking deviation results of different operation sections were analysed and the results are shown in

Table 4. Path tracking deviation results for different job segments.

Path Number	Lateral Deviation/m				Heading Angle Deviation/°
	Maximum Value	Minimum Value	Average absolute Value	Standard Deviation	Maximum Value	Minimum value	Average Absolute Value	Standard Deviation
1	0.076	−0.066	0.041	0.029	7.821	−4.941	0.028	2.133
2	0.102	−0.075	0.024	0.034	5.877	−6.864	0.021	1.892
3	0.095	−0.088	0.041	0.031	7.820	−4.822	0.017	1.945
4	0.103	−0.064	0.041	0.038	6.582	−7.167	0.017	1.714
5	0.061	−0.058	0.034	0.021	13.120	−5.732	0.011	2.432
6	0.068	−0.055	0.038	0.026	4.531	−6.035	0.015	1.521
7	0.103	−0.082	0.034	0.025	7.290	−5.082	0.021	2.140
8	0.045	−0.051	0.031	0.027	3.721	−6.786	0.008	1.273

.

The results in Table 4 show that the root mean square (RMS) and average absolute values of the lateral deviation for each path sequence do not exceed 0.042 m and 0.029 m, the average absolute error and RMS values of the overall heading angle results are 0.017 m and 1.88°, respectively, and more than 95 per cent of the segments have lateral deviation results that are within 0.11 m. The application evaluation of the two main agricultural environments under the fuzzy adaptive control algorithm used in this study is shown in Figure 15.

The work zone (or work area/work range) refers to the predetermined flight path or work area for unmanned aerial vehicles to perform agricultural operations. It is often divided based on the shape of the farmland and the surrounding environment. The selection and planning of the work zone are crucial for improving work efficiency and coverage. The results of Figure 15 indicate that when using the proposed fuzzy adaptive control algorithm to analyse the results of agricultural machinery operation planning, its tracking accuracy for natural and artificial agricultural environments is mainly above 70%, and the average clustering value is mainly distributed between 85% and 95%, which has good application adaptability. Subsequently, a farmland experimental environment was selected for further research on path control, and tracking experiments were conducted on agricultural machinery under different road conditions in the “S”-shaped experimental path. The instantaneous control error was recorded, and a forward distance of 0.5 m was set. The proposed method was compared with the efficiency-oriented path tracking control algorithm (EfiMPC) [35], the MPC path tracking control method based on agricultural machinery attitude (MPC) [5], and the improved A * algorithm combined with the fuzzy sliding mode variable structure control method (IA *—F-SMC) [36]. The tracking performance of the different algorithms under the control path of straight and curved segments is discussed separately. Figure 16 shows the control error results of the different tracking methods under a straight-line segment.

The results in Figure 16 indicate that the proposed method can effectively complete the tracking task of the entire path on the straight-line segment, and the motion trajectories are relatively close to the target path, with a direction roughly the same as the target path. The average distance error during tracking is 0.03 m, the standard deviation of the tracking distance error is 0.02 m, and the maximum distance error does not exceed 0.12 m. The tracking results of the IA *—F-SMC method and MPC method have significant differences compared to the target path, with their maximum tracking accuracy approaching 0.15 m. The tracking accuracy error of EfiMPC changes significantly, with a maximum value exceeding 0.15 m. Subsequently, the tracking error results of the control method in the curve path segment were analysed, and the results are shown in Figure 17.

The results in Figure 17 indicate that the proposed method has an average tracking error of 0.039 m, a standard deviation of distance error of 0.028 m, and a maximum error value of no more than 0.04 m during path tracking, demonstrating good path tracking performance. Secondly, the IA *—F-SMC algorithm performed well, with an average tracking error of 0.052 m. The standard deviation of distance error for both the EfiMPC and MPC methods exceeded 0.07 m, and the maximum tracking error values exceeded 0.16 m and 0.10 m, respectively, indicating a significant error margin compared to the proposed method.

5. Conclusions

This study analyses the automatic driving and operating system of unmanned agricultural machines and introduces an improved fuzzy adaptive control algorithm to enhance its operating accuracy. The designed system was experimentally analysed, and the results show that the PD control algorithm shows large ups and downs in the variation in lateral deviation curves at different driving distances, and its maximum and minimum values reach 0.046 and −0.032, respectively, with obvious fluctuations. The average absolute deviation exhibited by the proposed algorithm is less than 0.005 m, and the response speed is faster in the two test scenarios of concrete and paddy field. When the travelling distance is less than 10 m, the PD algorithm and the path algorithm have higher overshooting, with average absolute deviation values of 0.0715 and 0.0689, respectively, which are much larger than that of the research-proposed algorithm (0.0216). After the system reaches the steady state, the maximum absolute deviation of the fuzzy adaptive control algorithm reaches 0.2016, and the differences with the other algorithms are above 0.13, with smaller regulation coefficients and higher robustness. And in the concrete and paddy field scenarios, the average value of the deviation of the turning part of the control system platform reaches 2.05 m and 2.54 m, the root mean square and the average absolute value of the lateral deviation of each path sequence do not exceed 0.042 m and 0.029 m, and the average absolute error and the root mean square value of the overall heading angle results are 0.017 m and 1.88°, respectively. The proposed unmanned agricultural machine system in this study shows good stability and applicability during operation with high path tracking accuracy and control performance. Autonomous agricultural machinery requires a rapid response to environmental changes, real-time processing of sensor data, and path planning and control. Although the method proposed by this research institute has good performance, the randomness of terrain changes, weather conditions, and obstacles will increase the complexity of the agricultural environment. Therefore, strengthening the real-time recognition and adaptability of the algorithm is one of the challenges in the future. At the same time, dynamically adjusting the control system rules based on real-time environmental information and ensuring further coordination and communication of the overall intelligent control system are also technical challenges that the system needs to face in the future.