Next Article in Journal
Laser Stripe Centerline Extraction Method for Deep-Hole Inner Surfaces Based on Line-Structured Light Vision Sensing
Previous Article in Journal
Quality Evaluation for Colored Point Clouds Produced by Autonomous Vehicle Sensor Fusion Systems
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

ZPTM: Zigzag Path Tracking Method for Agricultural Vehicles Using Point Cloud Representation

1
School of Agricultural Engineering and Food Science, Shandong University of Technology, Zibo 255000, China
2
College of Biosystems Engineering and Food Science, Zhejiang University, Hangzhou 310058, China
*
Author to whom correspondence should be addressed.
Sensors 2025, 25(4), 1110; https://doi.org/10.3390/s25041110
Submission received: 28 December 2024 / Revised: 30 January 2025 / Accepted: 11 February 2025 / Published: 12 February 2025
(This article belongs to the Section Sensors and Robotics)

Abstract

:
Automatic navigation, as one of the modern technologies in farming automation, enables unmanned driving and operation of agricultural vehicles. In this research, the ZPTM (Zigzag Path Tracking Method) was proposed to reduce the complexity of path planning by using a point cloud consisting of a series of anchor points with spatial information, which are obtained from orthophotos taken by UAVs (Unmanned Aerial Vehicles) to represent the curved path in the zigzag. A local straight path was created by linking two adjacent anchor points, forming the local target path to be tracked, which simplified the navigation algorithm for zigzag path tracking. A nonlinear feedback function was established, using both lateral and heading errors as inputs for determining the desired heading angle of agricultural vehicles, which were guided along the local target path with minimal errors. A GUI (Graphic User Interface) was designed on the navigation terminal to visualize and monitor the working process of agricultural vehicles in automatic navigation, displaying interactive controls and components, including representations of the zigzag path and the agricultural vehicle using affine transformation. A high-clearance sprayer equipped with an automatic navigation system was utilized as the test platform to evaluate the proposed ZPTM. Zigzag navigation tests were conducted to explore the impact of path tracking parameters, including path curvature, moving speed, and spacing between anchor points, on zigzag navigation performance. Based on these tests, a regression model was established to optimize these parameters for achieving accurate and smooth movement. Field test results showed that the maximum error, average error, and RMS (Root Mean Square) error in the zigzag navigation were 3.30 cm, 2.04 cm, and 2.27 cm, respectively. These results indicate that the point cloud path-based ZPTM in this research demonstrates adequate stability, accuracy, and applicability in zigzag navigation.

1. Introduction

With the increasing application of cutting-edge technologies, precision agriculture has been experiencing significant development during the last decades [1,2]. Agricultural vehicles are becoming increasingly intelligent, with traditional agricultural vehicles gradually being replaced by smart ones [3,4]. Automatic navigation contributes as one of the core technologies in achieving precision agriculture, which has been widely used in agricultural vehicles for sowing, weeding, and harvesting to improve operation efficiency and reduce crop production costs [5,6,7].
The problem of automatic navigation for agricultural vehicles in complex field environments has received considerable research attention [8,9]. The purpose of automatic navigation is to control agricultural vehicles to travel accurately along predetermined paths and efficiently achieve full coverage of the operation area [10,11]. When the attitude information of agricultural machinery is accurately obtained through integrated navigation technology [12,13], the accuracy of agricultural vehicle automatic navigation is closely related to efficient path planning algorithms [14,15,16]. In order to deal with the challenges of path planning in complex environments, various path planning algorithms have been proposed. The A* algorithm effectively plans the optimal operation path for static operation scenarios with minimal cost, which has been widely used in the field of automatic navigation [17,18,19]. Jeon et al. [20] developed a novel path planner using the A* algorithm to generate an entry–exit path that enables the tractor to enter the farmland from the entrance, complete the agricultural task, and return. In two real paddy field tests, the entry–exit paths created by the proposed path planner effectively guided the tractor to reach the given target point with a lateral error of ≤ 8.5 cm. However, when the agricultural operation scene is large, and the navigation accuracy is high, the A* algorithm has the disadvantages of long search calculation times and high computer memory resources [21]. The pure tracking model is a geometric method to simulate the manual driving process, which uses the calculated arc to connect the vehicle body and the preview point [22,23,24]. A suitable local tracking path was planned by dynamically adjusting the forward-looking distance of the model according to the speed, lateral, and heading errors. Wu et al. [25] considered the impact of operating speed and target path curvature on the look-ahead distance of the agricultural vehicle, and the target path was planned by adjusting the look-ahead area and calculating the preview point. Curvature path tracking control experiments were conducted. The results showed that when the tractor was driving at speeds of 1.0 m/s, 1.5 m/s, 2.0 m/s, and 3.0 m/s, the average absolute errors were 2.7 cm, 2.7 cm, 3.3 cm, and 4 cm, respectively. He et al. [26] proposed a novel local adaptive tracking path planning method using pose error of agricultural vehicles by combining the pure tracking algorithm with fuzzy control. The straight-line path tracking tests in paddy fields were conducted to validate the stability and precision of the method. Compared with the traditional pure tracking algorithm, the proposed algorithm improves the performance of the straight-line path tracking of the tracked combine harvester. In recent years, meta-heuristic optimization algorithms have been widely applied to path planning due to their powerful global search capability and adaptability to deal with a wide range of constraints in complex farmland environments [27,28,29]. Khan et al. [30] proposed an RNN-based approach to the tracking control of redundant mobile manipulators, using a metaheuristic optimization process to find control parameters. This provides novel ideas for dealing with path planning in dynamic and complex farmlands.
In recent years, UAVs have been widely applied in various agricultural fields, including crop management [31,32], seed sowing, and pesticide spraying, due to their high efficiency and precision [33,34]. Wu et al. [35] proposed a navigation method that extracted the coordinate information of seedlings by utilizing UAVs to capture farmland images, recognizing the navigation path of seeding rows. The automatic navigation of the robot vehicle along the curved rice seedlings was realized in complex paddy field environments. Sun et al. [36] proposed an orchard path plan method based on UAV images to assist agricultural robots in pesticide spraying and fruit harvesting. For the automatic navigation of agricultural vehicles, path planning is required to cover both farmland and field roads, so that the agricultural vehicle can move along the field road to the entrance of the working area and return after completing its tasks in the fields. Therefore, the target path is usually complex and zigzag, rather than straight. The zigzag and changeable path should be reasonably planned to generate a smooth trajectory so that agricultural vehicles can track with minimum error. Therefore, we combine an efficient UAV with an image processing algorithm to plan the target path.
In this research, the ZPTM was proposed by using a point cloud to describe the target zigzag path. The point cloud contained anchor points with latitude and longitude, which were created from orthographic images obtained by an UAV and stored in turn as the navigation map. Zigzag path tracking was simplified by segment tracking between two adjacent anchor points. A nonlinear feedback function was established to determine the desired heading angle of agricultural vehicles using lateral and heading errors for highly precise path tracking. A GUI was developed to monitor the zigzag navigation process in real time and enable visualization of automatic navigation for agricultural vehicles. The path tracking parameters were optimized by establishing a regression model to explore their impact on zigzag navigation performance. Field tests were conducted to evaluate the stability and applicability of the proposed ZPTM.

2. Materials and Methods

A high-clearance sprayer equipped with an electronic HST (Integrated Hydrostatic Transmission), automatic throttle, and automatic navigation with a CAN (Controller Area Network) bus communication system was utilized as the test platform, as shown in Figure 1 [37]. Its main parameters are described in Table 1. The automatic navigation system was developed by SDUT (Shandong University of Technology, Zibo, China) and is composed of an RTK-GNSS (Real-Time Kinematic Global Navigation Satellite System) receiver, a navigation terminal, a remote controller, and an IMU (Inertial Measurement Unit). Control commands, including steering and speed, were issued by the automatic navigation system, which automatically guided the high-clearance sprayer to travel along the target path. The RTK-GNSS receiver integrated a GNSS positioning board UM982 by Unicore Technology (Beijing) Co., Ltd., Beijing, China, to obtain centimeter-level positioning and heading information. The IMU (a JY61P model by Shenzhen Wit-Motion Intelligent Technology Co., Ltd., Shenzhen, China) was used to sense the vehicle posture. The navigation terminal, with interfaces including CAN and an RS232 serial port, was capable of receiving sensor data, running the navigation program, and providing controls for basic operations, including command output, navigation process monitoring, path planning, and map management. The terminal was configured with an Intel® Core™ i5-13500H CPU @ 2.60 GHz and 16 GB RAM.

2.1. Path Planning

A DJI Phantom 4 RTK UAV was used to capture a series of aerial images with a resolution of 5472 × 3648 pixels at a height of about 60 m, flying at a speed of 4.7 m/s over the SDUT ecological unmanned farm (118°13′ E, 36°57′ N), as shown in Figure 2a. The aerial images, containing farmland, a field road, and a garage, were stitched to form an orthophoto with geographic information for extracting the target zigzag path, as shown in Figure 2b. Anchor points were selected sequentially from the orthophoto to create a zigzag path, ensuring that agricultural vehicles could be guided to start from the garage, run along the field road to the working area, and return after completing assigned tasks. These anchor points were treated as a point cloud Ω, and their global positions were stored in a navigation map file.
Ω = ω i | ω i E 3 , 1 i N ,
where ω i = ( L a t ,   L o n ,   C o d e ) contains the latitude, longitude, and operation information at each anchor point, and N is the number of anchor points in the point cloud.
The DPA (Douglas–Peucker Algorithm) [38,39] was utilized to remove redundant anchor points from the point cloud path to minimize the storage size of the map file while maintaining the original path shape, as shown in Figure 3. A minimum spacing e was given in the DPA to prevent over-simplification.
The original zigzag path consisted of multiple segments defined by connecting all anchor points in the point cloud Ω. The start and end points of the zigzag path were marked as retained anchor points, and a straight segment ω 1 ω N was drawn between them. The perpendicular distances Di from the other anchor points to this segment were calculated by Equations (2) and (3), and the maximum value Dmax was obtained.
D i = ω i ω 1 × ω i ω N · ω i ω N ω 1 ω N , 2 i N 1 ,
x i = R c o s L a t i · c o s L o n i y i = R c o s L a t i · s i n L o n i z i = R s i n L a t i ,
where R is the radius of the Earth, and ω i = ( x i , y i , z i ) is the cartesian coordinate converted from the geographical coordinate ω i = ( L a t i , L o n i ) .
In the case that Dmax was no more than the given threshold ε, all anchor points except the start point were considered to be insignificant and removed from the point cloud if the distance LA between the start and end points was no more than the minimum spacing e. If LA exceeded e, both the start and end points were kept, and the others were removed. In the case that Dmax exceeded ε, the anchor point ω k corresponding to Dmax was kept, and the zigzag path was divided into two sub-paths that are represented by Ω 1 and Ω 2 , which are defined in Equations (4) and (5), respectively. The sub-paths were processed iteratively using the DPA until there were no anchor points to be removed. In this way, a simplified zigzag path was obtained with fewer anchor points.
Ω 1 = ω i ω i E 3 , 1 i k
Ω 2 = ω i ω i E 3 , k i N
As shown in Figure 4, the zigzag path consists of eight segments formed by connecting nine anchor points, which is simplified by applying the DPA with a threshold ε of 1 m and a minimum spacing e of 1 m. By comparing the simplified zigzag path with the original zigzag path, the RMSE (Root Mean Squared Error), MAE (Mean Absolute Error), and ME (Maximum Error) are found to be 0.34 m, 0.23 m, and 0.86 m, respectively. These results indicate that the DPA effectively simplifies the zigzag path by giving certain constraints.

2.2. Point Cloud Path Tracking

In this research, the ZPTM was developed to guide an agricultural vehicle along the point cloud path by minimizing both lateral and heading errors. The desired steering angle is determined by calculating the lateral error and heading error of the agricultural vehicle relative to the point cloud path in real time. The anchor points ωi* = (Lat, Lon) near the vehicle is taken from the point cloud Ω to create a new subset Ω*, as defined in Equation (6).
Ω = ω i ω i E 2 , 1 i N
The target zigzag path, determined by the subset Ω*, is shown in Figure 5. The position C of the agricultural vehicle is obtained from an RTK-GNSS receiver. The anchor point ω c that is closest to position C of the agricultural vehicle is searched by using Equation (7).
ω c = ω c a r g m i n i C ω c 2 , ω c Ω
The local target path ω c ω c + 1 is constructed by connecting the nearest anchor point ω c and the next anchor point ω c + 1 . The lateral error ey is defined as the perpendicular distance from C to ω c ω c + 1 , as shown in Equation (8). η is the nearest point to the agricultural vehicle on the local target path.
e y = ω c ω c + 1 × ω c C ω c ω c + 1
The lookahead distance L is defined as the sum of the segments from the nearest point η to the target point Q. The target point Q is located between the anchor points ω k and ω k + 1 in the point cloud path. These anchor points are determined by Equation (9).
η ω c + 1 + i = c + 2 k ω i ω i 1 L η ω c + 1 + i = c + 2 k ω i ω i 1 L
The target point Q is represented by anchor points ω k and ω k + 1 , as shown in Equation (10).
Q = λ ω k + 1 λ ω k + 1 , 0 λ 1
The desired heading angle θ is given by Equation (11).
θ = c o s 1 C Q C Q · n ,
where n is the unit vector oriented to the north direction in the UTM (Universal Transverse Mercator) coordinate framework.
The heading error ∆θ is given by Equation (12).
θ = θ θ 0
The desired steering angle φ is calculated by lateral error ey and heading error ∆θ.
φ = K 1 e y + K 2 θ
where K1 and K2 are proportional gains.
The values of ey and Δθ represent both the magnitude and direction of the deviation relative to the target zigzag path. When the agricultural vehicle is to the left of the target zigzag path, ey is negative, and when it is to the right, ey is positive. When the heading of the agricultural vehicle deviates to the left relative to the vector direction of the target zigzag path, Δθ is negative, and when it deviates to the right, Δθ is positive.

2.3. Design of the GUI

To visualize the working process, a GUI was developed on the navigation terminal, which served as an interactive medium between the user and the navigation system. Controls were designed to help the user in path planning, navigation map managing, safety monitoring, and information acquisition, as shown in Figure 6. Navigation data were displayed on the left, including attitude and position information of the agricultural vehicle, navigation status, and operation width. The STOP/START button on the bottom right was pressed to switch between the manual and automatic modes.
To monitor the working process on the navigation terminal, different elements were displayed on its screen to represent the agricultural vehicle and the target zigzag path. Since the position coordinates obtained from the RTK-GNSS receiver are in the geographic coordinate system, an affine transformation is necessary to convert these coordinates and orientations from the global frame, defined by WGS84 (World Geodetic System 1984) and UTM, to the screen frame. This affine transformation enables the visualization of the target zigzag path and the agricultural vehicle heading on the screen, aligning the local target path with the top of the screen, as shown in Figure 7.
Both the UTM coordinate system and the screen coordinate system utilize a common origin O for their coordinate frameworks. The local target path P 1 P 2 , formed by adjacent anchor points, is rotated around the origin O by an angle τ so that its direction is the same as the UTM northing. Subsequently, in order to convert the rotated local target path P 1 P 2 from the UTM coordinate system to the screen coordinate system for full visualization, P 1 P 2 was scaled and translated with respect to the midpoint P′m of P 1 P 2 so that the midpoint P′m is located at the center of the screen. The scale factor s is determined based on the length of the local target path and the size of the screen, which is given by Equation (14). The translation vector (tx, ty) is calculated by Equation (15).
s = H s c r P 1 P 2 ,
t x = W s c r s x 1 + x 2 2 t y = H s c r s y 1 + y 2 2 ,
where the (x1, y1) and (x2, y2) are coordinates of anchor point P1 and P2, respectively.
Finally, the anchor points P1 and P2 from the UTM coordinate system are converted to anchor points Pscr1 and Pscr2 in the screen coordinate system using the affine transformation, as shown in Equations (16) and (17).
x s c r 1 y s c r 1 1 = s · c o s τ s · s i n τ t x s · s i n τ s · c o s τ t y 0 0 1 x 1 y 1 1 ,
x s c r 2 y s c r 2 1 = s · c o s τ s · s i n τ t x s · s i n τ s · c o s τ t y 0 0 1 x 2 y 2 1 ,
where (xscr1, yscr1) and (xscr2, yscr2) are the coordinates of Pscr1 and Pscr2 in the screen coordinate system.

2.4. Optimization of Zigzag Path Tracking Parameters

Compared with straight tracking, the navigation performance of zigzag paths was affected greatly by three path tracking parameters, including the path curvature φ, the moving speed v, and the spacing d between anchor points. Therefore, the RMS value of lateral errors between the actual trajectory and the target zigzag path was selected to explore the influence of φ, v, and d on the navigation performance, which is calculated by Equation (18).
e R M S = i = 1 N e i 2 N ,
where ei is the lateral error at the anchor point ωi.
Zigzag navigation tests were conducted for the same road at the ecological unmanned farm of SDUT using a high-clearance sprayer integrated with an automatic navigation system, as shown in Figure 8. During the test process, the values of three path tracking parameters were adjusted to investigate their relationship with lateral error, aiming to optimize the accuracy of zigzag navigation. Zigzag navigation maps were constructed by assigning diverse parameter values of curvature and spacing of anchor points. Considering the field road width and the minimum turning radius of the agricultural vehicle, the values of φ were set to 0.09 m−1, 0.11 m−1, 0.125 m−1, 0.17 m−1, and 0.25 m−1. The values of d were set to 1.0 m, 1.5 m, 2.0 m, 2.5 m, and 3.0 m. The high-clearance sprayer tracked the target zigzag path based on the proposed ZPTM and fed back the real-time position through the RTK-GNSS receiver. The high-clearance sprayer started from A′ and drove along the zigzag path AB at a constant speed, completing the turn at point B. The values of v were set to 3.0 km/h, 4.0 km/h, 5.5 km/h, 7 km/h, and 8 km/h, respectively. The navigation terminal logged the actual trajectories of the high-clearance sprayer, with the operator manually recording lateral errors at each anchor point.

3. Results and Discussion

3.1. Determination of Optimal Navigation Parameters

The actual trajectories following the five target zigzag paths from point A to B are shown in Figure 9. The high-clearance sprayer started from point A′ (588,528.67, 4,073,799.147) and moved along the target zigzag path AB toward B at a constant speed. The turn was executed precisely when reaching A and finished at B. From Figure 9, it can be seen that the actual trajectory of the high-clearance sprayer on these five target zigzag paths is generally smooth. Lateral errors from 28 sets of zigzag navigation tests under different navigation parameters were recorded and analyzed, as shown in Table 2. However, relatively larger lateral errors occurred for the zigzag paths with curvature of 0.25 m−1, mainly due to its proximity to the minimum turning radius of the high-clearance sprayer. During the turning process, the wheels need to be rotated to their limit position, making it difficult for the agricultural vehicle to quickly and accurately track the target zigzag path. For those five target zigzag paths with different curvatures from A to B, the minimum, maximum, minimum average, maximum average, minimum RMS, and maximum RMS values were 1.6 cm, 6.8 cm, 0.75 cm, 2.69 cm, 0.85 cm, and 2.72 cm, respectively. This indicated that the proposed ZPTM had good tracking accuracy to meet the requirements for zigzag paths with different curvatures.
Figure 10 shows a comparison between the RMS values for lateral errors obtained when the three path tracking parameters were changed. The RMS values significantly increased with the increase in the spacing of the anchor points and were proportional to the moving speed. Both excessively small and large curvature values resulted in increased RMS values. At a curvature of 0.125 m−1, the minimum RMS value reached 0.85 cm. The most significant variation in RMS values occurred with an anchor point spacing of 2 m, where errors ranged from 2.72 cm to 0.99 cm.
To obtain more robust analysis results, the ANOVA (Analysis of Variance) was conducted on the test results, and a regression model was established. The ANOVA results, after eliminating insignificant factors, are shown in Table 3. The results indicated that the regression model was statistically significant (F-value = 8.06, p-value < 0.05), and the effects of the curvature, moving speed, and the spacing of the anchor points on the lateral error of the zigzag navigation were statistically significant (p-value < 0.05). The regression model for the RMS value y was obtained as shown in Equation (19):
y = 0.170029 + 0.258064 v + 0.357539 d 1.51609 φ v + 34.72536 φ 2
The ANOVA results showed that spacing had the greatest impact on tracking accuracy based on the F-value. Path tracking parameters were optimized to minimize tracking errors. Based on actual tracking conditions and theoretical analysis, the constraint conditions are given by Equation (20).
y y m i n 0.09 φ 0.25 3 v 8 1 d 3
The values of the optimal path tracking parameters were finally determined as 0.125 m−1, 3 km/h, and 1 m for curvature, speed, and spacing, respectively.

3.2. Tests in the Field

To evaluate the proposed ZPTM for the following zigzag paths, field tests were conducted using the optimal path tracking parameters to complete the automatic navigation process with the high-clearance sprayer. The test scene included a garage, field roads, and the working area, as shown in Figure 11. After departing from the garage, the high-clearance sprayer made a left turn and traveled along the entry path, then made a right turn to reach the starting point and entered the first working line. Once the field operations were completed, the high-clearance sprayer made a right turn at the end point of the last working line and continued straight along the exit path to return to the garage.
Considering the position of the working line and the optimal path tracking parameters, the entry and exit paths were planned by manually selecting anchor points in the orthophoto at an anchor point spacing of 1 m. The curvature at turns was set to 0.125 m−1, and the paths were stored in the point cloud as a navigation map. After the navigation terminal successfully loaded the navigation map, the high-clearance sprayer automatically tracked the entry and exit paths at a speed of 3 km/h and recorded the actual trajectory in real time.
During field tests, the entry and exit paths consisted of four turns and two straight lines. The turns at the starting and end points of field operations were particularly crucial in affecting the accuracy of field operations and the return to the garage. Therefore, the zigzag navigation process at the turns of the entry and exit paths was the focus of attention. Figure 12 shows the actual trajectories of the high-clearance sprayer following the entry and exit paths and the enlarged images of the turns at the starting and end points. For the yellow trajectory obtained along the entry path, the high-clearance sprayer could follow the entry path and reach the starting point. Due to the influence of the terrain potholes at the junction of the field boundary and road, the high-clearance sprayer oscillated during the turn process. However, it was able to quickly adjust and smoothly track the entry path to enter the first working line. For the purple trajectory obtained along the exit path, including the last working line of the field, the high-clearance sprayer successfully returned from the end point through the exit path to the garage.
Table 4 shows the lateral errors at the entry path, exit path, and turns at the starting and end points, respectively. For the turn at the starting point, the maximum, average, and RMS values of lateral errors were 3.3 cm, 1.86 cm, and 2.14 cm, respectively. For the turn at the end point, the maximum, average, and RMS values of lateral errors were 2.4 cm, 1.75 cm, and 1.98 cm, respectively. For the entry path, the maximum, average, and RMS values of lateral errors were 3.3 cm, 2.04 cm, and 2.27 cm, respectively. For the exit path, the maximum, average, and RMS values of lateral errors were 2.4 cm, 1.82 cm, and 2.06 cm, respectively. The maximum, average, and RMS values of lateral errors were almost the same for the entry path and the turn at the starting point. Similarly, the maximum, average, and RMS values of lateral errors were almost the same for the exit path and the turn at the end point. The maximum values of lateral errors for the entry and exit paths appeared at the turns of the starting and end points, respectively. This indicated that the turns of the starting and end points of the field had an important impact on the zigzag navigation accuracy. For the zigzag navigation, the maximum, average, and RMS values of lateral errors were 3.3 cm, 2.04 cm, and 2.27 cm, respectively. These results confirmed that the proposed ZPTM effectively planned zigzag paths and guided the agricultural vehicle to cover the working area with high accuracy and stability.

4. Conclusions

In this research, ZPTM was proposed as a method to describe a zigzag path using a point cloud consisting of anchor points with spatial information. These anchor points are obtained from orthophotos taken by UAVs and stored in a navigation map. The desired heading angle of the agricultural vehicle along the local target path, linked to adjacent anchor points, was calculated based on both lateral and heading errors to achieve highly precise zigzag path tracking. A GUI was designed on the navigation terminal to visualize the zigzag path tracking process of agricultural vehicles. The proposed ZPTM was applied on a high-clearance sprayer with automatic navigation, and zigzag navigation tests were conducted to explore the impact of path tracking parameters, including path curvature, moving speed, and spacing between anchor points, on navigation performance. The optimal path tracking parameters were then determined and adopted for field tests. The results indicated that the proposed ZPTM in this research was of adequate stability and applicability for zigzag navigation.

Author Contributions

X.Y., J.D. and S.Y. conceived and designed the experiments; E.Z. and S.Y. performed the experiments; J.D. and Y.L. analyzed the data; S.Y. wrote the draft manuscript; X.Y., E.Z., J.D. and Y.L. revised the manuscript. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the National Key Research and Development Program of China (Grant No. 2021YFD2000502), the Key Research and Development Projects in Shandong Province (Grant No. 2022SFGC0201), the National Natural Science Foundation of China (Grant No. 32171910), the National Innovation Park for Forestry and Grass Equipments (Grant No. 2023YG08), and the Zhejiang Province Agricultural Machinery Research, Manufacturing and Application Integration project (Grant No. 2023-YT-06).

Institutional Review Board Statement

Not applicable.

Data Availability Statement

Data are contained within the article.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Fountas, S.; Mylonas, N.; Malounas, I.; Rodias, E.; Hellmann Santos, C.; Pekkeriet, E. Agricultural Robotics for Field Operations. Sensors 2020, 20, 2672. [Google Scholar] [CrossRef] [PubMed]
  2. Kutyrev, A.I.; Kiktev, N.A.; Smirnov, I.G. Laser Rangefinder Methods: Autonomous-Vehicle Trajectory Control in Horticultural Plantings. Sensors 2024, 24, 982. [Google Scholar] [CrossRef]
  3. Wang, N.; Yang, X.; Wang, T.; Xiao, J.; Zhang, M.; Wang, H.; Li, H. Collaborative path planning and task allocation for multiple agricultural machines. Comput. Electron. Agric. 2023, 213, 108218. [Google Scholar] [CrossRef]
  4. Zhang, S.; Liu, J.Y.; Du, Y.F.; Zhu, Z.X.; Mao, E.R.; Song, Z.H. Method on automatic navigation control of tractor based on speed adaptation. Trans. Chin. Soc. Agric. Eng. 2017, 33, 48–55. [Google Scholar]
  5. Xie, B.B.; Jin, Y.C.; Faheem, M.; Gao, W.J.; Liu, J.Z.; Jiang, H.K.; Cai, L.J.; Li, Y.X. Research progress of autonomous navigation technology for multi-agricultural scenes. Comput. Electron. Agric. 2023, 211, 107963. [Google Scholar] [CrossRef]
  6. Zhao, X.; Wang, K.; Wu, S.; Wen, L.; Chen, Z.; Dong, L.; Sun, M.; Wu, C. An obstacle avoidance path planner for an autonomous tractor using the minimum snap algorithm. Comput. Electron. Agric. 2023, 207, 107738. [Google Scholar] [CrossRef]
  7. Kim, W.-S.; Lee, D.-H.; Kim, T.; Kim, H.; Sim, T.; Kim, Y.-J. Weakly Supervised Crop Area Segmentation for an Autonomous Combine Harvester. Sensors 2021, 21, 4801. [Google Scholar] [CrossRef]
  8. Ye, L.; Wu, F.Y.; Zou, X.J.; Li, J. Path planning for mobile robots in unstructured orchard environments: An improved kinematically constrained bi-directional RRT approach. Comput. Electron. Agric. 2023, 215, 108453. [Google Scholar] [CrossRef]
  9. Urrea, C.; Muñoz, J. Path Tracking of Mobile Robot in Crops. J. Intell. Robot. Syst. 2015, 80, 193–205. [Google Scholar] [CrossRef]
  10. Liu, W.L.; Guo, R.; Zhao, J.Y. Predictive control method for the path tracking of agricultural machinery based on preview model. Trans. CSAE 2023, 39, 39–50. [Google Scholar]
  11. Sun, J.L.; Li, Q.S.; Ding, S.H.; Xing, G.Y.; Chen, L.P. Fixed-time generalized super-twisting control for path tracking of autonomous agricultural vehicles considering wheel slipping. Comput. Electron. Agric. 2023, 213, 108231. [Google Scholar] [CrossRef]
  12. Zhang, X.; Fu, Y.; Li, J.; Wei, Y.; Li, Y.; Zheng, L. Research on Combined Localization Algorithm Based on Active Screening–Kalman Filtering. Sensors 2024, 24, 2372. [Google Scholar] [CrossRef]
  13. Ji, P.; Duan, Z.; Xu, W. A Combined UWB/IMU Localization Method with Improved CKF. Sensors 2024, 24, 3165. [Google Scholar] [CrossRef] [PubMed]
  14. Takai, R.; Yang, L.L.; Noguchi, N. Development of a crawler-type robot tractor using RTK-GPS and IMU. Eng. Agric. Environ. Food 2014, 7, 143–147. [Google Scholar] [CrossRef]
  15. Li, S.; Zhang, M.; Ji, Y.H.; Zhang, Z.Q.; Cao, R.Y.; Chen, B.; Li, H.; Yin, Y.X. Agricultural machinery GNSS/IMU-integrated navigation based on fuzzy adaptive finite impulse response Kalman filtering algorithm. Comput. Electron. Agric. 2021, 191, 106524. [Google Scholar] [CrossRef]
  16. Jing, Y.P.; Li, Q.; Ye, W.S.; Liu, G. Development of a GNSS/INS-based automatic navigation land levelling system. Comput. Electron. Agric. 2023, 213, 108187. [Google Scholar] [CrossRef]
  17. Ou, Y.; Fan, Y.; Zhang, X.; Lin, Y.; Yang, W. Improved A* Path Planning Method Based on the Grid Map. Sensors 2022, 22, 6198. [Google Scholar] [CrossRef] [PubMed]
  18. You, Z.; Shen, K.; Huang, T.; Liu, Y.; Zhang, X. Application of A* Algorithm Based on Extended Neighborhood Priority Search in Multi-Scenario Maps. Electronics 2023, 12, 1004. [Google Scholar] [CrossRef]
  19. Wu, B.; Chi, X.; Zhao, C.; Zhang, W.; Lu, Y.; Jiang, D. Dynamic Path Planning for Forklift AGV Based on Smoothing A* and Improved DWA Hybrid Algorithm. Sensors 2022, 22, 7079. [Google Scholar] [CrossRef] [PubMed]
  20. Jeon, C.W.; Kim, H.J.; Yun, C.; Guang, M.S.; Han, X. An entry-exit path planner for an autonomous tractor in a paddy field. Comput. Electron. Agric. 2021, 191, 106548. [Google Scholar] [CrossRef]
  21. Xu, X.; Zeng, J.Z.; Zhao, Y.; Lv, X.S. Research on global path planning algorithm for mobile robots based on improved A*. Expert Syst. Appl. 2024, 243, 122922. [Google Scholar] [CrossRef]
  22. Ahn, J.; Shin, S.; Kim, M.; Park, J. Accurate Path Tracking by Adjusting Look-Ahead Point in Pure Pursuit Method. Int. J. Automot. Technol. 2021, 22, 119–129. [Google Scholar] [CrossRef]
  23. Nguyen, P.T.T.; Yan, S.W.; Liao, J.F.; Kuo, C.H. Autonomous Mobile Robot Navigation in Sparse LiDAR Feature Environments. Appl. Sci. 2021, 11, 5963. [Google Scholar] [CrossRef]
  24. Yang, Y.; Li, Y.; Wen, X.; Zhang, G.; Ma, Q.; Cheng, S.; Qi, J.; Xu, L.; Chen, L. An optimal goal point determination algorithm for automatic navigation of agricultural machinery: Improving the tracking accuracy of the Pure Pursuit algorithm. Comput. Electron. Agric. 2022, 194, 106760. [Google Scholar] [CrossRef]
  25. Wu, C.C.; Wu, S.X.; Wen, L.; Chen, Z.B.; Yang, W.Z.; Zhai, W.X. Variable curvature path tracking control for the automatic navigation of tractors. Trans. Chin. Soc. Agric. Eng. 2022, 38, 1–7. [Google Scholar]
  26. He, Y.Q.; Zhou, J.; Sun, J.W.; Jia, H.B.; Liang, Z.A.; Awuah, E. An adaptive control system for path tracking of crawler combine harvester based on paddy ground conditions identification. Comput. Electron. Agric. 2023, 210, 107948. [Google Scholar] [CrossRef]
  27. Lamini, C.; Benhlima, S.; Elbekri, A. Genetic algorithm based approach for autonomous mobile robot path planning. Procedia Comput. Sci. 2018, 127, 180–189. [Google Scholar] [CrossRef]
  28. Liu, Y.; Zhang, H.; Zheng, H.; Li, Q.; Tian, Q. A spherical vector-based adaptive evolutionary particle swarm optimization for UAV path planning under threat conditions. Sci. Rep. 2025, 15, 2116. [Google Scholar] [CrossRef] [PubMed]
  29. Wu, S.; Dong, A.; Li, Q.; Wei, W.; Zhang, Y.; Ye, Z. Application of ant colony optimization algorithm based on farthest point optimization and multi-objective strategy in robot path planning. Appl. Soft Comput. 2024, 167, 112433. [Google Scholar] [CrossRef]
  30. Khan, A.H.; Li, S.; Chen, D.; Liao, L. Tracking control of redundant mobile manipulator: An RNN based metaheuristic approach. Neurocomputing 2020, 400, 272–284. [Google Scholar] [CrossRef]
  31. Zhao, C.; Wu, D.; He, J.; Dai, C. A Visual Positioning Method of UAV in a Large-Scale Outdoor Environment. Sensors 2023, 23, 6941. [Google Scholar] [CrossRef] [PubMed]
  32. Wang, Q.; Yi, W. Composite Improved Algorithm Based on Jellyfish, Particle Swarm and Genetics for UAV Path Planning in Complex Urban Terrain. Sensors 2024, 24, 7679. [Google Scholar] [CrossRef] [PubMed]
  33. Zhang, B.Y.; Wu, W.B.; Zhou, J.P.; Dai, M.L.; Sun, Q.; Sun, X.G.; Chan, Z.; Gu, X.H. A spectral index for estimating grain filling rate of winter wheat using UAV-based hyperspectral images. Comput. Electron. Agric. 2024, 223, 109059. [Google Scholar] [CrossRef]
  34. Yang, J.K.; Zhai, Z.Q.; Li, Y.L.; Duan, H.W.; Cai, F.J.; Lv, J.D.; Zhang, R.Y. Design and research of residual film pollution monitoring system based on UAV. Comput. Electron. Agric. 2024, 217, 108608. [Google Scholar] [CrossRef]
  35. Wu, S.L.; Chen, Z.G.; Bangura, K.; Jiang, J.; Ma, X.G.; Li, J.Y.; Peng, B.; Meng, X.B.; Qi, L. A navigation method for paddy field management based on seedlings coordinate information. Comput. Electron. Agric. 2023, 215, 108436. [Google Scholar] [CrossRef]
  36. Sun, Q.L.; Zhang, R.R.; Chen, L.P.; Zhang, L.H.; Zhang, H.M.; Zhao, C.J. Semantic segmentation and path planning for orchards based on UAV images. Comput. Electron. Agric. 2022, 200, 107222. [Google Scholar] [CrossRef]
  37. Yin, X.; An, J.H.; Wang, Y.X.; Wang, Y.K.; Jin, C.Q. Development and experiments of the autonomous driving system for high-clearance spraying machines. Trans. Chin. Soc. Agric. Eng. 2021, 37, 22–30. [Google Scholar]
  38. Song, W.; Wang, C.; Dong, T.; Wang, Z.; Wang, C.; Mu, X.; Zhang, H. Hierarchical extraction of cropland boundaries using Sentinel-2 time-series data in fragmented agricultural landscapes. Comput. Electron. Agric. 2023, 212, 108097. [Google Scholar] [CrossRef]
  39. Lee, W.; Cho, S.-W. AIS Trajectories Simplification Algorithm Considering Topographic Information. Sensors 2022, 22, 7036. [Google Scholar] [CrossRef] [PubMed]
Figure 1. Components of test platform.
Figure 1. Components of test platform.
Sensors 25 01110 g001
Figure 2. The process of (a) image acquisition by DJI Phantom 4 and (b) zigzag path planning.
Figure 2. The process of (a) image acquisition by DJI Phantom 4 and (b) zigzag path planning.
Sensors 25 01110 g002
Figure 3. Flow chart of DPA.
Figure 3. Flow chart of DPA.
Sensors 25 01110 g003
Figure 4. An example of the DPA: (a) Original point-cloud path. (bd) Process using DPA. (e) Result after applying DPA. (f) Comparison of original and simplified point-cloud path.
Figure 4. An example of the DPA: (a) Original point-cloud path. (bd) Process using DPA. (e) Result after applying DPA. (f) Comparison of original and simplified point-cloud path.
Sensors 25 01110 g004
Figure 5. Point cloud path tracking algorithm: ω c 1 ,   ω c ,   ω c + 1 ,   ω k and ω k + 1 are anchor points in the point cloud path. C is the movement center of the agricultural vehicle chassis, representing the vehicle position. η is the nearest point from the vehicle position to the point cloud path. Q is the target point. θ0 is the heading angle of the agricultural vehicle. θ is the desired heading angle of the agricultural vehicle. ∆θ is the heading error defined as the angle between vector C Q and the heading of the vehicle. ey is the lateral error defined as the perpendicular distance from C to vector ω c ω c + 1 .
Figure 5. Point cloud path tracking algorithm: ω c 1 ,   ω c ,   ω c + 1 ,   ω k and ω k + 1 are anchor points in the point cloud path. C is the movement center of the agricultural vehicle chassis, representing the vehicle position. η is the nearest point from the vehicle position to the point cloud path. Q is the target point. θ0 is the heading angle of the agricultural vehicle. θ is the desired heading angle of the agricultural vehicle. ∆θ is the heading error defined as the angle between vector C Q and the heading of the vehicle. ey is the lateral error defined as the perpendicular distance from C to vector ω c ω c + 1 .
Sensors 25 01110 g005
Figure 6. The GUI.
Figure 6. The GUI.
Sensors 25 01110 g006
Figure 7. Coordinate transformation: Left-O-Up is the screen coordinate system. P1 and P2 are adjacent anchor points in the UTM coordinate system that form the local target path. Pm is the midpoint of the local target path. τ is the angle between the local target path and the UTM northing. P′1 and P′2 are the anchor points after rotation of P1 and P2, respectively. Pscr1 and Pscr2 are anchor points for converting the local target path to the screen. The screen resolution is Wscr × Hscr.
Figure 7. Coordinate transformation: Left-O-Up is the screen coordinate system. P1 and P2 are adjacent anchor points in the UTM coordinate system that form the local target path. Pm is the midpoint of the local target path. τ is the angle between the local target path and the UTM northing. P′1 and P′2 are the anchor points after rotation of P1 and P2, respectively. Pscr1 and Pscr2 are anchor points for converting the local target path to the screen. The screen resolution is Wscr × Hscr.
Sensors 25 01110 g007
Figure 8. Zigzag navigation tests with (a) the high-clearance sprayer in (b) different zigzag path curvatures for the same turn.
Figure 8. Zigzag navigation tests with (a) the high-clearance sprayer in (b) different zigzag path curvatures for the same turn.
Sensors 25 01110 g008
Figure 9. The actual trajectories of the high-clearance sprayer following five target zigzag paths.
Figure 9. The actual trajectories of the high-clearance sprayer following five target zigzag paths.
Sensors 25 01110 g009
Figure 10. Comparative analysis of RMS in response to changes in (a) curvature, (b) speed, and (c) spacing.
Figure 10. Comparative analysis of RMS in response to changes in (a) curvature, (b) speed, and (c) spacing.
Sensors 25 01110 g010
Figure 11. The test scene.
Figure 11. The test scene.
Sensors 25 01110 g011
Figure 12. Actual trajectories and enlarged images of the high-clearance sprayer following the entry and exit paths.
Figure 12. Actual trajectories and enlarged images of the high-clearance sprayer following the entry and exit paths.
Sensors 25 01110 g012
Table 1. Technical parameters of the high-clearance sprayer.
Table 1. Technical parameters of the high-clearance sprayer.
Technical ParameterValue
Motor power/kW20
Tank volume/L500
Wheelbase × Tread/m × m1.5 × 1.5
Sprinkling width/m12
Traving speed/km/h0–10
Ground clearance/m1.2
Minimum turn radius/m3.5
Table 2. Path tracking errors for different navigation parameters.
Table 2. Path tracking errors for different navigation parameters.
NumberCurvature (m−1)Speed (km/h)Spacing (m)Maximum (cm)Average (cm)RMS (cm)
10.09313.71.421.37
20.0941.53.31.321.29
30.095.523.11.462.67
40.0972.54.32.022.39
50.09834.82.242.52
60.11414.21.371.21
70.11813.21.331.86
80.1131.53.61.431.55
90.11724.41.762.11
100.115.52.53.51.641.65
110.11434.41.842.22
120.125312.10.840.85
130.12571.53.91.531.18
140.12581.52.81.231.82
150.1255.522.20.941.45
160.12542.53.91.571.78
170.1255.533.31.842.04
180.17714.32.021.79
190.175.51.53.71.671.98
200.17425.21.691.87
210.17823.51.430.99
220.1732.53.71.762.39
230.177352.242.36
240.25314.11.732.25
250.255.51.56.82.242.11
260.25724.21.722.72
270.2582.56.32.492.33
280.25436.52.652.72
Table 3. ANOVA results for regression model accuracy.
Table 3. ANOVA results for regression model accuracy.
SourceSum of SquaresdfMean SquareF-Valuep-Value
Model4.4041.108.060.0003significant
v-Speed0.802210.80225.880.0236\
d-Spacing1.6811.6812.310.0019\
φv0.728710.72875.340.0301\
φ21.3811.3810.100.0042\
Table 4. Path tracking lateral errors.
Table 4. Path tracking lateral errors.
PathMaximum (cm)Average (cm)RMS (cm)
Turn at the starting point3.301.862.14
Turn at the end point2.401.751.98
Entry path3.302.042.27
Exit path2.401.822.06
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Yang, S.; Zhang, E.; Liu, Y.; Du, J.; Yin, X. ZPTM: Zigzag Path Tracking Method for Agricultural Vehicles Using Point Cloud Representation. Sensors 2025, 25, 1110. https://doi.org/10.3390/s25041110

AMA Style

Yang S, Zhang E, Liu Y, Du J, Yin X. ZPTM: Zigzag Path Tracking Method for Agricultural Vehicles Using Point Cloud Representation. Sensors. 2025; 25(4):1110. https://doi.org/10.3390/s25041110

Chicago/Turabian Style

Yang, Shuang, Engen Zhang, Yufei Liu, Juan Du, and Xiang Yin. 2025. "ZPTM: Zigzag Path Tracking Method for Agricultural Vehicles Using Point Cloud Representation" Sensors 25, no. 4: 1110. https://doi.org/10.3390/s25041110

APA Style

Yang, S., Zhang, E., Liu, Y., Du, J., & Yin, X. (2025). ZPTM: Zigzag Path Tracking Method for Agricultural Vehicles Using Point Cloud Representation. Sensors, 25(4), 1110. https://doi.org/10.3390/s25041110

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop