Unmanned ploughing machine obstacle avoidance path planning method and related equipment

Abstract

The invention discloses an unmanned ploughing machine obstacle avoidance path planning method and related equipment. The method comprises the following steps: detecting an obstacle having an overlapping relationship with a current operation channel in the current operation channel, and calculating a width ratio of the obstacle in the current operation channel; determining an obstacle avoidance path planning mode of the unmanned ploughing machine based on the width ratio, wherein the obstacle avoidance path planning mode comprises a complete supplementary farming mode and a simplified bypassing mode; and under the condition that the obstacle avoidance path planning mode is determined to be a complete supplementary tillage mode, the unmanned ploughing machine is controlled to execute short-distance reciprocating tillage for a preset number of times behind and / or in front of the target obstacle so as to perform supplementary tillage on the non-tillage area around the obstacle. The problem that the current obstacle avoidance path planning algorithm applied to the ploughing operation of the agricultural unmanned aerial vehicle mostly directly refers to the obstacle avoidance strategy of a general mobile platform such as a self-service vehicle and the like and cannot meet the agricultural requirement can be solved.

Description

Technical Field

[0001] This application relates to the field of computers, and more specifically, to a method and related equipment for obstacle avoidance path planning for an unmanned ploughing machine. Background Technology

[0002] Currently, obstacle avoidance path planning algorithms applied to agricultural drones for plowing operations mostly directly borrow from obstacle avoidance strategies of general mobile platforms such as passenger cars. That is, based on a preset global path and the perceived outline of the obstacle, a one-time obstacle avoidance operation is performed with the goal of safe passage. However, this pass-through obstacle avoidance mode exposes significant defects in agricultural farming scenarios. This is because the primary goal of agricultural machinery operations is to ensure the complete plowing of the land rather than simply passing through. The plows carried by drones have a fixed width, and coupled with the physical limitation of their minimum turning radius, a large number of uncultivated triangular or arc-shaped blank areas are generated in front of and behind obstacles due to the obstacle avoidance operation. This seriously reduces the utilization rate of farmland and cannot meet farmers' strict requirements for cultivated coverage. Summary of the Invention

[0003] The summary section introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. The summary section of this invention is not intended to limit the key features and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0004] To address the issue that current obstacle avoidance path planning algorithms used in agricultural drones for plowing operations often directly borrow obstacle avoidance strategies from general-purpose mobile platforms such as passenger cars, which fail to meet agricultural needs, this invention proposes, firstly, an obstacle avoidance path planning method for unmanned plowing drones. This method includes: Detect obstacles that overlap with the current work channel within the current work channel, and calculate the width ratio of the obstacles in the current work channel; The obstacle avoidance path planning mode of the unmanned ploughing machine is determined based on the width ratio. The obstacle avoidance path planning mode includes a complete supplementary tillage mode and a simplified detour mode. When the obstacle avoidance path planning mode is determined to be a complete supplementary tillage mode, the unmanned ploughing machine is controlled to perform a preset number of short-distance reciprocating tillage operations behind and / or in front of the target obstacle to supplement the uncultivated area around the obstacle. Then, while keeping the plow in the lowered state, the plow bypasses the target obstacle and returns to the global operation path based on the detour return path.

[0005] Optional, also includes: The current position of the unmanned ploughing machine and the global operation path within a preset distance range ahead are obtained. Based on the global operation path, a current operation channel is generated. The current operation channel takes the center line of the global operation path within the preset distance range as a reference and expands to the left and right with a preset width to form a rectangular decision area. The preset width is half the width of the plow. The calculation of the width percentage of the obstacle in the current working channel includes: Within the current work channel, a set of obstacles intersecting with the current work channel is detected. For each obstacle, its projected width in the width direction of the current work channel is calculated. The projection ratio is determined based on the ratio of the projected width to the width of the current work channel to determine the width ratio.

[0006] Optionally, the obstacle avoidance path planning mode of the unmanned tiller is determined based on the width ratio. The obstacle avoidance path planning mode includes a complete supplementary tillage mode and a simplified detour mode, including: When the target obstacle is in an unprocessed state or the projection ratio is greater than a preset threshold, it is determined that the complete supplementary tillage mode is adopted. When the target obstacle is in a state of being supplemented with farmland and the projection ratio is less than or equal to the preset threshold, the simplified detour mode is determined to be adopted.

[0007] Optionally, controlling the unmanned tiller to perform a preset number of short-distance reciprocating tillage operations behind and / or in front of the target obstacle to replenish uncultivated areas around the obstacle, and then, while keeping the plow in the lowered state, to bypass the target obstacle and return to the global operation path based on a detour return path, includes: Control the unmanned tractor to reverse to the preset re-cultivation starting point behind the target obstacle, and perform 2 to 3 reciprocating tillage operations between the re-cultivation starting point and the preset re-cultivation ending point to cover the blank area behind the target obstacle; Control the unmanned tractor to move to the preset re-cultivation area in front of the target obstacle, and perform 2 to 3 reciprocating tillage operations to cover the blank area in front of the target obstacle; Then, while keeping the plow in the lowered state, a passable grid is constructed based on the bird's-eye view grid map, and the A* search algorithm is used to plan a detour and regression path that satisfies the minimum turning radius constraint and the safety distance constraint, so as to bypass the target obstacle with the plow and return to the global operation path.

[0008] Optional, also includes: When the obstacle avoidance path planning mode is determined to be the simplified detour mode, the plow is kept in the lowered state, and the shortest detour regression path is planned based on the bird's-eye view grid map and the A* search algorithm to bypass the target obstacle and return to the global operation path. The processing status of the target obstacle is updated to "replenished with cultivation" and the corresponding operation data is recorded.

[0009] Optional, also includes: Based on the geometric boundaries and safe distance of the target obstacle, a safe avoidance side is formed, prioritizing the detour side that is farthest from the main body of the target obstacle; and / or, Under the premise of meeting safety avoidance requirements, the curvature or tangential deviation of the regression curves on different detour sides is calculated to evaluate the regression smoothness, and the detour side with better regression smoothness is selected; and / or, By combining the area or coverage benefit of the adjacent uncultivated area on the bypass side, the bypass side with higher cultivation efficiency is selected as the bypass direction.

[0010] Secondly, the present invention also proposes an obstacle avoidance path planning device for an unmanned ploughing machine, comprising: The identification unit is used to detect obstacles that overlap with the current working channel within the current working channel, and to calculate the width ratio of the obstacles in the current working channel; The determining unit is used to determine the obstacle avoidance path planning mode of the unmanned tiller based on the width ratio. The obstacle avoidance path planning mode includes a complete supplementary tillage mode and a simplified detour mode. The planning unit is used to control the unmanned ploughing machine to perform a preset number of short-distance reciprocating tillage operations behind and / or in front of the target obstacle to supplement the uncultivated area around the obstacle when the obstacle avoidance path planning mode is determined to be a complete supplementary tillage mode. Then, while keeping the plow in the lowered state, the ploughing machine bypasses the target obstacle and returns to the global operation path based on the detour return path.

[0011] Thirdly, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program stored in the memory to implement the steps of the unmanned ploughing machine obstacle avoidance path planning method as described in any of the first aspects above.

[0012] Fourthly, the present invention also proposes a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the obstacle avoidance path planning method for the unmanned ploughing machine according to any of the above claims in the first aspect.

[0013] In summary, the obstacle avoidance path planning method for the unmanned ploughing machine proposed in this application detects obstacles that overlap with the current working channel and calculates the width ratio of the obstacle in the current working channel; based on the width ratio, it determines the obstacle avoidance path planning mode of the unmanned ploughing machine, which includes a complete supplementary tillage mode and a simplified detour mode; when the obstacle avoidance path planning mode is determined to be the complete supplementary tillage mode, it controls the unmanned ploughing machine to perform a preset number of short-distance reciprocating tillage operations behind and / or in front of the target obstacle to supplement the uncultivated area around the obstacle; then, while keeping the plow in the lowered state, it detours around the target obstacle with the plow based on the detour return path and returns to the global working path. Therefore, by limiting the focus to the current work channel, obstacles that could cause missed tillage are consistently identified. The impact of obstacles is mapped to mode selection through width proportion, prioritizing high-impact obstacles for re-tillage. Short-distance, back-and-forth re-tillage directly covers critical untended areas that traditional detours struggle to reach. During the detour return process, the plow remains in place, ensuring the transition zone also participates in tillage, further reducing new untilled areas. The increase in untilled area due to obstacle appearance is significantly suppressed, especially when multiple fixed obstacles exist in the plot. For larger obstacles, while re-tillage increases short-term mobility, it reduces subsequent remedial work and multiple back-tilling operations, resulting in better overall efficiency. For smaller obstacles, a simplified detour mode can be selected in the implementation example to avoid excessive re-tillage causing time loss and repeated soil disturbance, achieving a more reasonable trade-off between time and coverage for different obstacle sizes. The early warning preparation action ensures that the tool's state is controllable when approaching obstacles, reducing the probability of collision between the plow and the obstacle. The detour return path explicitly considers safety distance and kinematic constraints, making the path easier to execute and reducing the risks caused by repeated probing near obstacles. Attached Figure Description

[0014] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A schematic flowchart of an obstacle avoidance path planning method for an unmanned ploughing machine is provided in an embodiment of this application; Figure 2 A schematic diagram of an obstacle avoidance path planning device for an unmanned ploughing machine provided in this application embodiment; Figure 3 This is a schematic diagram of an electronic device for obstacle avoidance path planning of an unmanned ploughing machine, provided in an embodiment of this application. Detailed Implementation

[0015] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The technical solutions of the embodiments of this application will now be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0016] To address the issue that current obstacle avoidance path planning algorithms used in agricultural drone plowing operations often directly borrow obstacle avoidance strategies from general-purpose mobile platforms such as passenger cars, which cannot meet the needs of agricultural applications, please refer to [the relevant documentation / reference]. Figure 1 This is a flowchart illustrating an obstacle avoidance path planning method for an unmanned ploughing machine provided in an embodiment of this application, which may specifically include steps S110 to S130.

[0017] S110, detect obstacles that overlap with the current working channel within the current working channel, and calculate the width ratio of the obstacles in the current working channel.

[0018] S120, Based on the width ratio, determine the obstacle avoidance path planning mode of the unmanned ploughing machine. The obstacle avoidance path planning mode includes a complete supplementary tillage mode and a simplified detour mode.

[0019] S130, when the obstacle avoidance path planning mode is determined to be a complete supplementary tillage mode, the unmanned ploughing machine is controlled to perform a preset number of short-distance reciprocating tillage operations behind and / or in front of the target obstacle to supplement the uncultivated area around the obstacle. Then, while keeping the plow in the lowered state, the plow bypasses the target obstacle and returns to the global operation path based on the detour return path.

[0020] For example, when the unmanned ploughing machine is performing global operation path tracking and tillage, the controller uses the current position as a reference to obtain a local segment of the global operation path within a preset distance range ahead, and constructs the current operation channel in the vehicle coordinate system or map coordinate system. This channel is used to represent the area that the plow will cover in the short term. Its lateral boundary can be determined according to the effective width of the plow, and its longitudinal length can be determined according to the preset decision distance, thus forming a strip-shaped area or rectangular area for local decision-making. Subsequently, the unmanned ploughing machine obtains obstacle information through the environmental perception module. The obstacle information can come from the results of lidar point cloud clustering, the 3D reconstruction results of depth cameras or binocular cameras, the ground feature boundaries obtained by visual semantic segmentation, and the prior annotation of farmland features, etc., and converts each obstacle into a geometric occupancy area representation, for example, describing its occupancy boundary on the ground by a polygonal envelope or circumscribed rectangle. The safety distance can be incorporated into the obstacle-occupied area to form a safety expansion zone, avoiding the risk of edge collision in subsequent planning. After completing the geometric representation of the obstacle, the controller performs a spatial intersection judgment on the area occupied by each obstacle and the current working channel area. If there is a non-zero intersection area or the intersection length reaches a set threshold, it is determined that the obstacle and the current working channel have an overlapping relationship and enter the subsequent calculation. For the obstacle entering the calculation, the controller calculates the projection width of the obstacle-occupied area in the lateral direction of the channel. The projection width can be obtained by calculating the length of the projection interval of the obstacle-occupied area on the lateral axis of the channel, and the ratio of the projection width to the channel width is used as the width ratio. In order to improve the robustness to irregular obstacles or obliquely distributed obstacles, multiple cross-sections can be sampled in the channel's forward direction to calculate the lateral projection and take the maximum value or weighted average value as the projection width. By filtering out obstacles that directly affect cultivated cover through overlapping channels, and then quantifying their encroachment on channel cover by width ratio, the system transforms the previously difficult-to-measure risk of missed cultivation into a calculable decision variable. This effectively avoids the accidental triggering of processing for irrelevant obstacles outside the channel, while providing a stable quantitative basis for subsequent mode selection. In multi-obstacle scenarios, the system can prioritize processing targets based on their proportion or distance, reducing operational jitter caused by frequent strategy switching. For example, when there is a tree stump in the center of the plot and its safety expansion zone significantly encroaches on the channel, its width ratio is large, and the system can identify that it will lead to a large risk of missed cultivation. However, when there are only small stones at the edge of the channel and their projection width is very small, the system can determine that their impact on cover is limited, creating conditions for selecting a lighter processing method in the future.

[0021] For example, after obtaining the width proportion of the target obstacle, the controller determines the obstacle avoidance path planning mode of the unmanned tillage machine according to a pre-set mode determination rule. The obstacle avoidance path planning mode includes at least a complete supplementary tillage mode and a simplified detour mode. The determination rule can be implemented by threshold determination or segmented determination. For example, in the threshold determination embodiment, the controller compares the width proportion with a preset proportion threshold. When the width proportion is greater than the threshold, it is determined that the obstacle significantly encroaches on the passage coverage. If only detour is used, a large area of ​​blank space in front of and behind the obstacle will be generated. Therefore, the complete supplementary tillage mode is selected to supplement the tillage. The task is incorporated into the obstacle avoidance sequence. When the width proportion is less than or equal to the threshold, the obstacle is considered to have a weak encroachment on the passage coverage. A simplified detour mode is selected to reduce the time loss caused by reversing, back-and-forth and additional maneuvers. In the segmented judgment embodiment, the width proportion can be divided into a high proportion interval and a low proportion interval. A complete supplementary cultivation mode is triggered for the high proportion interval, and a simplified detour mode is triggered for the low proportion interval. If necessary, an intermediate interval can also be set and additional criteria can be introduced. For example, a better mode can be selected based on the comparison between the expected missed cultivation area and the expected supplementary cultivation time to avoid frequent mode jitter when the proportion is close to the threshold. The obstacle avoidance mechanism transforms the goal from simply ensuring passage to improving cultivated coverage while maintaining passage. By using width ratio as a quantitative indicator of obstacle impact intensity, supplementary tillage actions are assigned only to high-impact obstacles. This creates a controllable engineering trade-off between tillage quality and operational efficiency. When the obstacle impact is significant, it can significantly reduce the area of ​​missed tillage before and after the obstacle, and when the obstacle impact is minor, it avoids unnecessary complex actions that lead to repeated tillage and increased operation time. For example, fixed irrigation facilities in the center of the plot usually have a large projection ratio. Triggering a complete supplementary tillage can reduce the risk of missed tillage around the facility in one go, and subsequent passages passing through this area do not require repeated remediation. On the other hand, small objects that occasionally fall at the edge of the passage have a small projection ratio, and simplified detours can be used to quickly pass through and return to the stable main passage for tillage as soon as possible, resulting in higher overall efficiency.

[0022] For example, when the controller determines that the obstacle avoidance path planning mode is the complete supplementary tillage mode, it first calculates a safe distance threshold based on the occupied boundary and safe expansion zone of the target obstacle. When the unmanned tiller reaches a preset position outside this safe distance threshold, it triggers a warning preparation action. This warning preparation action may include controlling the vehicle to decelerate until it comes to a complete stop, controlling the tiller to lift so that the tiller tip leaves the ground at a safe height, and recording the start marker of the obstacle handling event within the controller to ensure that the tiller does not scrape against the obstacle or dig into the ground due to abnormal force during subsequent reversing and maneuvering, thus preventing mechanical damage. After completing the warning preparation, the controller determines the area behind the obstacle... The spatial range of the supplementary tillage area and the supplementary tillage area in front of the obstacle is defined. The rear supplementary tillage area can be used to cover the tail gap caused by the return transition after detouring, and the front supplementary tillage area can be used to cover the head gap caused by the advance yaw before detouring. These two areas can be obtained by expanding the obstacle safety expansion zone on the rear and front sides in the channel direction, respectively. The expansion length can be determined by combining the minimum turning radius and the expected return curve length, so that the supplementary tillage area covers the key strip area most prone to missed tillage. Subsequently, in the embodiment of supplementary tillage behind the obstacle, the unmanned ploughing machine reverses to the rear supplementary tillage starting point while keeping the plow raised. After reaching the starting point, it controls the plow to fall to the set tillage depth, along with the global... The tiller moves forward a short distance along a nearly parallel path to the end point of the subsequent tillage. At the end point, depending on the ground conditions and the machine's capabilities, it selects between lifting the plow or shallow lifting the plow and performs a reversing maneuver. It then returns to the vicinity of the starting point of the subsequent tillage and repeats the tillage for a preset number of times. A lateral offset can be set between adjacent tillage cycles to create multiple adjacent coverage trajectories, avoiding uneven tillage depth caused by repeated tillage on the same path. In the embodiment of tillage in front of obstacles, the unmanned tiller moves to the starting point of the preceding tillage area, similarly controlling the plow to drop and perform a preset number of short-distance reversing tillage cycles to fill in the untilled areas caused by premature turning before detouring. After completing the preceding and following steps... After re-cultivation, the controller generates a detour return path based on the current pose, obstacle occupancy boundaries and safety expansion zones, and the return target point on the global operation path. The detour return path must meet the vehicle's minimum turning radius constraint and the safety distance constraint to obstacles, and can further consider the geometric offset of the plow relative to the vehicle's reference point to avoid the risk of the plow sweeping and colliding with the vehicle while it passes through. When executing the detour return path, the unmanned tractor keeps the plow in the lowered state and moves along the detour return path, so that the detour and return process itself forms a continuous cultivation trajectory, thereby including the return transition zone in the cultivation coverage range, and finally returns to the global operation path and resumes regular channel cultivation.By employing early warning and preparation actions, the low-speed maneuvering and reversing processes near obstacles are ensured to be safe and controllable. Then, by using short-distance reciprocating tillage before and after the obstacle, key untapped areas that are difficult to cover with traditional single-pass detours are transformed into planable and executable tillage trajectories. Finally, an executable plowed detour return is achieved through a detour return path that satisfies kinematic and safety constraints, preventing the return transition area from becoming a new missed tillage zone. This significantly reduces the missed tillage area before and after the obstacle and in the return transition area, improves tillage coverage and tillage depth consistency around the obstacle, and reduces repeated trial runs due to path infeasibility or edge-to-edge risks. For example, when there is a tree stump in the center of the plot and the stump has a large diameter, simply going around it will create a fan-shaped or strip-shaped uncultivated area behind the stump. By plowing back and forth to fill in the uncultivated area, the uncultivated area can be covered first. Then, when returning with the plow, the area around the bend can be plowed as well, thus significantly reducing the missed cultivation around the stump. Another example is when the obstacle is a fixed irrigation well cover. The area in front of the well cover is often left blank due to early deviation. By plowing back and forth in front, this area can be filled in, so that subsequent passages do not need secondary remediation. In addition, the lifting and stopping of the plow during the early warning preparation can reduce the impact risk of the well cover edge protrusion on the plow.

[0023] In some examples, it also includes: The current position of the unmanned ploughing machine and the global operation path within a preset distance range ahead are obtained. Based on the global operation path, a current operation channel is generated. The current operation channel takes the center line of the global operation path within the preset distance range as a reference and expands to the left and right with a preset width to form a rectangular decision area. The preset width is half the width of the plow. The calculation of the width percentage of the obstacle in the current working channel includes: Within the current work channel, a set of obstacles intersecting with the current work channel is detected. For each obstacle, its projected width in the width direction of the current work channel is calculated. The projection ratio is determined based on the ratio of the projected width to the width of the current work channel to determine the width ratio.

[0024] For example, when the unmanned ploughing machine is performing a tillage task, the controller periodically obtains the current position and heading of the unmanned ploughing machine, and extracts a path segment within a preset distance range in front of the current position from the global operation path as a local reference. The preset distance range is used to cover areas that the unmanned ploughing machine may reach in a short time and require advance decision-making, thereby avoiding sudden turns, sudden stops, or missed tillage caused by triggering planning when approaching obstacles. After obtaining the local path segment, the controller uses the centerline of the local path segment as the center reference of the current operation channel, and expands it to a preset width in the left and right directions corresponding to the normal of the centerline to form a rectangular reference. The decision area, with a preset width equal to half the width of the plow, ensures that the rectangular decision area covers the effective coverage width of the plow in a single pass, allowing for successful tillage. This directly correlates whether an obstacle intrudes into the passage with whether the tillage coverage that should be completed by the current passage will be disrupted. In engineering implementation, the controller can represent the rectangular decision area as the coordinates of its four corner points or a polygonal region, and update this region by translating it as the unmanned tiller moves. Simultaneously, a small safety margin can be added to the preset width based on factors such as positioning errors, vehicle yaw, and sideslip caused by ground slopes to avoid missing detection of edge-intruding obstacles due to an excessively narrow passage. A local decision space is constructed using the short-term forward-looking segment of the global path, converging the obstacle avoidance planning input from arbitrary environmental obstacles to obstacles that would affect the current tillage coverage. Half the width of the plow is used as the lateral expansion, aligning the passage geometry with the tillage coverage geometry on scale, thus providing a consistent coordinate and width reference for subsequent obstacle projection ratio calculations. It enables forward perception and advance decision-making, reducing the risk of trajectory failure caused by emergency turning when approaching obstacles. At the same time, it gives clear physical meaning to subsequent proportion indicators based on channel width, thereby improving the stability of mode selection and re-cultivation planning. For example, when there are tree stumps or irrigation facilities in the middle of the plot, the rectangular decision area can be used to determine in advance whether the tree stumps have encroached on the current cultivation bandwidth, and then arrange re-cultivation and detour paths in advance, instead of waiting until the machinery approaches the obstacle before hurriedly lifting the plow to detour, resulting in a large area of ​​blank space.

[0025] For example, after the current working channel in the form of a rectangular decision area is generated, the controller performs obstacle detection and screening within the channel. Obstacle detection can be achieved through obstacle outlines obtained from LiDAR point cloud clustering, ground feature boundaries obtained from visual detection and segmentation, depth maps, or 3D occupancy areas reconstructed by structured light, etc. The detected obstacles are then uniformly converted into planar geometric occupancy area representations for spatial relationship calculation with the channel area. The controller first performs an intersection determination on all obstacle occupancy areas and the current working channel, including only obstacles that intersect with the channel area in the intersecting obstacle set. This set screening avoids interference from obstacles outside the channel on the farming decision, thus concentrating computational resources on objects that truly affect the current channel coverage. Subsequently, for each obstacle in the intersecting obstacle set, the controller performs an intersection determination on the obstacle within the channel width direction. A projection calculation caliber is established. Specifically, the direction of the channel centerline can be defined as the longitudinal axis, and its normal can be defined as the transverse axis. All boundary points or grid occupancy points of the obstacle-occupied area are projected onto the transverse axis to obtain the upper and lower boundaries of the transverse projection interval, thereby calculating the projected width of the obstacle in the channel width direction. To accommodate irregular obstacles, obliquely intruding obstacles, or slender obstacles, the controller can also select multiple cross-sectional positions in the longitudinal direction of the channel, sample the transverse occupancy width of the obstacle at each cross-section, and take the maximum value as the projected width to ensure that the projected width reflects the most unfavorable degree of encroachment and avoid underestimation caused by averaging. After obtaining the projected width, the controller uses the ratio of the projected width to the channel width as the projection proportion and outputs this projection proportion as the width proportion to the pattern determination module, thereby realizing the quantitative expression of the degree of obstacle encroachment. Channel intersection screening ensures that the calculation object is directly related to the tillage cover. Then, transverse projection is used to transform the obstacle's encroachment on the channel from a two-dimensional shape problem into a one-dimensional width index, so that the proportion index can directly correspond to the plow width and channel width and can be used for threshold-based decision-making. It can reliably distinguish between obstacles that significantly affect coverage and those that have a limited impact on coverage, thus supporting the subsequent selection of complete supplementary cultivation or simplified detour, and reducing misjudgments caused by the complexity of obstacle shapes. For example, for obstacles with many branches and irregular outlines around tree stumps, using the maximum projection width of multiple cross sections can capture their strong encroachment on the lateral width in a certain section of the channel, thus triggering the supplementary cultivation mode more reliably. For stones that only have a small local corner encroaching on the channel boundary, the projection width is smaller and the proportion is naturally lower, making the system tend to adopt a lighter detour strategy to improve overall operation efficiency.

[0026] In some examples, the obstacle avoidance path planning mode of the unmanned tillage machine is determined based on the width ratio. The obstacle avoidance path planning mode includes a complete supplementary tillage mode and a simplified detour mode, including: When the target obstacle is in an unprocessed state or the projection ratio is greater than a preset threshold, it is determined that the complete supplementary tillage mode is adopted. When the target obstacle is in a state of being supplemented with farmland and the projection ratio is less than or equal to the preset threshold, the simplified detour mode is determined to be adopted.

[0027] Understandably, after obtaining the projection percentage of the target obstacle, the controller, in order to implement the operational strategy of prioritizing the re-cultivation of high-impact obstacles and allowing low-impact obstacles to pass quickly, incorporates the processing status of the target obstacle into the mode determination logic. The processing status includes at least two states: unprocessed and re-cultivated. The controller maintains an obstacle status record table. This obstacle status record table can be indexed according to the spatial location and geometric characteristics of the obstacle. For example, the obstacle's center point coordinates in the map coordinate system, the size of its circumscribed rectangle, its category label, and a timestamp can be used to form an obstacle identifier. This identifier is recorded each time the unmanned tillage machine enters the field. When entering a new current work channel, newly detected obstacles are matched with existing obstacles in the status log table. Matching can be based on location neighborhood thresholds and contour similarity thresholds. If the match is successful, its historical processing status is read; if the match fails, it is written as a new obstacle in the log table and assigned an unprocessed status. Subsequently, the controller performs mode determination. When the processing status of the target obstacle is unprocessed, the controller directly determines to adopt the full supplementary tillage mode. This ensures that when the obstacle is first encountered, the risk of leaving gaps before and after it is systematically tilled, thereby avoiding the continuous presence of obstacles when the same area is repeatedly passed through. If a target obstacle is found to have been partially tilled, the controller further compares its projected proportion with a preset threshold. If the projected proportion is still greater than the preset threshold, it indicates that the obstacle still significantly encroaches on the current channel coverage, or that the current channel coverage differs from the previous supplementary tillage coverage due to factors such as positioning errors, tillage zone switching, or channel offset, potentially creating new partially tilled areas. Therefore, the controller still determines to use the full supplementary tillage mode to perform more thorough supplementary tillage again. Conversely, if the target obstacle's processing status is that it has been supplemented and its projected proportion is less than or equal to the preset threshold, the controller determines... The system adopts a simplified detour mode, which allows the unmanned tractor to avoid repeated reversing and back-and-forth tillage after completing a sufficient re-tilling operation, thereby reducing time costs and minimizing excessive disturbance to the already tilled soil. In terms of threshold settings, the preset threshold can be set comprehensively based on factors such as the width of the tractor, the minimum turning radius of the unmanned tractor, the safe expansion distance of obstacles, and the upper limit of the positioning error. This allows the system to characterize whether simply detouring will result in a non-negligible blank area. Furthermore, the controller can write the threshold as a configurable parameter into the operation configuration file to optimize parameters for different plots, soil conditions, and different tractor combinations.The projected proportion is used as an instantaneous indicator of the intensity of obstacle encroachment on the current passage coverage, and the processing status is used as a historical memory indicator of whether the area around the obstacle has been systematically re-cultivated. By jointly determining the two, the re-cultivation action is accurately allocated to situations that truly require re-cultivation. This avoids the traditional obstacle avoidance strategy of repeatedly leaving the same obstacle blank for a long time or repeatedly performing heavy actions at low-impact obstacles. It can prioritize triggering complete re-cultivation when the obstacle is first encountered or when the obstacle encroachment is significant, thus significantly reducing the probability of missing cultivation before and after the obstacle. When the obstacle has been re-cultivated and the encroachment is small, it automatically switches to simplified detour to improve overall operation efficiency and reduce the tillage depth caused by repeated tillage. Unevenness and soil structure damage; for example, when a fixed tree stump in the center of the plot first appears, it is marked as untreated and triggers full re-cultivation. After the system completes the back-and-forth re-cultivation, its status is written as re-cultivated. When the subsequent passage passes through the area again and the tree stump only slightly intrudes into the edge of the passage, the projection ratio is lower than the threshold. The system can directly use simplified detour to quickly pass through and return to the main path. However, when the operation path is laterally offset due to boundary constraints, causing the tree stump to strongly intrude into the center of the passage again, the projection ratio exceeds the threshold again. The system will still trigger full re-cultivation to cover the newly appeared blank risk area, thereby maintaining a high cultivation coverage rate under different passage geometry conditions.

[0028] In some examples, the unmanned tiller is controlled to perform a preset number of short-distance reciprocating tillage operations behind and / or in front of a target obstacle to replenish uncultivated areas around the obstacle. Then, while keeping the tiller in a lowered state, it navigates around the target obstacle using a detour return path and returns to the global work path, including: Control the unmanned tractor to reverse to the preset re-cultivation starting point behind the target obstacle, and perform 2 to 3 reciprocating tillage operations between the re-cultivation starting point and the preset re-cultivation ending point to cover the blank area behind the target obstacle; Control the unmanned tractor to move to the preset re-cultivation area in front of the target obstacle, and perform 2 to 3 reciprocating tillage operations to cover the blank area in front of the target obstacle; Then, while keeping the plow in the lowered state, a passable grid is constructed based on the bird's-eye view grid map, and the A* search algorithm is used to plan a detour and regression path that satisfies the minimum turning radius constraint and the safety distance constraint, so as to bypass the target obstacle with the plow and return to the global operation path.

[0029] For example, when the controller determines that a target obstacle requires a complete supplementary tillage mode, to prioritize eliminating the tail gap most easily formed during the detour return phase, the controller first determines the preset supplementary tillage start point and preset supplementary tillage end point behind the target obstacle. The supplementary tillage start point can be set behind the obstacle's safety expansion boundary with a reversing alignment margin, and the supplementary tillage end point can be set near the obstacle's rear side but without encroaching on the safety expansion boundary. The controller also determines the main coverage direction of the rear gap area by combining the plow width, the expected return transition bandwidth due to the minimum turning radius, and the boundary of the previous tillage trajectory. During execution, the controller controls the unmanned tiller to decelerate, stop, and raise the plow at a safe distance before approaching the obstacle, ensuring that the plow does not dig into the ground or collide with the obstacle due to reverse force during reversing. Then, under path tracking control, the machine reverses to the preset re-cultivation starting point. Upon reaching the starting point, the plow is lowered to the set cultivation depth to enter the cultivation state. It then moves forward along the preset re-cultivation direction to the preset re-cultivation endpoint to complete the first short-distance cultivation. After reaching the endpoint, the controller controls the unmanned tractor to select either lifting or shallow lifting the plow according to the ground conditions to reduce the lateral resistance of the plow during reversal. It then performs a reversal maneuver to return to the vicinity of the starting point and lower the plow again for cultivation. This short-distance back-and-forth movement from the starting point to the endpoint and from the endpoint to the starting point is repeated. The number of back-and-forth movements is limited to 2 or 3 times. Between two adjacent back-and-forth movements, the controller can control the unmanned tractor to make a small lateral offset. The offset amount can be set according to the width of the plow or a reasonable fraction thereof to form adjacent cover strips, avoiding repeated plowing on the same trajectory, which would result in excessive cultivation depth or excessive soil fragmentation. This method transforms the fan-shaped or strip-shaped blank areas behind obstacles caused by traditional detours into controllable short-distance coverage strips. By reversing and aligning the vehicle, the re-cultivation trajectory can cover the blank areas one by one, starting from the outer edge of the blank area. At the same time, by using a limited number of reciprocations, a balance is struck between sufficient coverage and time cost. This can significantly reduce the uncultivated areas behind obstacles caused by the outward expansion of the regression curve, premature plowing, or regression correction. Furthermore, because the number of reciprocations is controlled and lateral offset can be introduced, the re-cultivation is closer to coverage than repeated plowing, thereby improving the consistency of cultivation coverage and cultivation depth behind obstacles. For example, when the obstacle is a tree stump or a fixed facility located in the center of the passage, detours often leave an uncultivated strip outside the curve behind the obstacle. By performing two to three short reciprocations behind the obstacle and appropriately offsetting laterally, this uncultivated strip can be divided into coverable strips and the obvious blank areas can be basically eliminated.

[0030] For example, to eliminate headspace gaps caused by premature yaw, premature plowing, or safe avoidance before detouring, the controller further determines a preset re-cultivation area in front of the target obstacle after completing the rear re-cultivation. This re-cultivation area can be set in front of the obstacle's safety expansion boundary and extends along the current working channel direction or the tangential direction of the global working path to form an operable short-distance cultivation strip. The selection of the re-cultivation start and end points can be determined based on the obstacle's encroachment position in the channel, the trimming results of the channel boundary and the plot boundary, and the current posture of the unmanned ploughing machine, to ensure that the main headspace gaps are covered without encroaching on the safety expansion zone or crossing the boundary during reciprocating cultivation. During execution, the controller controls the unmanned ploughing machine. Move to the starting position of the forward re-cultivation area at an appropriate speed, adjust the plow to the set cultivation depth and enter the cultivation state. Complete a short distance of cultivation in the re-cultivation area along the preset direction to the end of the re-cultivation, then change direction and return to carry out the second cultivation. Repeat the repetition 2 or 3 times. Lateral fine-tuning can also be introduced between adjacent repetitions so that multiple repetitions form parallel covering strips instead of single-line repetition. In cases where the boundary space of some plots is limited, the controller can limit the forward re-cultivation area to the side of the passage away from the boundary to ensure that the re-cultivation action will not cause excessive turning or repeated fine-tuning due to insufficient space. The controllability of the implement can be maintained by reducing the speed and shortening the single repetition distance. Therefore, the abandoned farmland before the detour can be filled in with controlled strips. Especially for the triangular or trapezoidal gaps commonly seen in front of obstacles, key gaps can be covered with a limited number of short-distance back-and-forth movements without significantly increasing the operation time. This reduces the probability of missed farmland in front of obstacles due to safety avoidance, and avoids the need for repeated maneuvers when adjacent passages pass by again. Furthermore, since the number of back-and-forth movements and the amount of offset are controllable, excessive disturbance to the already cultivated area can be reduced, improving the uniformity of cultivation in front of obstacles. For example, when the obstacle is an irrigation well cover or a protruding rock, vehicles usually lift the plow and veer before reaching the obstacle, resulting in an uncultivated strip in front of the obstacle. Two to three short-distance back-and-forth movements in the area to be recultivated in front can cover this uncultivated strip, making the cultivated boundary near the detour starting point more regular, thereby improving land utilization.

[0031] For example, after completing the pre- and post-tillage replanting, the controller acquires or generates a bird's-eye view grid map with reference to the unmanned tillage machine to generate an executable and cover-friendly detour return path. This bird's-eye view grid map can be constructed by sensor fusion, for example, mapping obstacle-occupied areas detected by LiDAR or depth cameras as occupied grids, mapping arable ground as passable grids, and mapping obstacle safety expansion zones as impassable or high-cost grids. Simultaneously, plot boundaries are mapped as impassable grids to prevent boundary crossings. In constructing the bird's-eye view grid map... When traversing a grid, the controller can further consider the geometric outline of the plow relative to the vehicle body. By morphologically expanding the grid occupied by obstacles, the vehicle outline and the plow outline are included in the avoidance range, ensuring that the planned path not only guarantees the vehicle's passage but also prevents the plow from colliding with obstacles. Subsequently, the controller selects the path planning start point as the current pose of the unmanned tractor or the pose at the end of the re-plowing, and the planning endpoint as the regression target point located in front of the obstacle on the global operation path. The controller then uses the A search algorithm on the grid map to search for a feasible path from the start point to the endpoint. The path, where the cost design of A can include travel distance cost, penalty cost for approaching obstacles, and penalty cost for deviating from the global operation path, to guide the path to return to the main path safely and as quickly as possible. To meet the minimum turning radius constraint, the controller can introduce kinematic reachability constraints during the search process. For example, the search node can be expanded into a state node containing heading information and the heading change rate of adjacent nodes can be limited. Alternatively, curvature constraint smoothing can be performed on the discrete path obtained from A, so that the minimum curvature radius of the path curve is not less than the minimum turning radius of the unmanned tractor. At the same time, a minimum gap of not less than the safe distance can be maintained between the path and obstacles, thereby ensuring that the path is trackable and executable. During the path execution phase, the controller keeps the tractor in the lowered state, so that the unmanned tractor continues to till the soil while traveling along the planned detour and return path. The detour and return transition area, which was originally only used for maneuvering, is transformed into a tillage coverage area. The speed is reduced in sections with large curvature through speed adaptive control to stabilize the tillage depth and trajectory tracking accuracy. Finally, the system returns to the global operation path and resumes regular channel tillage.By using a bird's-eye view grid map to discretize the free space of the environment and the safe expansion zone of obstacles, A* can search for a connected detour path under clear drivability constraints. Then, through kinematic constraints and smoothing, geometric feasibility is improved to be executable by the vehicle and plow. By keeping the plow in place, the detour process can simultaneously complete the supplementary coverage. This can significantly reduce the probability of pauses, repeated attempts and return failures caused by unexecutable paths. At the same time, it reduces the newly added blank area outside the return curve and improves the farming continuity and coverage of the area around the obstacle while ensuring the safety distance. For example, when the obstacle is located in the center of the channel and the space on both sides is asymmetrical, A* can automatically select the safer and faster return side to generate a detour path under the guidance of grid cost. Curvature constraints avoid execution deviations caused by planning sharp curves, so that the unmanned plowing machine is more stable in maneuvering near obstacles, with less blank space and smoother return.

[0032] In some examples, it also includes: When the obstacle avoidance path planning mode is determined to be the simplified detour mode, the plow is kept in the lowered state, and the shortest detour regression path is planned based on the bird's-eye view grid map and the A* search algorithm to bypass the target obstacle and return to the global operation path. The processing status of the target obstacle is updated to "replenished with cultivation" and the corresponding operation data is recorded.

[0033] In some examples, it also includes: Based on the geometric boundaries and safe distance of the target obstacle, a safe avoidance side is formed, prioritizing the detour side that is farthest from the main body of the target obstacle; and / or, Under the premise of meeting safety avoidance requirements, the curvature or tangential deviation of the regression curves on different detour sides is calculated to evaluate the regression smoothness, and the detour side with better regression smoothness is selected; and / or, By combining the area or coverage benefit of the adjacent uncultivated area on the bypass side, the bypass side with higher cultivation efficiency is selected as the bypass direction.

[0034] For example, when the controller needs to generate a detour path for a target obstacle, to avoid safety risks, unsmooth return, or additional missed areas due to relying solely on experience to fixate on detours from the left or right, the controller processes the geometric boundary and safety distance of the target obstacle and forms a calculable detour direction selection rule. Specifically, the controller first expands the occupancy boundary of the target obstacle by a preset safety distance to obtain a safety expansion boundary, and then constructs candidate detour corridors or candidate detour sides on both sides of the obstacle based on this safety expansion boundary. The candidate detour sides can be defined in the channel coordinate system as the free space to the left and right of the obstacle's centerline, or in the grid map as a set of detour side grids; the determination of the safe avoidance side... In terms of obstacle avoidance, the controller calculates the minimum safety margin that can be provided on both sides while ensuring that the safety expansion boundary is not intruded, based on the convex direction of the target obstacle's main body, the relative positional relationship between the obstacle's main body and the channel centerline, and the minimum clearance from the safety expansion boundary to the left and right boundaries of the channel. It prioritizes the candidate bypass side with a larger safety margin and further away from the obstacle's main body as the safe avoidance side, thereby reducing the probability of edge-grabbing collisions caused by tracking errors, sideslip, and ground undulations. Regarding regression smoothness evaluation, when all candidate bypass sides meet the safety avoidance requirements, the controller generates a candidate regression curve or candidate regression path for each of the left and right bypass sides to return to the global operation path, and calculates the corresponding smoothness. Evaluation metrics, including smoothness metrics, can be characterized by the peak curvature, rate of curvature change, or deviation between the tangential direction at the end of the candidate path and the tangential direction of the global operation path. This allows for the selection of candidate bypass sides with smaller curvature, gentler curvature changes, and smaller tangential deviations, making the regression process more closely match the minimum turning radius and tracking capability of the unmanned tiller, reducing path tracking overshoot, redundant corrections, and new blank areas caused by overly steep regression curves. Regarding tillage efficiency evaluation, the controller further incorporates the spatial distribution of adjacent uncultivated areas on the bypass side to construct a cover benefit index. This cover benefit index can be determined by the number of uncultivated grids near the bypass path, the area of ​​uncultivated areas, or the predicted cover of uncultivated areas overlapping with the bypass path. The coverage area is calculated, and under the premise of satisfying safety avoidance and return to workability, the detour side with higher coverage benefits is selected first, so that the detour process is as close as possible to the uncultivated area and necessary maneuvers are converted into effective cultivated coverage as much as possible. In engineering implementation, the above rules can be implemented through weighted comprehensive scoring. When the safety margin is insufficient, the corresponding detour side is directly eliminated as a hard constraint. The return to workability and coverage benefits are compared among the remaining candidate sides to determine the final detour direction. The controller can fix the safety avoidance priority to the highest according to the operation configuration, and dynamically adjust the weights of return to workability and cultivation efficiency according to factors such as plot shape, soil conditions, and machinery traction capacity, so as to realize a configurable strategy from safety priority to efficiency priority.This approach elevates the detour direction selection from a simple geometric bypass problem to a decision-making problem under multi-objective constraints. First, a safety expansion boundary ensures the physical feasibility and safety of the detour side. Then, regression curvature or tangential deviation guarantees the feasibility and trajectory continuity of the regression process. Finally, the coverage benefits of uncultivated areas ensure that detour maneuvers positively contribute to cultivated cover. This significantly improves the quality and efficiency of operations near obstacles without increasing hardware costs. It reduces the risk of edge-grazing caused by improper detour selection, lowers the probability of tracking failures and repeated maneuvers due to excessively steep regression curves, and reduces new blank areas and repeated tillage caused by deviating from uncultivated areas. For example, when the right side of the obstacle is adjacent to the plot boundary and free space is narrow. If the gap between the safety expansion boundary and the right boundary of the channel is insufficient, the system will reject the right detour and select the left detour, thereby avoiding the plow outline from sweeping into the boundary or obstacle safety zone. When both sides are safe and feasible, if the curvature required for the left detour is smaller and the end tangent is easier to align with the global operation path, the system will prioritize the left detour to achieve a smoother return and reduce missed tillage in the return transition zone. For example, if there is a large uncultivated area near the detour path on one side and the other side has been basically tilled, the system will select the uncultivated area side for detour under the premise of safety and smoothness, so that the tillage trajectory of the detour process generates higher coverage benefits for the uncultivated area, thereby improving land utilization and reducing the number of subsequent re-tillage times.

[0035] In some cases, considering the common micro-topography around obstacles, such as minor ridges, ruts, shallow ditches, and slippery soil patches, lateral slippage and heading deviation can occur during short-distance back-and-forth tillage. Although the planned tillage trajectory geometrically covers the untilled strips, due to slippage, the actual plow tip trajectory does not coincide with the target blank area, creating a hidden omission that appears to be filled but is actually misaligned. Therefore, it is necessary to explicitly introduce the micro-topography slope field and rolling resistance field into the tillage trajectory generation and execution layers, and to implement heading and lateral micro-pulse correction based on real-time yaw angle and slippage estimation during execution, ensuring that the plow tip trajectory continuously conforms to the untilled low-confidence area. Based on this, some examples also include: Before performing short-distance reciprocating tillage behind and / or in front of the target obstacle, a micro-topographic slope field and rolling resistance field around the obstacle are constructed based on sensor data and historical operation data. During the trajectory generation and execution of short-distance reciprocating tillage, a sideslip compensation heading and lateral micro-pulse correction mechanism are introduced. The correction is based on real-time yaw angle, sideslip estimation and online update of the coverage confidence grid to maximize the overlap between the tilled strip and the target untilled low-confidence area and reduce strip drift.

[0036] For example, data such as lidar reflection intensity, stereo vision dense point cloud, historical wheel slip rate, and motor current are collected to reconstruct a digital surface model of the recultivated area; local slope and aspect are calculated, and a rolling resistance field is constructed by combining wheel slip rate and current fluctuations; weighted amplification is applied to areas with significant disturbances to obtain a micro-topographic cost map for trajectory generation. Historical cultivation trajectories are projected onto a bird's-eye view grid, and the cumulative number of cultivation passes and cultivation depth stability are used to form a coverage confidence level; uncultivated or low-confidence areas are the primary targets for recultivation. Two to three short-distance reciprocating candidate lines are generated on the target uncultivated areas. During online optimization, the slope field and rolling resistance field are used as cost terms to pre-compensate the offset of the candidate lines relative to the terrain normal in high sideslip areas; the solution with the maximum coverage of low-confidence areas and the minimum crossing of high-resistance areas is selected. The sideslip angle and yaw rate are estimated based on the vehicle's attitude and wheel speed difference. Small-frequency, micro-amplitude lateral displacement pulses are superimposed on the trajectory tracking output, with the pulse amplitude and period adjusted by a real-time sideslip estimation closed-loop. When a continuous deviation between the strip and the low-confidence grid is detected, the heading compensation coefficient is increased and the forward speed is appropriately reduced. The growth rate of the low-confidence covered area is statistically analyzed in real time. When the growth rate is below a threshold within two round trip cycles, the strip is deemed effectively covered, and the process moves to the next strip or ends the replanting. Thus, the slope and resistance costs during the trajectory generation stage ensure that the replanting direction inherently aligns with the executable direction; the sideslip compensation and micro-pulses during the execution period transform passive disturbances of error within the strip scale into active corrections. Repeated deep digging at the same location is avoided, reducing soil structure damage. Thresholded termination conditions reduce ineffective repetitions.

[0037] In some cases, objects such as straw piles, loose plastic film residue, and small mounds of loose soil may geometrically occupy the path, but mechanically they may be safely compacted or driven over. Judging obstacles solely based on geometric occupancy can easily trigger unnecessary detours and re-cultivation, reducing efficiency. An intrudability assessment layer is introduced to distinguish between geometric occupancy and mechanical intrudability. This is determined through a comprehensive assessment using low-speed contact probing, plow pressure response, and visual compliance estimation. The cost layer is dynamically adjusted downwards or upwards based on the results, and strategies for straight-line compaction or enhanced re-cultivation are switched. Based on this, some examples also include: An indentability assessment layer is set up during obstacle detection and mode selection: the indentability of obstacles or surface deposits is comprehensively determined by the normal acceleration response of low-speed contact probing, plow pressure sensing, and visual texture compliance estimation; for occupied areas determined to be indentable, their cost in the bird's-eye view grid is reduced and simplified detour or straight-through strategies are prioritized; for occupied areas determined to be non-indentable, their safe expansion distance is increased and corresponding local re-cultivation templates are triggered.

[0038] For example, occupied grids that are small in size, loosely textured, have low reflectivity, and are below the safety threshold are marked as compactable candidates. Approaching at low speed, the plow or pressure roller makes light contact with the target at a safe height; normal acceleration, plow force, and displacement response curves are collected to calculate a compactability score. Visual compliance indicators such as texture fineness and edge softness are weighted and fused with the mechanical compactability score to obtain a comprehensive compactability level. If the level reaches the compactability threshold, the grid cost for that area is reduced, allowing straight-through driving or classifying it as a simplified detour. If the level does not reach the threshold, the safe expansion radius is increased, triggering the corresponding re-cultivation template and plow-assisted detour regression. By analyzing post-pass sensor readings and cover confidence changes, it is assessed whether the compaction result causes new blanking or equipment malfunction; the results are fed back to adjust the fusion weights and thresholds. This frees a large number of geometrically occupied but compactable targets from complex paths and re-cultivation, shortening maneuver time. Post-event verification, combined with cover confidence write-back, ensures that straight-through driving does not create hidden missed cultivation. The pressure is only implemented in low-risk areas through contact and probing.

[0039] In some cases, considering that wind, dust, and vegetation movement can cause inter-frame jitter in perception, the obstacle occupancy probability fluctuates repeatedly within a short period. If this jitter is directly input into the planning, A* will be frequently recalculated, resulting in a jagged execution trajectory and discontinuous replanting strips. By using double-threshold hysteresis and critical stage raster snapshot freezing, the planning input is stabilized without sacrificing safety, and replanning is minimized only when a new occupancy conflicting with the current path occurs. Based on this, some examples also include: When constructing a bird's-eye view raster map for detour regression planning, a time-stabilized occupancy update strategy is adopted: a dual-threshold hysteresis is implemented on the occupancy probability to make occupancy expansion take effect first and occupancy reduction take effect only after a certain number of frames. In the critical stages of recultivation and detour regression, the raster snapshot used for planning is frozen for a short time. Local minimization replanning is only triggered when the newly added occupancy intersects with the current planned path, so as to reduce the frequent recalculation of the path and jagged tillage trajectory caused by perceived jitter.

[0040] For example, exponential smoothing is applied to each grid cell; two thresholds are configured: a lower threshold for entering an occupancy state and a higher threshold for exiting an occupancy state; occupancy expansion takes effect immediately, while shrinking requires confirmation over several frames. During back-end and front-end recultivation and rerouting path tracking, a short time window is frozen for the occupancy map snapshot used for planning; within the window, only whether a new occupancy intersects with the current path is monitored. Local window replanning is triggered only when a new occupancy intersects with the current path; otherwise, the path remains unchanged until the window ends or the task phase switches. After the window ends, the global occupancy map is updated with the stable occupancy accumulated within the window, ensuring that subsequent decisions inherit changes in the real environment. This avoids frequent path switching and jagged edges caused by short-term jitter. Occupancy expansion takes effect immediately, ensuring that new risks quickly intervene in decision-making; shrinking takes effect later, which can resist noise. Local minimization replanning reduces unnecessary global searches.

[0041] In some cases, considering the short-distance reciprocating tillage and plow return process near obstacles, ground undulations, lateral resistance, and traction fluctuations are transmitted to the plow attachment through the three-point suspension or trailer connection, causing the plow attachment to oscillate low-frequency around the connection point. This oscillation manifests as a phase lag and amplitude amplification of the plow tip relative to the vehicle's reference trajectory, resulting in a systematic deviation between the actual tillage trajectory and the target tillage strip. By establishing a coupled estimation model of plow attachment oscillation, vehicle attitude, and traction load, the oscillation phase and amplitude are estimated online. Based on this, a feedforward bias is applied to the reference trajectory, and dynamic filtering and amplitude limiting control are implemented on the execution heading. Simultaneously, when the oscillation is too large, the operating speed window is narrowed, allowing the actual plow tip trajectory to realign with the target tillage strip, improving coverage accuracy and tillage depth consistency. Based on this, some examples also include: during the trajectory generation and execution of re-tillage and detour regression, establishing a coupled estimation model of plow oscillation-vehicle attitude-traction load, estimating the oscillation phase and amplitude of the plow relative to the vehicle reference trajectory online; applying a feedforward bias to the target reference trajectory based on the estimation results and performing dynamic filtering and amplitude limiting control on the execution heading so that the plow tip actually aligns with the target re-tillage strip through the trajectory; and temporarily narrowing the operating speed window when the oscillation amplitude is detected to reduce the impact of oscillation on trajectory alignment.

[0042] For example, a small inertial measurement unit, an angular displacement sensor, and a tension / compression sensor at the traction pin are arranged on the plow frame or plow beam to simultaneously read the vehicle body IMU, steering angle, drive wheel speed, and engine or motor output torque. The equivalent moment of inertia of the plow, the equivalent stiffness of the suspension, and approximate values ​​of damping are obtained through offline calibration, forming a second-order small-perturbation model. In the online stage, an extended Kalman filter or sliding mode observer is used to jointly estimate the plow oscillation angle, oscillation angular velocity, and load perturbation terms. The model uses the low-frequency components of vehicle body pitch and yaw as external excitation inputs and the traction force change as a perturbation input, achieving a considerable understanding of the plow's oscillation state relative to the vehicle body reference. The target reference trajectory is represented at the equivalent point of the plow tip rather than the geometric center of the vehicle body, and a feedforward bias is calculated based on the estimated sway phase and amplitude. When a leftward sway trend is estimated, a small lateral and directional bias opposite to the sway direction is applied to the reference trajectory within a short future time window. The bias amplitude and phase are obtained by mapping the observer output through the feedforward channel and are limited within a safety boundary to ensure that no new edge-grabbing risks are introduced. For regression segments with large curvature, static compensation based on the plow tip-vehicle body geometric bias is also superimposed to make the plow tip path on the curve segment closer to the planned centerline. During the low-speed fine operation phase, the controller introduces a bandpass dynamic filter for the steering command: high-frequency small disturbances are not excessively amplified, and low-frequency sway components are canceled out through the feedforward channel; at the same time, upper and lower limits are set for the directional change rate to ensure that the time derivative of the steering command does not exceed the set threshold, avoiding resonance between the vehicle body attitude and the plow sway. If necessary, slope limits are set for the drive wheel differential and hydraulic lifting response to reduce traction fluctuations and facilitate sway attenuation. When the estimated sway amplitude exceeds the threshold or the sway energy does not decay within several sampling windows, a speed management strategy is triggered: the operating speed window is narrowed, reducing the longitudinal speed to a range favorable for sway damping; within this range, the relative motion between the plow and the vehicle body tends to be steady-state, which helps to ensure accurate alignment of feedforward compensation. The speed threshold and recovery conditions are calibrated experimentally to avoid frequent switching.

[0043] In some cases, considering the spatial gradient of soil hardness and stone content often present near obstacles, the regression curve with plow may exhibit traction overload, increased wheel slip, and insufficient tillage depth on the high-hardness side, forming asymmetrical shallow tillage or untilled strips; while on the low-hardness side, excessive tillage depth and increased energy consumption may occur. By constructing soil hardness and traction load layers on the regression curve, dividing the regression curve into several control zones, and implementing short-segment lifting plowing or tillage depth reduction and torque limiting strategies in the overload zone, while maintaining the target tillage depth in the carry-bearing zone and generating compensating tillage strips along the carry-bearing side after regression, overall optimization of cover and tillage depth consistency can be achieved. Based on this, some examples also include: constructing a soil hardness layer and a traction load layer on the regression curve segment of the regression path, and dividing the regression curve segment into several control zones: for overloaded zones where the hardness or traction load exceeds the zone threshold, implementing a short-segment plowing or reduced tillage depth strategy and limiting the upper limit of traction torque; for susceptible zones, maintaining the set tillage depth and generating a compensating tillage strip on one side of the susceptible zone after the regression is completed, to compensate for the shallow tillage or untilled areas caused by the overloaded zones, thereby ensuring the consistency of tillage coverage and tillage depth of the entire regression segment.

[0044] For example, by combining the penetration resistance estimate inferred from traction force, plow force, wheel slip ratio, and vibration indicators, historical land hardness maps, and surface stone content information, a hardness layer representing the neighborhood of the regression curve is generated in the bird's-eye view coordinate system; simultaneously, a load layer representing the regression curve is estimated from real-time traction torque, drive current, wheel slip ratio, and engine load. Both layers are synchronized to the planning controller after unified normalization. The regression curve is divided into several control zones of equidistant or adaptive lengths according to the path arc length. The mean hardness and mean load are statistically analyzed for each zone. If either indicator exceeds a preset threshold, it is marked as an overload zone; otherwise, it is marked as a bearable zone. Adjacent overload zones can be merged into overload segments to reduce frequent actions. In overload zones, the controller executes short-segment plowing or depth reduction according to the principle of safety first, then coverage, reducing traction peaks through torque limiting and speed reduction. When plowing inevitably leads to uncultivated strips, the spatial location, width, and length estimates of these strips are recorded and written into a compensation queue for compensation operations after regression. The lifting height and duration are triggered by a combined threshold of torque, wheel slip, and vibration to prevent excessive lifting from causing increased cover loss. Maintaining the target tillage depth and normal speed in the load-bearing zone, and after completing the regression and entering the straight section, one or more short-distance compensation strips are generated on the load-bearing side according to the compensation queue, prioritizing coverage of shallowly tilled or untilled areas caused by overloaded zones. The direction of the compensation strips is consistent with the main channel or the tangential direction at the end of the regression. The strip spacing is set according to the plow width, and the number of strips is determined by the area to be compensated and the allowable time overhead.

[0045] In some cases, plastic residue, damp surfaces, and strong light angles can cause abnormal lidar echo intensity and sparse point clouds, creating false airspace in the bird's-eye view occupancy map. If path generation is still primarily radar-based, A* might be misled by voids and plan to traverse inaccessible areas, leading to collision risks and missed cultivated zones. By constructing a high-reflectivity area mask using polarization imaging and combining it with lidar intensity confidence gating, this area is elevated to a high-risk grid in the cost map and subjected to safety distance inflation. When false airspace signs are detected, such as the co-occurrence of sparse point clouds and abnormal reflectivity, a vision-driven boundary tracking regression mode is switched to generate regression paths along visible ground boundary lines or cultivated lines, thus avoiding the misleading influence of false airspace. Based on this, some examples also include: introducing high-reflectivity area masks and sensor intensity confidence gating into the bird's-eye view raster map used for planning: identifying high-reflectivity areas such as plastic film residue based on polarization features obtained from polarization imaging and LiDAR echo intensity confidence, and elevating the area to a high-risk raster in the cost map and adding a safety distance inflation; when the high-reflectivity area causes the point cloud to be sparse and false passable airspace appears, switching to a regression path generation mode based on visual boundary tracking to avoid entering false airspace and reduce the risk of collision and missed cultivation in the detour regression stage.

[0046] For example, a polarizer is added in front of a visible light camera or a polarization camera is used to calculate the degree of polarization and polarization angle map in real time. Combined with lidar echo intensity and echo confidence, highly reflective materials and specular reflection areas are identified, and a binary mask and confidence map are generated. The masked area is marked as high-risk in the cost map, and a safety distance expansion is applied to its periphery, with the expansion radius determined based on the positioning error and the plowshare outline. On the planning input side, the co-occurrence of two types of features—sudden decrease in local point cloud density and echo intensity saturation or abnormal consistency—is monitored. When the co-occurrence persists for more than several frames and the overlap ratio with the polarization mask area exceeds a threshold, a false spatial domain is determined, triggering a switch from radar-dominated search to visual boundary tracking mode, while preserving the hard obstacle constraints occupied by the radar. In switching mode, semantic segmentation or edge detection is used to obtain the boundaries of arable land and the outer edges of obstacles. The tangent of the regression target is extracted by combining the texture of cultivated lines. A safe passage is generated along the boundary or lines. Within this passage, splines or polylines with limited curvature are spliced ​​to form the regression trajectory. The tangent at the end of the trajectory is aligned with the tangent of the global operation path to facilitate stable continuation after regression. Polarization masks and radar hard barriers are still used as impassable constraints during the planning process. The passage generated by visual tracking is converted into a low-cost band and written into the cost map, and a safety expansion related to the mask is maintained on both sides. Image quality monitoring is added to the execution layer. If the visual quality is lower than the threshold, it is downgraded to low-speed conservative straight-line movement and manual review or delay and retry is requested. If the visual quality recovers, it automatically returns to the vision-dominated mode until it passes through the high-reflectivity area.

[0047] Please see Figure 2One embodiment of the obstacle avoidance path planning device for the unmanned ploughing machine in this application may include: The identification unit 21 is used to detect obstacles that overlap with the current working channel within the current working channel, and to calculate the width ratio of the obstacles in the current working channel; The determining unit 22 is used to determine the obstacle avoidance path planning mode of the unmanned tiller based on the width ratio. The obstacle avoidance path planning mode includes a complete supplementary tillage mode and a simplified detour mode. Planning unit 23 is used to control the unmanned ploughing machine to perform short-distance reciprocating plowing a preset number of times behind and / or in front of the target obstacle to supplement the uncultivated area around the obstacle when the obstacle avoidance path planning mode is determined to be a complete supplementary plowing mode. Then, while keeping the plow in the lowered state, it bypasses the target obstacle with the plow based on the detour return path and returns to the global operation path.

[0048] like Figure 3 As shown, this application embodiment also provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, it implements the steps of any of the above-described methods for obstacle avoidance path planning of the unmanned ploughing machine.

[0049] Since the electronic device described in this embodiment is the device used to implement the obstacle avoidance path planning device for an unmanned ploughing machine in the embodiments of this application, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the method described in the embodiments of this application. Therefore, how the electronic device implements the method in the embodiments of this application will not be described in detail here. Any device used by those skilled in the art to implement the method in the embodiments of this application falls within the scope of protection of this application.

[0050] In practical implementation, when the computer program 311 is executed by the processor, it can achieve the following: Figure 1 Any of the corresponding implementation methods in the embodiments.

[0051] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0052] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for obstacle avoidance path planning for an unmanned ploughing machine, characterized in that, include: Detect obstacles that overlap with the current work channel within the current work channel, and calculate the width ratio of the obstacles in the current work channel; The obstacle avoidance path planning mode of the unmanned ploughing machine is determined based on the width ratio. The obstacle avoidance path planning mode includes a complete supplementary tillage mode and a simplified detour mode. When the obstacle avoidance path planning mode is determined to be a complete supplementary tillage mode, the unmanned ploughing machine is controlled to perform a preset number of short-distance reciprocating tillage operations behind and / or in front of the target obstacle to supplement the uncultivated area around the obstacle. Then, while keeping the plow in the lowered state, the plow bypasses the target obstacle and returns to the global operation path based on the detour return path.

2. The method as described in claim 1, characterized in that, Also includes: The current position of the unmanned ploughing machine and the global operation path within a preset distance range ahead are obtained. Based on the global operation path, a current operation channel is generated. The current operation channel takes the center line of the global operation path within the preset distance range as a reference and expands to the left and right with a preset width to form a rectangular decision area. The preset width is half the width of the plow. The calculation of the width percentage of the obstacle in the current working channel includes: Within the current work channel, a set of obstacles intersecting with the current work channel is detected. For each obstacle, its projected width in the width direction of the current work channel is calculated. The projection ratio is determined based on the ratio of the projected width to the width of the current work channel to determine the width ratio.

3. The method as described in claim 1, characterized in that, The obstacle avoidance path planning mode of the unmanned tiller is determined based on the width ratio. The obstacle avoidance path planning mode includes a complete supplementary tillage mode and a simplified detour mode, including: When the target obstacle is in an unprocessed state or the projection ratio is greater than a preset threshold, it is determined that the complete supplementary tillage mode is adopted. When the target obstacle is in a state of being supplemented with farmland and the projection ratio is less than or equal to the preset threshold, the simplified detour mode is determined to be adopted.

4. The method as described in claim 1, characterized in that, The process of controlling the unmanned tractor to perform a preset number of short-distance reciprocating tillage operations behind and / or in front of the target obstacle to replenish uncultivated areas around the obstacle, and then, while keeping the plow in the lowered state, to bypass the target obstacle and return to the global operation path based on a detour return path, includes: Control the unmanned tractor to reverse to the preset re-cultivation starting point behind the target obstacle, and perform 2 to 3 reciprocating tillage operations between the re-cultivation starting point and the preset re-cultivation ending point to cover the blank area behind the target obstacle; Control the unmanned tractor to move to the preset re-cultivation area in front of the target obstacle, and perform 2 to 3 reciprocating tillage operations to cover the blank area in front of the target obstacle; Then, while keeping the plow in the lowered state, a passable grid is constructed based on the bird's-eye view grid map, and the A* search algorithm is used to plan a detour and regression path that satisfies the minimum turning radius constraint and the safety distance constraint, so as to bypass the target obstacle with the plow and return to the global operation path.

5. The method as described in claim 1, characterized in that, Also includes: When the obstacle avoidance path planning mode is determined to be the simplified detour mode, the plow is kept in the lowered state, and the shortest detour regression path is planned based on the bird's-eye view grid map and the A* search algorithm to bypass the target obstacle and return to the global operation path. The processing status of the target obstacle is updated to "replenished with cultivation" and the corresponding operation data is recorded.

6. The method as described in claim 1, characterized in that, Also includes: Based on the geometric boundaries and safe distance of the target obstacle, a safe avoidance side is formed, and the bypass side that is far away from the main body of the target obstacle is preferred; And / or, Under the premise of ensuring safe avoidance, the curvature of the regression curve or the regression tangential deviation of different detour sides are calculated to evaluate the regression smoothness, and the detour side with better regression smoothness is selected. And / or, By combining the area or coverage benefit of the adjacent uncultivated area on the bypass side, the bypass side with higher cultivation efficiency is selected as the bypass direction.

7. The method according to any one of claims 1 to 6, characterized in that, Also includes: Before performing short-distance reciprocating tillage behind and / or in front of the target obstacle, construct the micro-topography slope field and rolling resistance field around the obstacle based on sensor data and historical operation data; In the process of generating and executing short-distance reciprocating tillage trajectories, a sideslip compensation heading and lateral micro-pulse correction mechanism are introduced. Based on real-time yaw angle, sideslip estimation and online update of the coverage confidence grid, the overlap between the re-tilled strip and the target untilled low-confidence area is maximized and strip drift is reduced.

8. A path planning device for obstacle avoidance in an unmanned ploughing machine, characterized in that, include: The identification unit is used to detect obstacles that overlap with the current working channel within the current working channel, and to calculate the width ratio of the obstacles in the current working channel; The determining unit is used to determine the obstacle avoidance path planning mode of the unmanned tiller based on the width ratio. The obstacle avoidance path planning mode includes a complete supplementary tillage mode and a simplified detour mode. The planning unit is used to control the unmanned ploughing machine to perform a preset number of short-distance reciprocating tillage operations behind and / or in front of the target obstacle to supplement the uncultivated area around the obstacle when the obstacle avoidance path planning mode is determined to be a complete supplementary tillage mode. Then, while keeping the plow in the lowered state, the ploughing machine bypasses the target obstacle and returns to the global operation path based on the detour return path.

9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program stored in the memory, implements the steps of the obstacle avoidance path planning method for an unmanned tiller as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the obstacle avoidance path planning method for the unmanned ploughing machine as described in any one of claims 1-7.

Images