Artificial intelligence-based agricultural machine efficiency collaborative optimization control method

Abstract

The application relates to the technical field of intelligent control of agricultural machinery, and discloses a farm machine efficiency collaborative optimization control method based on artificial intelligence. The method is characterized in that, around the recovery control process after the end of manual takeover, recovery control state recovery processing is first performed on multi-source operation state data, then takeover scene attribution processing is performed in combination with the geometric relationship and the environmental relative relationship of the target operation zone, on the basis of which, a recovery target set and a recovery risk index are collaboratively generated, and phased release coordination control and stability determination closed-loop updating are further performed, so that the recovery speed, the recovery direction and the timing of machine recovery can be adaptively adjusted according to different takeover scenes, and problems such as state distortion, hard recovery target, recovery sequence mismatch, machine recovery timing loss and control output mutation caused by direct recovery of automatic operation after the end of manual takeover are reduced.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology for agricultural machinery, and in particular to a collaborative optimization control method for agricultural machinery efficiency based on artificial intelligence. Background Technology

[0002] With the continuous development of agricultural machinery's autonomous driving, assisted driving, electro-hydraulic control of implements, and information technology for farmland operations, agricultural machinery such as seeders, fertilizer applicators, sprayers, and harvesters have gradually acquired the ability to automatically drive along preset work paths and simultaneously perform implement lifting, start-stop, and intensity adjustment. In large-scale planting scenarios, agricultural machinery usually needs to operate continuously within long work zones and complete operations such as deceleration, steering, belt changing, and re-entering the work zone in the field area.

[0003] In existing technologies, research and applications of automatic control for agricultural machinery mainly focus on automatic navigation, path tracking, automatic steering, speed regulation, and implement motion control. Some solutions can switch control authority between automatic and manual modes, and some can disconnect or connect the corresponding control loop after receiving a switching command, thereby enabling manual takeover or restoring automatic control. However, most of these solutions emphasize the switching action itself and lack systematic processing of the continuity of control states before and after manual takeover. In practical applications, after agricultural machinery enters the manual takeover stage, its speed, steering angle, heading, gear, implement lifting status, and other parameters are affected. The start and stop status of operations often changes, and when existing technologies resume automatic operation, they usually still re-engage control according to the target path, target speed, or target equipment status before takeover, or simply re-connect automatic control based on the current position. This fails to fully consider the state differences under different takeover scenarios such as turning at the field, temporary obstacle avoidance, and low-speed correction, and also fails to coordinate the recovery sequence and change rate between chassis control and equipment control. As a result, it is easy to cause sudden changes in steering, speed, equipment action lag, and operation trajectory deviation at the moment of recovery, ultimately leading to an increase in non-operation time at the field, fluctuations in operation quality, and a decrease in overall operation efficiency. Summary of the Invention

[0004] This application proposes an artificial intelligence-based collaborative optimization control method for agricultural machinery efficiency to address the problems mentioned in the background art.

[0005] To achieve the above objectives, this application adopts the following technical solution: an artificial intelligence-based collaborative optimization control method for agricultural machinery efficiency, comprising the following steps: The system acquires multi-source operational status data before and after manual takeover. Based on this data, it performs recovery control state restoration processing to generate a recovery control state vector, a pre-takeover automatic operation reference state, and geometric and environmental relationships for the target operation zone. Based on this recovery control state vector, geometric and environmental relationships, it performs pre-defined takeover scenario attribution processing to generate takeover type labels and scenario constraint labels. Based on these labels, the recovery control state vector, and the pre-takeover automatic operation reference state, it performs pre-defined recovery target collaborative reconstruction processing and pre-defined recovery risk quantification processing to generate a recovery target set and a recovery risk index. Based on the recovery target set, recovery risk index, and scenario constraint labels, it performs pre-defined phased mitigation and coordination control, outputs recovery control quantities, and updates the recovery control process in a closed loop based on the stability determination results obtained from pre-defined stability determination rules.

[0006] Furthermore, the multi-source operational status data includes vehicle motion status data, vehicle attitude status data, equipment execution status data, target work zone geometry data, and environmental status data. Vehicle motion status data and vehicle attitude status data are used to generate the recovery control state vector; equipment execution status data, vehicle motion status data, and vehicle attitude status data are used to generate the automatic operation reference state before takeover; target work zone geometry data are used to generate the target work zone geometric relationship; and environmental status data are used to generate the relative environmental relationship. The multi-source operational status data is also used to characterize the continuous change process of control before and after manual takeover, so as to provide a cross-time period state association basis for recovery control state restoration processing.

[0007] Furthermore, the recovery control state restoration process includes: based on preset state homogeneity constraints, performing cross-source correspondence organization and time-series mapping processing on multi-source operation state data to obtain candidate associated state data; based on preset control continuity constraints, performing abnormal segment elimination and reference state backtracking processing on candidate associated state data to obtain continuous state segments; based on preset state credibility fusion rules, performing consistency verification and restoration fusion processing on continuous state segments to generate a recovery control state vector, pre-takeover automatic operation reference state, target operation zone geometric relationship, and environmental relative relationship.

[0008] Furthermore, the pre-defined takeover scenario attribution processing includes: extracting takeover behavior evolution features, target work zone deviation features, environmental interference features, and equipment linkage features based on the recovery control state vector, the geometric relationship of the target work zone, and the relative relationship of the environment; performing pre-trained scenario attribution model inference processing based on the takeover behavior evolution features, target work zone deviation features, environmental interference features, and equipment linkage features; and performing constraint correction on the scenario attribution model inference results based on pre-defined scenario consistency verification rules to generate takeover type labels and scenario constraint labels.

[0009] Furthermore, the takeover type label includes at least one of the following: ground turn takeover label, temporary obstacle avoidance takeover label, vehicle stuck and extrication takeover label, and manual fine-tuning takeover label; the scenario constraint label includes at least one of the following: recovery speed upper limit constraint label, recovery course convergence priority label, equipment recovery access label, recovery order constraint label, and safety maintenance constraint label; the scenario constraint label is used to limit the target generation boundary in the recovery target collaborative reconstruction process and the stage execution boundary in the phased mitigation coordination control.

[0010] Furthermore, the pre-defined recovery target collaborative reconstruction processing includes: performing recovery phase division processing based on takeover type label, scenario constraint label, recovery control state vector, and pre-takeover automatic operation reference state to obtain a recovery phase sequence corresponding to the current takeover scenario; performing recovery target association mapping processing and target consistency correction processing based on the recovery phase sequence to generate a recovery target set; the recovery target set includes at least recovery target speed, recovery target steering angle, recovery target equipment position status, and recovery sequence label, and the recovery target speed, recovery target steering angle, and recovery target equipment position status constitute a mutually coupled recovery target combination under the constraint of the recovery sequence label.

[0011] Furthermore, the pre-defined recovery risk quantification process includes: determining the trajectory deviation evolution risk, environmental interference risk, and equipment recovery mismatch risk based on the recovery control state vector, relative environmental relationships, and pre-takeover automatic operation reference state; and performing joint characterization processing on the trajectory deviation evolution risk, environmental interference risk, and equipment recovery mismatch risk based on a pre-defined hierarchical collaborative quantification method to generate a recovery risk index.

[0012] Furthermore, the pre-defined collaborative reconstruction processing of recovery targets also includes: based on the recovery risk index, performing risk level adaptive constraint correction processing, and performing convergence enhancement correction, amplitude suppression correction, stage delay correction, or safety maintenance correction on at least one recovery target in the recovery target set; wherein, the risk level adaptive constraint correction processing includes at least one of the following: performing risk suppression correction on the recovery target speed, performing convergence enhancement correction on the recovery target steering angle, performing access delay correction on the recovery target equipment position status, and performing order rearrangement correction on the recovery sequence label.

[0013] Furthermore, the pre-defined phased mitigation and coordination control includes: determining recovery control quantities based on the recovery target set, including steering control quantities, longitudinal control quantities, and equipment control quantities; performing phase enable allocation and mitigation output adjustment on steering control quantities, longitudinal control quantities, and equipment control quantities based on recovery sequence labels and scenario constraint labels; and triggering subsequent phase control objects to enter the effective recovery phase when the controlled objects in the preceding phase meet the pre-defined transition stability conditions, thereby enabling the steering control quantities, longitudinal control quantities, and equipment control quantities to form a recovery control chain that is sequentially controlled, amplitude-limited, and mutually coordinated.

[0014] Furthermore, the closed-loop update recovery control process based on the stability determination result obtained from the preset stability determination rules includes: performing a joint determination on the trajectory convergence state, velocity convergence state, and equipment readiness state of the current recovery control process based on the preset stability determination rules to obtain the stability determination result; when the stability determination result meets the preset recovery completion conditions, switching to the normal automatic operation control mode; when the stability determination result does not meet the preset recovery completion conditions and the preset safety risk conditions are not triggered, re-executing the preset takeover scenario attribution processing, preset recovery target collaborative reconstruction processing, preset recovery risk quantification processing, and preset phased mitigation coordination control based on the updated recovery control state vector; and executing the preset safety maintenance control or manual maintenance control when the preset safety risk conditions are triggered.

[0015] The beneficial effects of this invention are as follows: This invention focuses on the recovery control process after manual takeover. First, it performs recovery control state restoration processing on multi-source operational state data. Then, it performs takeover scenario attribution processing by combining the geometric relationship of the target operational zone and the relative relationship of the environment. On this basis, it collaboratively generates a recovery target set and a recovery risk index, and further performs phased slow-release coordinated control and stability judgment closed-loop update. This enables adaptive adjustment of recovery speed, recovery direction, and machine recovery timing according to different takeover scenarios. It reduces problems such as state distortion, rigid recovery targets, recovery sequence mismatch, machine recovery time delay, and sudden changes in control output caused by directly resuming automatic operation after manual takeover. This improves the stability, safety, and operational continuity of the recovery control process after manual takeover, thereby enhancing the overall efficiency of automatic agricultural machinery operation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort: Figure 1This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example

[0019] like Figure 1 As shown, this invention discloses an artificial intelligence-based collaborative optimization control method for agricultural machinery efficiency, comprising the following steps: Step S1: Obtain multi-source operation status data before and after manual takeover triggering; perform recovery control status restoration processing based on multi-source operation status data to generate recovery control status vector, automatic operation reference status before takeover, geometric relationship of target operation zone and relative relationship of environment; Step S2: Based on the recovery control state vector, the geometric relationship of the target work zone, and the relative relationship of the environment, perform the preset takeover scenario attribution processing to generate takeover type labels and scenario constraint labels; Step S3: Based on the takeover type label, scenario constraint label, recovery control state vector, and pre-takeover automatic operation reference state, perform preset recovery target collaborative reconstruction processing and preset recovery risk quantification processing to generate a recovery target set and a recovery risk index. Step S4: Based on the recovery target set, recovery risk index and scenario constraint labels, execute the preset phased mitigation coordination control, output the recovery control quantity, and update the recovery control process in a closed loop based on the stability judgment result obtained from the preset stability judgment rule.

[0020] Step S1 addresses the issues of input state distortion, unstable reference state, and asynchronous environmental and work zone constraints when directly resuming automatic operation after manual takeover. To this end, this step focuses on the recovery control process, performing recovery control state restoration processing based on multi-source work state data to generate a recovery control state vector, the automatic operation reference state before takeover, the geometric relationship of the target work zone, and the relative relationship of the environment. These serve as the unified input basis for step S2 to perform takeover scenario attribution processing.

[0021] Specifically, the multi-source operational status data includes vehicle motion status data, vehicle attitude status data, implement execution status data, target work zone geometry data, and environmental status data. The vehicle motion status data and vehicle attitude status data are used to generate the recovery control state vector. The implement execution status data, vehicle motion status data, and vehicle attitude status data are used to generate the pre-takeover automatic operation reference state. The target work zone geometry data is used to generate the target work zone geometric relationship, and the environmental status data is used to generate the environmental relative relationship. The above data can be provided by satellite positioning unit, inertial measurement unit, wheel speed acquisition unit, drive control unit, steering sensing unit, implement controller, hydraulic position sensing unit, and environmental perception unit. This part belongs to the supporting technical means that can be directly adopted in this field. The innovation of this implementation method lies in the recovery control state restoration processing, rather than the specific acquisition hardware itself.

[0022] In a preferred embodiment, a data processing window is established using the start and end times of manual takeover as time anchors. A preset time period before the start time of manual takeover is used to extract the reference state of automatic operation before takeover, and the current data corresponding to the end time of manual takeover is used to generate the recovery control state vector. The preset time period is preferably set to 2 to 8 seconds, more preferably 3 to 5 seconds. The reason is that if the time period is too short, it is difficult to cover a sufficiently long stable automatic operation segment; if the time period is too long, it is easy to introduce early operation states that are unrelated to the current recovery control process.

[0023] To ensure that data from different sources truly correspond to the same agricultural machinery, the same work task, and the same control process, cross-source correspondence is first performed based on preset state-source constraints. State-source constraints include at least equipment identifier consistency constraints, work task identifier consistency constraints, and spatiotemporal proximity consistency constraints. Specifically, when two data records satisfy the following conditions: identical equipment identifier, identical work task identifier, time difference not exceeding a preset source-source time threshold, and spatial distance not exceeding a preset source-source distance threshold, they are determined to belong to the same state evolution segment of the same control object. The preset source-source time threshold is preferably 0.02 seconds to 0.30 seconds, with 0.02 seconds to 0.10 seconds preferred for vehicle motion state data and vehicle attitude state data, and 0.05 seconds to 0.30 seconds preferred for implement execution state data. The preset source-source distance threshold is preferably 0.10 meters to 1.50 meters. If the work task identifier is missing, alternative determinations can be made based on equipment identifier, continuous trajectory consistency, and target work zone affiliation to ensure that state-source constraints can be implemented in different agricultural machinery control systems.

[0024] After completing the cross-source correspondence processing, time-series mapping is performed based on the unified control time base. The control period of the unified control time base is preferably 0.02 seconds to 0.10 seconds, more preferably 0.02 seconds to 0.05 seconds. For continuous state variables such as vehicle speed, vehicle position, vehicle heading angle, steering state parameters, and equipment position status, the most recent valid sampled value of the corresponding state variable is read with the current control time as the target time. Combined with the local rate of change of the state variable at the most recent sampled time, a first-order time-series mapping is performed on the state variable to obtain the candidate state value at the current control time. The local rate of change can be obtained by the difference between two adjacent valid sampled values, or by... The slope of the linear fitting within a local window of a preset length is obtained. To prevent extrapolation distortion caused by excessive time delay, a maximum allowable mapping delay is set for different state variables. For vehicle speed and vehicle heading angle, the maximum allowable mapping delay is preferably 0.05 seconds to 0.20 seconds; for vehicle position, the maximum allowable mapping delay is preferably 0.05 seconds to 0.30 seconds; and for equipment position status, the maximum allowable mapping delay is preferably 0.10 seconds to 0.50 seconds. If the time difference between a certain state variable and the most recent valid sampling time exceeds the maximum allowable mapping delay corresponding to that state variable, then that candidate state value will not enter the subsequent restoration and fusion process.

[0025] After time-series mapping, based on preset control continuity constraints, abnormal segment removal and reference state backtracking are performed on the candidate associated state data. Control continuity constraints are used to identify and remove abnormal segments that do not conform to the physical laws of low-speed recovery control of agricultural machinery. Specifically, continuity detection quantities are constructed around vehicle speed, steering state parameters, implement position status, and vehicle heading angle. Vehicle speed continuity is obtained by dividing the difference in vehicle speed between adjacent control times by the control period to obtain the speed change rate; steering state parameter continuity is obtained by dividing the difference in steering state parameters between adjacent control times by the control period to obtain the steering change rate; implement position status continuity is obtained by dividing the difference in implement position status between adjacent control times by the control period to obtain the implement position change rate; vehicle heading angle continuity is obtained by dividing the difference in heading angle between adjacent control times by the control period after angle retracement processing to obtain the heading change rate. If any of the rates of change exceeds the corresponding preset continuity threshold, the corresponding data segment is determined to be an abnormal segment. The vehicle speed continuity threshold is preferably 2.0 m / s² to 6.0 m / s², the steering continuity threshold is preferably 0.20 radians / second to 1.20 radians / second, the equipment continuity threshold is preferably 0.05 radians / second to 0.50 radians / second, and the heading continuity threshold is preferably 0.10 radians / second to 1.00 radians / second. In addition to continuity constraints, candidate segments must also meet preset safety constraints in the relative environmental relationships. The preset safety constraints include at least that the minimum distance from the vehicle to the nearest obstacle is greater than the preset obstacle safety distance, and the minimum distance from the vehicle to the edge of the terrain is greater than the preset edge turning warning distance. Only segments that simultaneously meet the control continuity constraints and the preset safety constraints are used as continuous state segments for subsequent reference state backtracking and state credibility fusion.

[0026] Based on the continuous state segments after removing abnormal segments, further reference state backtracking processing is performed to generate an automatic operation reference state before takeover. In existing technologies, the last sampled state before manual takeover is often directly used as the control recovery reference. However, this method is easily affected by instantaneous deceleration before takeover, obstacle approach, or operator pre-intervention, causing the reference state to fail to accurately reflect the stable automatic operation baseline before takeover. To solve this problem, this implementation selects continuous state segments that meet control continuity constraints and preset safety constraints within a preset time period before the start of manual takeover, and assigns different values ​​to each valid state sample according to temporal proximity and state stability. The weights are determined as follows: the closer the effective state sample is to the start time of manual takeover and the smaller the overall continuity deviation, the higher its weight; the farther the effective state sample is from the start time of manual takeover or the larger the overall continuity deviation, the lower its weight. The time decay coefficient is preferably 0.2 to 2.0, and the stability decay coefficient is preferably 0.5 to 5.0. Then, the vehicle motion state data, vehicle attitude state data, and equipment execution state data in the candidate segment are weighted and aggregated according to the above weights to obtain the automatic operation reference state before takeover. The automatic operation reference state before takeover obtained in this way is no longer a single-moment sampling result, but a representative state of the stable automatic operation segment before takeover.

[0027] After obtaining continuous state segments and the pre-takeover automatic operation reference state, a consistency check and restoration fusion process is further performed based on preset state reliability fusion rules to generate a restored control state vector. The restored control state vector preferably includes vehicle position, vehicle heading angle, vehicle speed, steering state parameters, equipment position state, operation start / stop state, and wheel slip state. Since vehicle position, vehicle heading angle, vehicle speed, steering state parameters, and equipment position state are continuous state variables, while operation start / stop state is a discrete state variable, this embodiment restores both continuous and discrete state variables separately. For continuous state variables, candidate mapping values ​​from all valid data sources at the current control moment are used as input, and normalized weighted fusion is performed according to the state reliability weights of each data source to obtain the restored result of the continuous state variable. The state reliability weights are jointly determined by source reliability, time delay freshness, and consistency deviation. Reliability reflects the historical stability, calibration accuracy, or type reliability of the corresponding data source for that state variable. Time delay freshness reflects the weight of the data source as it is closer to the current control moment. Consistency deviation reflects the weight of the data source as it is smaller than the state difference with other data sources of the same type. The source reliability base weight can be preset according to the sensor type, historical failure rate, calibration accuracy, or factory configuration. Consistency deviation can be obtained based on the state deviation between the current data source and other similar valid data sources. The time delay suppression coefficient is preferably 3 to 20 seconds, and the deviation suppression coefficient is preferably 2 to 10. For the vehicle heading angle, since it belongs to the circular angle state, the heading angle is restored by fusing the sine and cosine components separately and then back-calculating the heading angle to avoid generating false average results when approaching the positive and negative circular boundaries. For discrete state variables such as operation start and stop states, the final state is determined by the weighted voting results of each valid data source.

[0028] In this embodiment, the wheel slip state is also determined based on vehicle speed, drive wheel angular velocity, and equivalent radius of the drive wheel to enhance the ability of the recovery control state vector to express the current ground adhesion. Specifically, the difference between the theoretical linear velocity of the drive wheel and the actual vehicle speed is first calculated, and then this difference is divided by the maximum value among the theoretical linear velocity of the drive wheel, the actual vehicle speed, and the preset minimum protection speed. If the calculation result is less than zero, it is treated as zero, thereby obtaining the wheel slip ratio. The preset minimum protection speed is preferably 0.05 m / s to 0.20 m / s to avoid denominator degradation under low-speed limit conditions. Furthermore, the wheel slip state can be determined according to the wheel slip ratio. The system is divided into three states: stable adhesion, slight wheel slip, and significant wheel slip. The stable adhesion state corresponds to a wheel slip ratio of less than 0.08, the slight wheel slip state corresponds to a wheel slip ratio of 0.08 to 0.20, and the significant wheel slip state corresponds to a wheel slip ratio of not less than 0.20. The wheel slip state is not the final control target of this step, but rather an important component of the recovery control state vector, providing support for the subsequent step S2 to identify the stuck vehicle escape takeover label and its corresponding scenario constraint label. Preferably, the wheel slip state is directly calculated only under driving conditions; under braking conditions or coasting conditions in neutral, the previous effective wheel slip state can be maintained or the current wheel slip state can be marked as invalid.

[0029] In addition to restoring the control state vector and the pre-takeover automatic operation reference state, this step also generates the geometric relationship and environmental relative relationship of the target operation zone. The geometric relationship of the target operation zone preferably includes at least the nearest point of the target operation zone corresponding to the current vehicle position, the tangential direction of the target operation zone, and the local curvature of the target operation zone. The environmental relative relationship preferably includes at least the minimum distance from the vehicle to the edge of the field, the minimum distance from the vehicle to the nearest obstacle, and the orientation relationship of the obstacle relative to the vehicle's forward operation direction. The geometric relationship of the target operation zone can be calculated from the centerline of the target operation zone and the current position of the vehicle. The centerline of the target operation zone can be obtained from the preset AB operation line, the land boundary vector, the historical operation trajectory polyline, or the operation planning file. When the centerline of the target operation zone is a discrete polyline, the nearest point of the target operation zone can be determined by the perpendicular projection distance from the current position of the vehicle to each polyline segment, and the direction of the nearest polyline segment can be used as the tangential direction of the target operation zone. When the centerline of the target operation zone is a spline curve, the nearest point of the target operation zone can be determined by the search result of the nearest point of the spline curve from the current position of the vehicle, and the tangential direction of the target operation zone can be determined by the first derivative of the spline curve. The local curvature of the target work zone can be estimated using the ratio of the angle between adjacent centerline segments to the corresponding arc length, or it can be calculated using the second derivative of the spline curve. The relative environmental relationships can be calculated using the obstacle center point, the boundary line of the work area, and the vehicle pose output by the environmental perception unit. Specifically, the minimum distance from the vehicle to the boundary line can be determined using the shortest distance from a point to the boundary line segment, and the minimum distance from the vehicle to the nearest obstacle can be determined using the shortest distance between the vehicle's circumscribed rectangle and the obstacle's outline. The orientation of the obstacle relative to the vehicle's forward work direction can be determined by transforming the obstacle's center point to the vehicle coordinate system and then using the lateral coordinate sign and longitudinal projection distance. The aforementioned centerline projection, spline fitting, point-to-line segment distance calculation, and coordinate system transformation are all commonly used engineering implementation methods in the field of agricultural machinery navigation and mobile robot positioning. The innovation of this implementation lies in using the geometric relationship of the target work zone and the relative environmental relationships as the direct output of the recovery control state restoration process, placing them under the same reference basis and control semantics as the recovery control state vector and the pre-takeover automatic operation reference state.

[0030] Through the above processing, step S1 ultimately outputs a recovery control state vector, a pre-takeover automatic operation reference state, a target operation zone geometric relationship, and an environmental relative relationship. The recovery control state vector is the core output of step S1, while the pre-takeover automatic operation reference state, target operation zone geometric relationship, and environmental relative relationship are associated outputs generated around the recovery control state vector. These three elements work together to support the takeover scenario attribution processing in step S2. Through recovery control state restoration processing, a recovery control state vector consistent with the current recovery control process is generated from asynchronous, multi-source operation state data with varying credibility. Simultaneously, the pre-takeover automatic operation reference state, target operation zone geometric relationship, and environmental relative relationship are generated, providing a unified input basis for step S2.

[0031] Step S2 addresses the issue of inconsistent handling of different manual takeover scenarios, leading to a lack of scenario adaptability in the recovery control boundary. In existing technologies, after a manual takeover ends, a simple judgment is usually made based on whether a manual takeover occurred or the vehicle's status, followed by direct restoration to the automatic operation mode. This fails to distinguish whether the manual takeover was caused by turning at the edge of the terrain, temporary obstacle avoidance, vehicle extrication from a stuck situation, or manual fine-tuning. Furthermore, it fails to transform the takeover scenario into a recovery boundary that can be directly invoked in subsequent recovery control processes. Therefore, based on the recovery control state vector, the geometric relationship of the target work zone, and the relative relationship of the environment output in step S1, a preset takeover scenario attribution processing is performed to generate a takeover type label and a scenario constraint label. The takeover type label is the core output of step S2, used to characterize the scenario category to which the current manual takeover belongs. The scenario constraint label is an associated output generated around the takeover type label, used to characterize the recovery boundary and stage execution boundary corresponding to the takeover scenario, and is invoked by subsequent steps S3 and S4.

[0032] Specifically, the pre-defined takeover scenario attribution processing includes: extracting takeover behavior evolution features, target work zone deviation features, environmental interference features, and equipment linkage features based on the recovery control state vector, the geometric relationship of the target work zone, and the relative relationship of the environment; performing pre-trained scenario attribution model inference processing based on the takeover behavior evolution features, target work zone deviation features, environmental interference features, and equipment linkage features; and performing constraint correction on the scenario attribution model inference results based on pre-defined scenario consistency verification rules to generate takeover type labels and scenario constraint labels. The takeover type labels include at least one of the following: takeover at the edge of the field turning point, temporary obstacle avoidance, vehicle stuck and extricated, and manual fine-tuning takeover. The scenario constraint labels include at least one of the following: recovery speed upper limit constraint label, recovery heading convergence priority label, equipment recovery access label, recovery order constraint label, and safety maintenance constraint label.

[0033] In a preferred embodiment, an attribution window for the takeover scenario is established backward, using the end time of manual takeover as the attribution reference time. The attribution window preferably covers a continuous state sequence from 0.5 to 3.0 seconds before the end time of manual takeover, and more preferably covers a continuous state sequence from 1.0 to 2.0 seconds before the end time of manual takeover. When the attribution window is too short, it is difficult to fully characterize the evolution of vehicle speed changes, heading corrections, steering corrections, machine actions, and environmental relationship changes during manual takeover. When the attribution window is too long, it is easy to introduce earlier stages of states unrelated to the current recovery control process. Therefore, when the duration of manual takeover is shorter than the minimum length of the attribution window, the start time of manual takeover can be used as the starting point of the window. When the duration of manual takeover is longer than the minimum length of the attribution window, it is preferable to extract the most recent valid state segment before the end of manual takeover as the attribution input.

[0034] In this embodiment, the evolutionary features of the takeover behavior are first extracted. These features characterize the temporal changes in vehicle control behavior during manual takeover. Preferably, they include at least speed change features, heading change features, steering change features, and takeover duration features. Speed ​​change features can be obtained by dividing the difference in vehicle speed between adjacent moments within the takeover scenario attribution window by the corresponding time interval, further forming the mean speed change, peak speed change, and speed change fluctuation. Heading change features can be obtained by dividing the difference in vehicle heading angle between adjacent moments after angle folding by the corresponding time interval, further forming the mean heading change, peak heading change, and heading change direction consistency. Specifically, heading change... Directional consistency can be obtained by the proportion of samples with the same heading direction within the attribution window of the takeover scenario. The turning change characteristics can be obtained by dividing the difference in turning state parameters at adjacent moments by the corresponding time interval, and further forming the mean turning change, peak turning change, and degree of turning change repetition. Among them, the degree of turning change repetition can be obtained by the number of times the turning change direction changes within the attribution window of the takeover scenario. The takeover duration characteristics are obtained by the length of time from the start time to the end time of the manual takeover. By introducing the evolution characteristics of takeover behavior, the manual takeover can be transformed from a single event into a behavioral segment with temporal evolution meaning, thereby enhancing the ability of takeover scenario attribution processing to distinguish complex takeover causes.

[0035] While extracting the evolutionary features of the takeover behavior, the deviation features of the target work zone are also extracted. The deviation features of the target work zone are used to characterize the degree, direction, and trend of deviation of the vehicle relative to the geometric relationship of the target work zone during manual takeover. Preferably, they include at least the lateral deviation features of the target work zone, the heading deviation features of the target work zone, and the curvature correlation features of the target work zone. The lateral deviation features of the target work zone can be directly characterized by the normal distance from the vehicle's current position to the nearest point of the target work zone. The heading deviation features of the target work zone can be characterized by the angle difference between the vehicle's current heading and the tangential direction of the target work zone. The curvature correlation features of the target work zone can be characterized by the consistency between the local curvature direction of the target work zone and the vehicle's heading correction direction. On this basis, the lateral deviation change rate, the heading deviation change rate, and the convergence or divergence trend of the deviation amount of the target work zone can be further extracted. By introducing the deviation features of the target work zone, the attribution processing of the takeover scenario can be given a working geometric background, thereby avoiding the simple transfer of the ordinary vehicle takeover identification logic to the agricultural machinery scenario.

[0036] Furthermore, environmental disturbance features are extracted. These features characterize the degree to which the relative environmental relationships during manual takeover restrict the recovery control process. Preferably, they include at least boundary approach features, obstacle disturbance features, and disturbance direction features. Boundary approach features can be determined by the minimum distance from the vehicle to the edge of the work area and its changing trend within the takeover scenario attribution window. Obstacle disturbance features can be determined by the minimum distance from the vehicle to the nearest obstacle, the projection distance of the nearest obstacle in the vehicle's forward working direction, and the duration of obstacle disturbance. The duration of obstacle disturbance can be determined by the distance from the vehicle to the nearest obstacle within the takeover scenario attribution window. The cumulative duration during which the distance to the nearest obstacle is continuously less than the preset obstacle safety distance is determined. The interference direction characteristics can be determined by the relationship between the obstacle and the left, right, or front offset relative to the vehicle's forward operating direction. The preset obstacle safety distance is preferably set to 1.0 meter to 6.0 meter, more preferably 1.5 meter to 4.0 meter; the preset turning warning distance is preferably set to 2.0 meter to 12.0 meter, more preferably 3.0 meter to 8.0 meter. Through environmental interference characteristics, the external causes of manual takeover and the external constraints faced by the current recovery control process can be simultaneously included in the scene attribution processing.

[0037] In addition, machine linkage features are extracted to characterize the linkage relationship between the machine's execution state and the vehicle's control state during manual takeover. Preferably, these features include at least machine position change features, operation start-stop change features, and temporal relationship features between machine actions and vehicle control state changes. Machine position change features can be obtained from the amount and rate of change of machine position state within the takeover scenario attribution window. Operation start-stop change features can be obtained from the number of times and the time of switching of operation start-stop state within the takeover scenario attribution window. The temporal relationship features between machine actions and vehicle control state changes can be determined according to the order of the time when the machine position state changes significantly relative to the time of vehicle deceleration, steering change, and heading change. By using machine linkage features, the attribution processing of the takeover scenario can avoid starting only from the vehicle's own state and ignoring the actual impact of machine actions on the recovery control boundary in the agricultural machinery scenario.

[0038] After extracting the above four types of features, the pre-trained scene attribution model is used for inference processing. The pre-trained scene attribution model can be implemented using gated recurrent networks, temporal convolutional networks, lightweight transformer networks, or other models suitable for short-term behavior recognition. This part belongs to the supporting technical means that can be directly adopted in this field. The innovation of this implementation method does not lie in the model structure itself, but in the composition of the model input features and the control semantic association between the model output and the subsequent recovery control process. Preferably, the takeover behavior evolution features, target operation zone deviation features, environmental interference features, and equipment linkage features are arranged in chronological order to form a multi-dimensional model. To avoid the impact of differences in the dimensions of different features on the model's inference results, the feature sequence input scene attribution model can be normalized before input. Speed ​​change features, heading change features, and turning change features can be normalized according to their respective preset physical upper limits. Target operation zone deviation features and environmental interference features can be normalized according to the corresponding safety thresholds or warning thresholds. Equipment position change features can be directly input according to the normalized range of equipment position status. The scene attribution model outputs the initial confidence scores corresponding to four types of takeover scenarios, which correspond to the ground turning takeover label, temporary obstacle avoidance takeover label, stuck vehicle escape takeover label, and manual fine-tuning takeover label, respectively.

[0039] To ensure that the scenario consistency verification generates executable confidence correction results, this implementation calculates a scenario verification score for each takeover scenario, with the score ranging from 0 to 1. Boundary consistency verification, obstacle consistency verification, attachment consistency verification, and equipment timing consistency verification each output corresponding sub-verification scores. When a verification condition is fully satisfied, the corresponding sub-verification score is 1; when it is not satisfied at all, it is 0; and when only some conditions are satisfied, the score is determined by the ratio of the number of satisfied conditions to the total number of verification conditions of that type. The scenario verification score for each takeover scenario is obtained by weighting the sub-verification scores related to that scenario. Specifically, for takeover scenarios involving turning at the edge, boundary consistency verification scores and equipment timing consistency verification scores are prioritized; for temporary obstacle avoidance takeover scenarios, obstacle consistency verification scores are prioritized; for vehicle stuck and escaping takeover scenarios, attachment consistency verification scores are prioritized; and for manual fine-tuning takeover scenarios, verification scores corresponding to the target work zone deviation, vehicle attitude fine-tuning, and equipment linkage strength are prioritized. The effective confidence score is obtained by normalizing the weighted sum of the initial confidence score output by the scene attribution model and the scene verification score. The initial confidence score weight is preferably between 0.50 and 0.80, and the scene verification score weight is preferably between 0.20 and 0.50. When historical calibration samples are lacking, the initial confidence score weight is set to 0.60, and the scene verification score weight is set to 0.40. In this way, scene consistency verification no longer merely performs textual corrections to the model inference results, but instead generates a directly comparable effective confidence score.

[0040] In one possible implementation, the pre-trained scene attribution model is obtained through offline training. Specifically, historical operation segments of multiple agricultural machines are collected from the start to the end of manual takeover in scenarios such as turning at the edge of the field, temporary obstacle avoidance, getting stuck and escaping, and manual fine-tuning. Each historical operation segment is converted into a multi-dimensional feature sequence corresponding to takeover behavior evolution features, target operation zone deviation features, environmental interference features, and machine linkage features, using the same processing method as in this step. Furthermore, each historical operation segment is labeled with a takeover type tag based on manual operation records, obstacle detection records, roller skating status records, and machine action records. The labeled multi-dimensional feature sequence is then used as training data. The samples, with the takeover type label as the supervision label, are used to perform offline training on the scene attribution model. The training objective is preferably to reduce the classification loss between the model output scene probability distribution and the supervision label. The classification loss is preferably cross-entropy loss or weighted cross-entropy loss. For takeover scenes with a small number of samples, the influence of class imbalance can be suppressed by increasing the corresponding class loss weight. After the model training is completed, the trained model parameters are deployed on the agricultural machinery controller or edge computing unit. In the recovery control stage after the manual takeover ends, the multi-dimensional feature sequence corresponding to the takeover scene attribution window is read and the initial confidence level corresponding to each takeover scene is output.

[0041] Before being included in the training sample set, historical operation segments used for offline training undergo sample validity screening. This screening includes at least four criteria: completeness of start and end times of takeover, continuity of positioning, physical rationality of attitude, synchronization of equipment status, and credibility of environmental perception. The completeness of start and end times of takeover is used to remove historical segments lacking the start or end time of manual takeover. The continuity of positioning is used to remove historical segments with missing positioning data for more than three consecutive control cycles. The physical rationality of attitude is used to remove historical segments with a heading angle jump or a rate of change in steering parameters exceeding the upper limit of the agricultural machinery's physical steering. The synchronization of equipment status is used to remove historical segments with missing synchronization timestamps for equipment position status or where the start and stop status of the operation cannot be aligned with the vehicle status. The credibility of environmental perception is used to remove historical segments with obstacle detection confidence levels below a preset lower limit of perception confidence, preferably between 0.50 and 0.80. When the number of valid training samples for any takeover scenario is lower than the preset minimum number of samples, the scenario attribution model output is not used as the primary criterion. Instead, candidate scenario matching rules and scenario consistency verification rules are used to generate takeover type labels. The preset minimum number of samples is preferably no less than 30 historical operation segments for each type of takeover scenario. This process avoids instability in the scenario attribution model output caused by insufficient historical samples or the inclusion of abnormal samples.

[0042] In another optional implementation, the takeover scenario attribution processing may not employ a pre-trained scenario attribution model. Instead, candidate scenario matching rules are constructed based on boundary consistency checks, obstacle consistency checks, attachment consistency checks, and equipment timing consistency checks. Rule matching scores are then calculated for each takeover scenario. Takeover type labels are determined based on the rule matching scores. When multiple takeover scenarios simultaneously meet the type confirmation conditions, the takeover type label is determined according to the principle of prioritizing takeover scenarios with higher recovery constraint strength. When none of the type confirmation conditions are met, the takeover type label is set to a manually fine-tuned takeover label, or the previous valid takeover type label is retained and a safety maintenance constraint label is generated. Subsequently, scenario constraint labels are generated based on the takeover type labels.

[0043] In this embodiment, the output of the scene attribution model is not directly used as the final takeover type label. Instead, the initial confidence level is constrained and corrected based on preset scene consistency verification rules. Specifically, the scene consistency verification rules include at least boundary consistency verification, obstacle consistency verification, attachment consistency verification, and equipment timing consistency verification. Boundary consistency verification is used to determine whether the minimum distance from the vehicle to the edge of the work area, its changing trend, and the deviation trend of the target work area support the takeover label for turning at the edge of the work area. Preferably, when the minimum distance from the vehicle to the edge of the work area is less than the preset turn warning distance, and this distance continuously decreases within the takeover scene attribution window, and the deviation feature of the target work area has a consistent directional relationship with the local curvature change of the target work area, the effective confidence level corresponding to the turn takeover label for turning at the edge of the work area is increased. When the above conditions are obviously not met, the effective confidence level corresponding to the turn takeover label for turning at the edge of the work area is decreased. Obstacle consistency verification is used to determine whether the minimum distance from the vehicle to the nearest obstacle, the obstacle's orientation, and the manual steering correction direction support the temporary obstacle avoidance takeover label for turning at the edge of the work area. Preferably, when the minimum distance from the vehicle to the nearest obstacle is less than the preset obstacle safety distance, and the relative orientation of the obstacle is less than the manual steering correction direction, the effective confidence level corresponding to the turn takeover label for turning at the edge of the work area is increased. When the steering correction direction is consistent, the effective confidence level corresponding to the temporary obstacle avoidance takeover label is increased; when the above conditions are clearly not met, the effective confidence level corresponding to the temporary obstacle avoidance takeover label is decreased. The attachment consistency check is used to determine whether the wheel slip condition, vehicle speed recovery capability, and changes in driving state support the stuck vehicle takeover label. Preferably, when the wheel slip condition is significant and the vehicle speed increase within the takeover scenario attribution window is lower than a preset escape speed increase threshold, the effective confidence level corresponding to the stuck vehicle takeover label is increased. The preset escape speed increase threshold is preferably set to 0.05 meters per second. Up to 0.50 meters per second, the machine timing consistency check is used to determine whether the sequence of machine position changes and operation start-stop changes relative to vehicle deceleration, steering correction, and heading correction supports the headway turn takeover label or manual fine-tuning takeover label. Preferably, when machine lifting or operation start-stop closing occurs before or approximately synchronously with the vehicle steering change, and the target work zone deviation characteristics continue to increase, the effective confidence level corresponding to the headway turn takeover label is increased; when the machine linkage characteristics are weak and the vehicle only shows a small attitude correction, the effective confidence level corresponding to the manual fine-tuning takeover label is increased.

[0044] After scenario consistency verification, takeover type labels are generated. Preferably, when the effective confidence level corresponding to a certain takeover scenario reaches the preset type confirmation threshold and is higher than the effective confidence level corresponding to other takeover scenarios, the takeover scenario is determined as the final takeover type label. The preset type confirmation threshold is preferably set to 0.45 to 0.80, more preferably 0.55 to 0.70. If the effective confidence levels corresponding to multiple takeover scenarios reach the preset type confirmation threshold at the same time, and the difference between them is less than the preset conflict threshold, the final takeover type label is determined according to the priority order of recovery restriction strength. The strength of the recovery constraint is determined based on the conservatism of the safety maintenance constraint, the equipment recovery access constraint, the recovery speed upper limit constraint, and the recovery course convergence priority. In this embodiment, the priority order of the recovery constraint strength is as follows: temporary obstacle avoidance takeover label, stuck vehicle escape takeover label, ground turn takeover label, and manual fine-tuning takeover label. The preset conflict threshold is preferably set to 0.05 to 0.20. If the effective confidence of all takeover scenarios is lower than the preset type confirmation threshold, a conservative strategy is preferred, which is to set the takeover type label to the manual fine-tuning takeover label, or to maintain the previous effective takeover type label and generate a safety maintenance constraint label simultaneously, so as to avoid using an overly aggressive recovery boundary when there is insufficient scenario evidence.

[0045] After determining the takeover type label, scenario constraint labels are generated based on the takeover type label, the constraint strength in the relative environmental relationship, and the temporal relationship in the equipment linkage characteristics. The scenario constraint labels are not repetitive names of the takeover scenario, but rather a structured expression of the constraints imposed by the takeover scenario on the subsequent recovery control process. Preferably, the recovery speed upper limit constraint label is used to limit the maximum allowable recovery speed in the recovery control process; the recovery heading convergence priority label is used to limit the priority of heading convergence in the recovery control process; the equipment recovery access label is used to limit the conditions under which the equipment position status and operation start / stop status are allowed to enter the operation state; the recovery sequence constraint label is used to limit the stage execution order of steering control, longitudinal control, and equipment control; and the safety maintenance constraint label is used to limit the priority of maintaining the current low-risk state in high-risk scenarios and temporarily not promoting the recovery of regular automatic operations. Furthermore, the higher the constraint strength in the relative environmental relationship, the more conservative the recovery speed upper limit constraint label and the safety maintenance constraint label tend to be; the more significant the temporal coupling relationship in the equipment linkage characteristics, the more the equipment recovery access label and the recovery sequence constraint label need to be consistent with the timing of equipment actions.

[0046] In a preferred embodiment, the scenario constraint labels corresponding to the "Temporary Obstacle Avoidance Takeover" label include: the recovery speed upper limit constraint label is set to a low-to-medium speed recovery level; the recovery heading convergence priority label is set to a high priority; the equipment recovery access label is set to access after heading deviation convergence; the recovery sequence constraint label is set to first heading convergence, then speed recovery, and then equipment recovery; and the safety maintenance constraint label is set to boundary safety maintenance. The scenario constraint labels corresponding to the "Temporary Obstacle Avoidance Takeover" label include: the recovery speed upper limit constraint label is set to a low-speed recovery level; the recovery heading convergence priority label is set to a high priority; the equipment recovery access label is set to access after obstacle is removed from a safe distance; the recovery sequence constraint label is set to first obstacle avoidance convergence, then speed recovery, and then equipment recovery; and the safety maintenance constraint label is set to... The scenario constraint labels corresponding to the obstacle safety maintenance and vehicle entrapment takeover labels include: recovery speed upper limit constraint label set to low-speed conservative level, recovery heading convergence priority label set to medium priority, equipment recovery access label set to access after adhesion recovery and stabilization, recovery sequence constraint label set to drive stabilization first, then attitude convergence, and then equipment recovery, and safety maintenance constraint label set to low adhesion safety maintenance. The scenario constraint labels corresponding to the manual fine-tuning takeover labels include: recovery speed upper limit constraint label set to normal recovery level, recovery heading convergence priority label set to medium priority, equipment recovery access label set to allow synchronous evaluation access, recovery sequence constraint label set to attitude fine-tuning and speed recovery coordinated, and safety maintenance constraint label set to fine-tuning maintenance.

[0047] Through the above processing, step S2 finally outputs the takeover type label and the scenario constraint label. The takeover type label is used to characterize which category the current manual takeover belongs to: ground turning takeover, temporary obstacle avoidance takeover, vehicle stuck get-out takeover, and manual fine-tuning takeover. The scenario constraint label is used to characterize the recovery boundary and stage execution boundary corresponding to the takeover type.

[0048] By processing the attribution of the takeover scenario, the behavioral evolution information, target operation zone deviation information, environmental interference information, and equipment linkage information before the end of the manual takeover are uniformly transformed into takeover type labels and scenario constraint labels, providing an input basis with scenario meaning and recovery boundary meaning for step S3.

[0049] Step S3 addresses the issues of a rigid recovery target, mismatch between recovery speed and recovery posture, improper timing of equipment recovery, and a fixed recovery sequence caused by directly switching back to the pre-operation automatic operation target after manual takeover. Therefore, this step does not simply call the reference speed, reference steering state parameters, and equipment position state from the pre-operation automatic operation reference state at the end of manual takeover. Instead, based on the takeover type label and scenario constraint label generated in step S2, and the recovery control state vector, pre-operation automatic operation reference state, and relative relationship with the environment generated in step S1, it performs preset recovery target collaborative reconstruction processing and preset recovery risk quantification processing to generate a recovery target set and a recovery risk index. The recovery target set is the core output of step S3, and the recovery risk index is a related output used to constrain the generation and correction of the recovery target set. Both serve as the direct input basis for the preset phased mitigation coordination control executed in step S4.

[0050] Specifically, step S3 includes: based on the takeover type label, scenario constraint label, recovery control state vector, and pre-takeover automatic operation reference state, performing preset recovery target collaborative reconstruction processing and preset recovery risk quantification processing to generate a recovery target set and a recovery risk index. The preset recovery target collaborative reconstruction processing includes: based on the takeover type label, scenario constraint label, recovery control state vector, and pre-takeover automatic operation reference state, performing recovery phase division processing to obtain a recovery phase sequence corresponding to the current takeover scenario; based on the recovery phase sequence, performing recovery target association mapping processing and target consistency correction processing to generate a recovery target set; the recovery target set includes at least recovery target speed, recovery target steering angle, recovery target equipment position status, and recovery sequence label, and the recovery target speed, recovery target steering angle, and recovery target equipment position status constitute a mutually coupled recovery target combination under the constraint of the recovery sequence label. The risk quantification process includes: determining the trajectory deviation evolution risk, environmental interference risk, and equipment recovery mismatch risk based on the recovery control state vector, relative environmental relationships, and pre-takeover automatic operation reference state; performing joint characterization processing on the trajectory deviation evolution risk, environmental interference risk, and equipment recovery mismatch risk based on a preset hierarchical collaborative quantification method to generate a recovery risk index; further, the preset recovery target collaborative reconstruction processing also includes: performing risk level adaptive constraint correction processing based on the recovery risk index, and performing convergence enhancement correction, amplitude suppression correction, stage delay correction, or safety maintenance correction on at least one recovery target in the recovery target set; wherein, the risk level adaptive constraint correction processing includes at least one of the following: performing risk suppression correction on the recovery target speed, performing convergence enhancement correction on the recovery target steering angle, performing access delay correction on the recovery target equipment position status, and performing order rearrangement correction on the recovery sequence label.

[0051] In this embodiment, the recovery risk index is represented by a weighted sum of three types of risks: trajectory deviation evolution risk, environmental disturbance risk, and equipment recovery mismatch risk. All three types of risks are first normalized to 0 to 1. The trajectory deviation evolution risk is obtained by weighting the normalized values ​​of the lateral deviation of the target work zone, the normalized values ​​of the heading deviation of the target work zone, the deviation change trend indicator value, and the local curvature correction value of the target work zone. The environmental disturbance risk is obtained by weighting the risk values ​​of the distance to the ground boundary, the distance to obstacles, and the proportion of disturbance duration. The equipment recovery mismatch risk is obtained by weighting the equipment position deviation value, the inconsistency indicator value of operation start-stop, and the mismatch indicator value of the recovery phase. When historical calibration samples are lacking, the default weights for trajectory deviation evolution risk, environmental interference risk, and equipment recovery mismatch risk in the field turn takeover scenario are 0.40, 0.25, and 0.35, respectively; in the temporary obstacle avoidance takeover scenario, the default weights are 0.25, 0.55, and 0.20, respectively; in the stuck vehicle rescue takeover scenario, the default weights are 0.25, 0.35, and 0.40, respectively; and in the manual fine-tuning takeover scenario, the default weights are 0.50, 0.20, and 0.30, respectively. After accumulating at least 30 valid historical operation segments for each type of takeover scenario, the default weights can be updated based on recovery failure records, manual secondary takeover records, and safety maintenance trigger records from the most recent operation season. Each update increment shall not exceed 0.05, and the sum of the weights for the three risk categories remains 1 after the update.

[0052] In a preferred embodiment, a recovery phase division process is first performed. This process transforms the recovery control process after manual takeover from a single-target recovery process into a phased recovery process constrained by both the takeover type label and the scenario constraint label. Specifically, the current recovery control process is first determined based on the takeover type label to belong to one of the following scenarios: a ground-turn takeover, a temporary obstacle avoidance takeover, a vehicle stuck in a rut takeover, or a manual fine-tuning takeover. Then, based on the recovery speed upper limit constraint label, recovery heading convergence priority label, equipment recovery access label, and recovery sequence constraint label in the scenario constraint labels, a recovery phase sequence corresponding to the takeover scenario is generated. Preferably, the recovery phase sequence corresponding to the ground-turn takeover includes an attitude convergence phase, a speed recovery phase, and an equipment recovery phase; the recovery phase sequence corresponding to the temporary obstacle avoidance takeover includes an obstacle removal phase, an attitude return phase, a speed recovery phase, and an equipment recovery phase; the recovery phase sequence corresponding to the vehicle stuck in a rut takeover includes a drive stabilization phase, an attitude correction phase, and an equipment recovery phase; and the recovery phase sequence corresponding to the manual fine-tuning takeover includes an attitude fine-tuning phase and a speed collaborative recovery phase.

[0053] To avoid the recovery phase sequence remaining merely conceptual, it is preferable to further define the phase switching criteria. The conditions for switching from the attitude convergence phase to the speed recovery phase can be set as follows: the lateral deviation of the current vehicle relative to the target work area is less than a preset permissible lateral deviation value, and the deviation between the vehicle's current heading and the tangential direction of the target work area is less than a preset permissible heading deviation value. Furthermore, these two deviations remain unchanged for several consecutive control cycles. The permissible lateral deviation value is preferably 0.05 meters to 0.50 meters, and the permissible heading deviation value is preferably 0.02 radians to 0.15 radians. The conditions for switching from the speed recovery phase to the equipment recovery phase can be set as follows: The recovery risk index is lower than the preset machine recovery risk threshold, and the access conditions corresponding to the machine recovery access label are met. The machine recovery risk threshold is preferably 0.25 to 0.70, more preferably 0.35 to 0.55. For the switch from obstacle removal stage to attitude return stage, it is preferred that the minimum distance from the vehicle to the nearest obstacle is greater than the preset obstacle safety distance, and this condition is maintained for several consecutive control cycles. For the switch from drive stabilization stage to attitude correction stage, it is preferred that the wheel slip state is reduced from a significant wheel slip state to a slight wheel slip state or an adhesion stable state, and the vehicle speed has a continuous positive improvement trend.

[0054] After completing the recovery phase division process, the recovery risk quantification process is performed, which includes determining the trajectory deviation evolution risk, the environmental interference risk, and the equipment recovery mismatch risk.

[0055] In a preferred embodiment, the trajectory deviation evolution risk is used to characterize the degree of deviation of the current recovery control state vector from the pre-takeover automatic operation reference state and the geometric relationship of the target operation zone, as well as the development trend of this deviation within a short time window. The trajectory deviation evolution risk is preferably determined based on the lateral deviation of the target operation zone, the heading deviation of the target operation zone, the deviation change trend, and the local curvature of the target operation zone. The lateral deviation of the target operation zone can be directly determined by the normal distance from the vehicle's current position to the nearest point on the target operation zone; the heading deviation of the target operation zone can be determined by the angle difference between the vehicle's current heading and the tangential direction of the target operation zone; the local curvature of the target operation zone is used to characterize the geometric requirements of the target operation zone for steering convergence during the current recovery control process. The deviation change trend is preferably determined by the direction of change of the lateral deviation and heading deviation of the target operation zone at several consecutive sampling points before and after the manual takeover end time. When the lateral deviation of the target operation zone is within 3 to 8 consecutive sampling points... When the target work zone's heading deviation continuously increases in the same direction, it is determined to be a continuous divergence; when the target work zone's lateral deviation and heading deviation continuously decrease within 3 to 8 consecutive sampling points, it is determined to be a continuous convergence. Preferably, the number of consecutive sampling points is 3 to 5. When the target work zone's lateral deviation increases, the target work zone's heading deviation increases, and the deviation trend shows continuous divergence, or when the target work zone has a large local curvature and the current steering state parameters are insufficient to support rapid regression, the risk of trajectory deviation evolution is increased. Preferably, the target work zone's lateral deviation is normalized according to a preset lateral deviation benchmark, and the target work zone's heading deviation is normalized according to a preset heading deviation benchmark. Then, the trajectory deviation evolution risk is formed by combining the deviation trend. The preset lateral deviation benchmark is preferably 0.10 meters to 1.20 meters, more preferably 0.20 meters to 0.80 meters; the preset heading deviation benchmark is preferably 0.03 radians to 0.35 radians, more preferably 0.05 radians to 0.20 radians.

[0056] Furthermore, environmental disturbance risk is used to characterize the degree of constraint imposed by relative environmental relationships on the current recovery control process. The environmental disturbance risk is preferably determined based on the minimum distance from the vehicle to the edge of the field, the minimum distance from the vehicle to the nearest obstacle, the orientation of the nearest obstacle relative to the vehicle's forward operating direction, and the duration of the disturbance. The intensity of environmental constraints can be determined based on the ratio of the minimum distance from the vehicle to the edge of the field to the preset turning warning distance, the ratio of the minimum distance from the vehicle to the nearest obstacle to the preset safe distance from the obstacle, and the proportion of the disturbance duration to the length of the current analysis window. Preferably, the environmental disturbance risk is divided into three levels: low, medium, and high, according to the intensity of environmental constraints. When the minimum distance from the vehicle to the nearest obstacle and the minimum distance from the vehicle to the edge of the field are both within their respective preset safe thresholds, the environmental disturbance risk is preferably directly upgraded to the high level to reflect the significant limiting effect of the superposition of boundary constraints and obstacle constraints on the recovery control process.

[0057] In a preferred embodiment, the equipment recovery mismatch risk is used to characterize the degree of matching between the current equipment execution state and the equipment state in the pre-takeover automatic operation reference state, as well as the current recovery stage. The equipment recovery mismatch risk is preferably determined jointly based on the current equipment position state, the current operation start / stop state, the equipment position state in the pre-takeover automatic operation reference state, the equipment recovery access label in the scenario constraint label, and the priority control object of the current recovery stage. When there is a large deviation between the current equipment position state and the equipment position state in the pre-takeover automatic operation reference state, or when the current operation start / stop state is inconsistent with the current operation task requirements, the equipment recovery mismatch risk is increased. Furthermore, when the current recovery stage is the attitude convergence stage, obstacle removal stage, or drive stabilization stage, and the equipment execution state has entered the operation recovery action, it is preferred to determine that the equipment action does not match the priority control object of the current recovery stage, thereby further increasing the equipment recovery mismatch risk.

[0058] After determining the risks of trajectory deviation evolution, environmental interference, and equipment recovery mismatch, a recovery risk index is generated based on a preset hierarchical collaborative quantification method. This method maps the three types of risks into a comprehensive risk representation of the current recovery control process. Preferably, the three types of risks are first normalized and classified into levels, and then joint representation processing is performed according to preset weights. The joint representation processing can be implemented by weighted summation followed by level classification based on a threshold, or by segmented mapping or rule-based reasoning. The preferred method is weighted summation followed by level classification based on a threshold. The weights of the three types of risks can be adaptively adjusted according to the takeover type label. For example, in a takeover scenario involving a field turn, the weights of the trajectory deviation evolution risk and the equipment recovery mismatch risk are preferably... The weight of environmental interference risk should be appropriately increased in temporary obstacle avoidance takeover scenarios; the weight of both environmental interference risk and equipment recovery mismatch risk should be increased in vehicle stuck takeover scenarios; and the risk of trajectory deviation evolution should be prioritized in manual fine-tuning takeover scenarios. Preferably, the sum of all risk weights is 1, and each risk weight is preferably between 0.15 and 0.60. The final recovery risk index is preferably between 0 and 1. Furthermore, the recovery risk index can be divided into low-risk, medium-risk, and high-risk levels, where the low-risk level preferably corresponds to a recovery risk index less than 0.30, the medium-risk level preferably corresponds to a recovery risk index between 0.30 and 0.65, and the high-risk level preferably corresponds to a recovery risk index greater than 0.65.

[0059] After completing the quantification of recovery risks, the collaborative reconstruction of recovery targets is performed. Instead of obtaining several independent recovery targets separately, the collaborative reconstruction of recovery targets generates a set of recovery targets based on the takeover type label, scenario constraint label, recovery control state vector, pre-takeover automatic operation reference state, and recovery risk index, under the constraints of the recovery phase sequence. The set of recovery targets includes at least the recovery target speed, recovery target steering angle, recovery target equipment position status, and recovery sequence label.

[0060] In a preferred embodiment, the recovery target speed is used to characterize the target speed that can be pursued under the current recovery phase. Preferably, the recovery target speed is first generated by a weighted transition between the current vehicle speed and the reference speed in the automatic operation reference state before takeover. Then, an upper limit constraint label for the recovery speed is applied to it by the recovery speed upper limit constraint label in the scenario constraint label. Finally, a risk suppression correction is applied to it by the recovery risk index. The reference speed maintenance coefficient in the weighted transition is preferably between 0.20 and 0.85. When the head turn takeover label or the temporary obstacle avoidance takeover label is established, the reference speed maintenance coefficient should preferably take a lower value; when the manual fine-tuning takeover label is established, the reference speed maintenance coefficient can be appropriately increased. The upper limit constraint label for recovery speed can correspond to low-speed recovery level, medium-low-speed recovery level, and normal recovery level. The upper limit of the recovery target speed corresponding to the low-speed recovery level is preferably 0.2 m / s to 1.5 m / s, the upper limit of the recovery target speed corresponding to the medium-low-speed recovery level is preferably 0.8 m / s to 3.0 m / s, and the upper limit of the recovery target speed corresponding to the normal recovery level is preferably 1.5 m / s to 5.0 m / s. When the recovery risk index is at a high-risk level, the recovery target speed is preferably taken from the lower range of the corresponding upper limit of recovery speed. When the recovery risk index is at a low-risk level, the recovery target speed can be closer to the reference speed in the automatic operation reference state before takeover.

[0061] Furthermore, the recovered target steering angle is used to characterize the convergence direction and intensity of the target attitude during the current recovery control process. The recovered target steering angle is preferably determined by the local curvature in the geometric relationship of the target work zone, the lateral and directional deviations of the current vehicle relative to the target work zone, the recovery heading convergence priority label in the scenario constraint labels, and the recovery risk index. Specifically, a basic steering guidance amount can first be formed based on the local curvature of the target work zone, then a deviation correction amount can be formed based on the current lateral and directional deviations, and finally, convergence enhancement correction can be performed based on the recovery heading convergence priority label and the recovery risk index. The lateral deviation feedback coefficient is preferably... The values ​​are selected as 0.10 to 1.50, and the heading deviation feedback coefficient is preferably 0.20 to 2.00. When the heading turn takeover label or temporary obstacle avoidance takeover label is established, and the heading recovery convergence priority label is high priority, it is preferable to increase the lateral deviation feedback coefficient and the heading deviation feedback coefficient. When the manual fine-tuning takeover label is established, it is preferable to decrease the above feedback coefficients. Furthermore, when the heading recovery convergence priority label is high priority and the recovery risk index is at a medium or high risk level, a convergence enhancement correction should be performed on the recovery target turning angle to enhance its convergence effect on the lateral deviation and heading deviation of the target operation zone.

[0062] In a preferred embodiment, restoring the target machine's position status is used to characterize whether the machine's execution system should maintain a safe position status, an operational position status, or maintain its current position status during the current recovery phase. The restored target machine's position status is preferably determined jointly by the current machine's position status, the machine's position status in the pre-takeover automatic operation reference status, the machine recovery access label in the scenario constraint label, and the recovery risk index. When the recovery risk index is higher than a preset machine recovery risk threshold, or the environmental relative relationship indicates that the minimum distance from the vehicle to the nearest obstacle is still less than a preset obstacle safety distance, or the current recovery phase has not yet met the machine recovery access conditions, it is preferable to restore the target machine's position status. The target machine position is set to a safe position state. When the recovery risk index is lower than the machine recovery risk threshold, and the lateral and directional deviations of the current vehicle relative to the target work zone have converged to the preset allowable range, and the access conditions corresponding to the machine recovery access label are met, it is preferable to set the recovery target machine position state to the work position state. In the intermediate case, that is, when the recovery risk index is at the medium risk level, and the current lateral or directional deviation has not yet reached the machine recovery access conditions, but has not reached the level where it must be immediately maintained as a safe position state, it is preferable to set the recovery target machine position state to maintain the current position state. By setting this intermediate state, the machine can be avoided from repeatedly switching in the boundary state.

[0063] After obtaining the target recovery speed, target recovery steering angle, and target recovery equipment position status, a recovery sequence label is further generated. The recovery sequence label is used to clarify the sequential relationship between steering convergence, speed recovery, and equipment recovery in the current takeover scenario. The recovery sequence label is preferably initially generated directly from the takeover type label, and then the sequence is rearranged and corrected according to the recovery risk index. Preferably, the recovery sequence label for a ground-turn takeover is set to attitude convergence first, then speed recovery, and then equipment recovery; the recovery sequence label for a temporary obstacle avoidance takeover is set to obstacle removal first, then attitude return, then speed recovery, and then equipment recovery; the recovery sequence label for a vehicle stuck and escaping takeover is set to drive stabilization first, then attitude correction, and then equipment recovery; the recovery sequence label for a manual fine-tuning takeover is set to attitude fine-tuning and speed recovery be promoted in tandem, and equipment recovery is performed only after the equipment recovery access conditions are met. When the recovery risk index is at a high risk level, it is preferable to perform a sequence rearrangement correction, prioritizing safety maintenance and attitude convergence, and delaying speed recovery and equipment recovery; when the recovery risk index is at a low risk level, the initial recovery sequence label is maintained or only a slight correction is performed.

[0064] In a preferred embodiment, based on the recovery risk index, an adaptive risk level constraint correction process is also performed. This process is triggered by the level of the recovery risk index. Low risk level corresponds to slight correction or no correction; medium risk level corresponds to amplitude suppression correction and access delay correction; and high risk level corresponds to significant amplitude suppression correction, convergence enhancement correction, and order rearrangement correction. Specifically, for the recovery target speed, risk suppression correction is performed so that the higher the recovery risk index, the more conservative the recovery target speed becomes. For the recovery target steering angle, convergence enhancement correction is performed so that when the recovery risk index is higher and the recovery heading convergence priority is higher, the recovery target steering angle is more biased towards prioritizing attitude convergence. For the recovery target equipment position state, access delay correction is performed so that when the recovery risk index is higher, the timing of the equipment entering the working position state from the safe position state or maintaining the current position state is later. For the recovery sequence label, order rearrangement correction is performed so that when the recovery risk index is higher, safety maintenance and attitude convergence take priority.

[0065] Through the above processing, step S3 finally outputs the recovery target set and the recovery risk index. The recovery target set includes at least the recovery target speed, the recovery target steering angle, the recovery target equipment position status, and the recovery sequence label. The recovery risk index is used to characterize the comprehensive risk level in the current recovery control process. The recovery risk index does not exist independently, but rather acts inversely on the recovery target collaborative reconstruction processing and the risk level adaptive constraint correction processing. The recovery sequence label is also not an independent label, but rather constitutes a mutually coupled recovery target combination together with the recovery target speed, the recovery target steering angle, and the recovery target equipment position status.

[0066] By performing recovery risk quantification and recovery target collaborative reconstruction, a set of recovery targets and a recovery risk index that are adapted to the current takeover scenario, current recovery status, and current risk boundary are generated, providing a direct control basis with phase sequence and risk constraints for step S4.

[0067] Step S4 addresses the issues of abrupt changes in control output, mismatch in phase progression, and lack of closed-loop correction after recovery failure caused by directly entering the execution layer after the recovery target is generated. In existing technologies, after generating the recovery target, the recovery target is usually directly converted into actuator control commands, or only simple amplitude and speed limiting are superimposed on the outside of the recovery target. This fails to handle the phase dependencies between steering control, longitudinal control, and machine control, and also fails to adjust the control progression strategy in real time based on stability judgment results and safety risk conditions during the recovery control process. As a result, problems such as abrupt changes in recovery control output, recovery sequence mismatch, improper timing of machine recovery, and continued progression according to the original control strategy after recovery failure are prone to occur. To solve the above problems, this step executes a preset phased mitigation and coordination control based on the recovery target set and recovery risk index output in step S3 and the scenario constraint label output in step S2, outputs the recovery control quantity, and updates the recovery control process in a closed loop based on the stability judgment result obtained from the preset stability judgment rules. This enables the recovery control process to simultaneously possess phase progression capability, risk convergence capability, and anomaly protection capability.

[0068] Specifically, step S4 includes: based on the recovery target set, recovery risk index, and scenario constraint labels, executing a preset phased mitigation coordination control, outputting recovery control quantities, and updating the recovery control process in a closed loop based on the stability judgment result obtained from the preset stability judgment rules. The preset phased mitigation coordination control includes: determining recovery control quantities based on the recovery target set, including steering control quantities, longitudinal control quantities, and equipment control quantities; performing phased enable allocation and mitigation output adjustment on the steering control quantities, longitudinal control quantities, and equipment control quantities based on the recovery sequence labels and scenario constraint labels; triggering the subsequent stage control objects to enter the effective recovery stage when the controlled objects in the preceding stage meet the preset transition stability conditions, thereby forming a sequentially controlled, amplitude-limited, and mutually cooperating recovery control chain of steering control quantities, longitudinal control quantities, and equipment control quantities, and obtaining the stability judgment result based on the preset stability judgment rules. The closed-loop update and recovery control process for the stability determination result includes: based on preset stability determination rules, performing joint determination on the trajectory convergence state, velocity convergence state, and equipment readiness state of the current recovery control process to obtain the stability determination result; when the stability determination result meets the preset recovery completion conditions, switching to the regular automatic operation control mode; when the stability determination result does not meet the preset recovery completion conditions and the preset safety risk conditions are not triggered, re-executing the preset takeover scenario attribution processing, preset recovery target collaborative reconstruction processing, preset recovery risk quantification processing, and preset phased mitigation coordination control based on the updated recovery control state vector; when the preset safety risk conditions are triggered, executing the preset safety maintenance control or manual maintenance control, where the recovery control quantity is the core output of step S4, and the stability determination result is the associated output used to update the recovery control process. When any key state quantity in the recovery control state vector is marked as invalid, or the effective confidence of the takeover type label is lower than the preset type confirmation threshold, or the input risk item of the recovery risk index lacks any necessary input, the current recovery control result is not directly issued as an effective recovery control quantity, but instead enters the safety maintenance control or manual maintenance control. Safety-maintaining controls preferably include maintaining low speed, keeping the machine in a non-operating position, limiting the rate of change of steering angle, and prohibiting the machine from re-entering the operating state. This process prevents low-confidence recovery target sets or low-confidence recovery risk indices from continuing to trigger automatic recovery controls.

[0069] In a preferred embodiment, step S4 uses the recovery target set output in step S3 as the target layer input, the recovery risk index output in step S3 as the risk adjustment input, and the scene constraint label output in step S2 as the stage boundary input. The recovery target set includes at least the recovery target speed, the recovery target steering angle, the recovery target equipment position status, and the recovery sequence label. The recovery target speed corresponds to the target boundary of the longitudinal control quantity, the recovery target steering angle corresponds to the target boundary of the steering control quantity, the recovery target equipment position status corresponds to the target boundary of the equipment control quantity, the recovery sequence label is used to determine the order in which various recovery control quantities enter the effective recovery stage, the recovery risk index is used to adjust the intensity and rate of various recovery control quantities advancing from the current state to the recovery target set, and the scene constraint label is used to limit the executable boundary and stage advancement boundary of the recovery control quantities.

[0070] In this embodiment, the recovery control quantity is first determined based on the recovery target set. Specifically, the steering control quantity is preferably formed by the deviation between the recovery target steering angle and the current steering state parameter to form the target steering adjustment quantity; the longitudinal control quantity is preferably formed by the deviation between the recovery target speed and the current vehicle speed to form the target drive adjustment quantity or the target braking adjustment quantity; the implement control quantity is preferably formed by the deviation between the recovery target implement position state and the current implement position state to form the target implement execution adjustment quantity. The above target adjustment quantities can be implemented by calling the actuator closed-loop controller commonly used in the art. This part belongs to the supporting prior art. The innovation of this embodiment is not in the underlying actuator controller structure, but in the fact that after the recovery control quantity is formed, it still needs to be subject to the common constraints of stage enable allocation and slow-release output adjustment.

[0071] In a preferred embodiment, the phased release coordinated control first performs phase enable allocation. Phase enable allocation refers to determining, based on the recovery sequence label and scenario constraint label, whether various recovery control quantities are allowed to advance towards their corresponding target boundaries according to the release output adjustment rules within the current control cycle. Entering the effective recovery phase means that the corresponding recovery control quantity is allowed to continuously approach the recovery target set based on the current effective release coefficient. The holding state means that the corresponding recovery control quantity maintains the output value of the previous control cycle or maintains a safe position without advancing. The restricted advancement state means that the corresponding recovery control quantity is allowed to make small adjustments within a preset restricted boundary, but is not allowed to advance at the normal recovery rate. For example, in a scenario where the recovery sequence label is first attitude convergence, then velocity recovery, and then equipment recovery, it is preferable to enable only the steering control quantity to enter the effective recovery phase in the initial stage, while the longitudinal control quantity... Only the low-speed restricted propulsion state is maintained, and the implement control quantity remains in a safe position or maintains its current position. When the attitude convergence target corresponding to the steering control quantity reaches the preset transition stability condition, the longitudinal control quantity is then enabled to enter the effective recovery stage. When the longitudinal control quantity further reaches the corresponding preset transition stability condition and the condition corresponding to the implement recovery access label is met, the implement control quantity is then enabled to enter the effective recovery stage. For scenarios where the recovery sequence label is obstacle removal, then attitude return, then speed recovery, and finally implement recovery, it is preferable that in the obstacle removal stage, only the steering control quantity and the necessary low-speed longitudinal control quantity are allowed to be in a restricted propulsion state, and the implement control quantity remains in a safe position. When the obstacle constraint is removed, attitude return, speed recovery, and implement recovery are then enabled in sequence. Through stage enable allocation, the problem of mutual interference caused by the simultaneous release of various recovery control quantities in the prior art can be avoided.

[0072] Furthermore, the phase enable allocation is controlled not only by the recovery sequence label but also by the scenario constraint label. Specifically, the recovery speed upper limit constraint label is used to limit the maximum advance amplitude of the longitudinal control quantity after entering the effective recovery phase; the recovery heading convergence priority label is used to limit the advance priority of the steering control quantity in the current control cycle; the implement recovery access label is used to limit whether the implement control quantity is allowed to enter the effective recovery phase from the holding state or the restricted advance state; the recovery sequence constraint label is used to specify the legal advance sequence between each phase; and the safety holding constraint label is used to apply a mandatory holding boundary to at least one recovery control quantity in high-risk scenarios. For example, when the safety holding constraint label is in the active state, the longitudinal control quantity is preferably restricted to the low-speed holding range, the implement control quantity is preferably forced to be held in a safe position state, and the steering control quantity is preferentially used to control the attitude convergence of the vehicle relative to the target work zone.

[0073] After determining the stage states of various recovery control quantities in the current control cycle, a slow-release output adjustment is further performed. This adjustment limits the rate at which the recovery control quantity progresses from the current state to the target state, thus preventing abrupt changes in the recovery control output. Preferably, basic slow-release coefficients are set for steering control, longitudinal control, and implement control quantities. Then, risk correction is applied to these basic slow-release coefficients based on the recovery risk index to obtain the effective slow-release coefficient for the current control cycle. The preferred values ​​for the basic slow-release coefficients are 0.05 to 0.80, with the basic slow-release coefficient for steering control preferably between 0.10 and 0.70, the basic slow-release coefficient for longitudinal control preferably between 0.05 and 0.60, and the basic slow-release coefficient for implement control preferably between 0.05 and 0.50. When the recovery risk index increases, the effective slow-release coefficient is preferably decreased; when the recovery risk index decreases, the effective slow-release coefficient can be gradually increased. Furthermore, when the recovery risk index is at a high-risk level, it is preferable to adjust the longitudinal control... The quantity and equipment control quantity are subject to a stronger mitigation correction than the steering control quantity to reflect the control principle that attitude convergence takes precedence over speed recovery and equipment recovery. Preferably, the mitigation coefficient correction range for low-risk levels is 0 to 20% of the base mitigation coefficient, for medium-risk levels it is 20% to 50% of the base mitigation coefficient, and for high-risk levels it is 50% to 80% of the base mitigation coefficient. In a preferred embodiment, the mitigation output adjustment can be performed according to the following logic: In the current control cycle, the output value of a certain recovery control quantity is obtained by adding "a portion of the difference between the target value of the recovery control quantity and the output value of the previous control cycle" to the output value of the previous control cycle; this portion of the difference is determined by the current effective mitigation coefficient. If a certain recovery control quantity is in a hold state, it is preferable to keep the output value of the previous control cycle unchanged; if a certain recovery control quantity is in a restricted propulsion state, it is preferable to only allow it to propulse within a preset restricted boundary.

[0074] To ensure clear switching criteria for each stage of the process, this embodiment further sets preset transition stabilization conditions. These conditions are used to determine whether the controlled object in the preceding stage has met the requirements for triggering the controlled object in the subsequent stage to enter the effective recovery stage. Preferably, the preset transition stabilization conditions include at least one of the following: the lateral deviation of the target work zone entering a preset lateral deviation allowable range; the heading deviation of the target work zone entering a preset heading deviation allowable range; the speed deviation between the current vehicle speed and the recovery target speed entering a preset speed deviation allowable range; and the state deviation between the current equipment position state and the recovery target equipment position state entering a preset equipment deviation allowable range. The preset lateral deviation allowable range is preferably 0.05 meters to 0.50 meters, the preset heading deviation allowable range is preferably 0.02 radians to 0.15 radians, and the preset speed deviation... The allowable range is preferably 0.05 m / s to 0.60 m / s, and the preset allowable range for machine deviation is preferably 0.02 to 0.20 for the normalized difference of machine position status. Further, it is preferred that the corresponding deviation amount remains within the allowable range for 2 to 10 consecutive control cycles, more preferably for 3 to 5 consecutive control cycles. For ease of unified description, it can be stipulated that when the main deviation amount corresponding to the controlled object in the preceding stage enters the preset allowable range and continues to meet the requirements for several control cycles, the preceding stage is determined to have reached the preset transition stability condition. Here, the main deviation amount refers to the deviation amount directly corresponding to the dominant task of the current stage. For example, in the attitude convergence stage, the main deviation amount is the lateral deviation of the target work zone and the heading deviation of the target work zone; in the speed recovery stage, the main deviation amount is the speed deviation; in the machine recovery stage, the main deviation amount is the machine position status deviation.

[0075] After the controlled object in the preceding stage meets the preset transition stability conditions, the controlled object in the subsequent stage is triggered to enter the effective recovery stage. This allows the steering control quantity, longitudinal control quantity, and implement control quantity to form a sequentially controlled, amplitude-limited, and mutually coordinated recovery control chain. Taking the turn-over scenario at the field as an example, the steering control quantity is preferred to first drive the vehicle's attitude convergence relative to the target work area. When the lateral deviation and directional deviation of the target work area simultaneously enter the preset allowable range and remain so for several control cycles, the longitudinal control quantity is then triggered to enter the effective recovery stage, causing the vehicle speed to gradually approach the recovery target speed. When the longitudinal control quantity further advances... Once the speed deviation falls within the preset allowable range, the recovery risk index is lower than the preset machine recovery risk threshold, and the conditions corresponding to the machine recovery access label are met, the machine control quantity is triggered to enter the effective recovery phase. Taking a temporary obstacle avoidance takeover scenario as an example, it is preferable that when the minimum distance between the vehicle and the nearest obstacle is still less than the preset obstacle safety distance, only the steering control quantity and the necessary low-speed longitudinal control quantity are allowed to work within the restricted boundary; when the minimum distance between the vehicle and the nearest obstacle is again greater than the preset obstacle safety distance and remains there, attitude return and speed recovery are then promoted; finally, the machine control quantity is allowed to recover after the recovery risk index is further reduced.

[0076] After the recovery control output of the current control cycle, a stability judgment result is obtained based on the preset stability judgment rules. The preset stability judgment rules are used to determine whether the current recovery control process has been completed, whether it needs to continue, or whether it should switch to safe holding control or manual holding control. Specifically, the preset stability judgment rules include at least trajectory convergence state judgment, speed convergence state judgment, and equipment readiness state judgment. The trajectory convergence state judgment is used to determine whether the lateral deviation and heading deviation of the current vehicle relative to the target work zone have entered the preset allowable range and remained stable. The speed convergence state judgment is used to determine whether the speed deviation between the current vehicle speed and the recovery target speed has entered the preset allowable range and remained stable. The equipment readiness state judgment is used to determine whether the current equipment position status, work start / stop status, and equipment recovery access label corresponding conditions meet the requirements. It should be noted that when the current recovery control process has not yet entered the equipment recovery stage, the equipment readiness state judgment is preferably not used as a mandatory item for the recovery completion condition, but is used as a mandatory judgment item when the equipment recovery stage has been enabled or is close to being enabled.

[0077] In a preferred embodiment, when the trajectory convergence state, speed convergence state, and equipment readiness state meet preset recovery completion conditions, the recovery control process is determined to be complete, and the system switches to a conventional automatic operation control mode. The preset recovery completion conditions preferably include: the lateral deviation of the target work zone is less than a preset lateral deviation completion threshold; the heading deviation of the target work zone is less than a preset heading deviation completion threshold; the deviation between the vehicle speed and the recovery target speed is less than a preset speed completion threshold; and the deviation between the equipment position state and the recovery target equipment position state is less than a preset equipment completion threshold. Furthermore, the above conditions are maintained for a preset completion holding time. The preset lateral deviation completion threshold is preferably 0.03 meters to 0.30 meters; the preset heading deviation completion threshold is preferably 0.01 radians to 0.10 radians; the preset speed completion threshold is preferably 0.03 meters per second to 0.40 meters per second; the preset equipment completion threshold is preferably a normalized difference in equipment position state of 0.01 to 0.10; and the preset completion holding time is preferably 0.10 seconds to 2.00 seconds, more preferably 0.20 seconds to 1.00 seconds.

[0078] When the stability determination result does not meet the preset recovery completion conditions and the preset safety risk conditions are not triggered, it is preferable to perform a closed-loop update on the recovery control process. The closed-loop update can be triggered at each control cycle or at a preset refresh cycle. Among them, when the change of the recovery risk index exceeds the preset risk refresh threshold, it is preferable to trigger the closed-loop update immediately. The preset risk refresh threshold is preferably 0.05 to 0.20. The specific content of the closed-loop update includes: re-executing the preset takeover scenario attribution processing, the preset recovery target collaborative reconstruction processing, the preset recovery risk quantification processing, and the preset phased mitigation coordination control based on the updated recovery control state vector. By updating the recovery control process through closed-loop updates, the problem of the initial recovery strategy remaining unchanged in the prior art, but still proceeding according to the old strategy when external constraints and vehicle status have changed, can be avoided.

[0079] In a preferred embodiment, if a preset safety risk condition is triggered, a preset safety holding control or manual holding control is executed. The preset safety risk condition preferably includes at least the following: the minimum distance from the vehicle to the nearest obstacle is less than a preset emergency safety distance; the wheel slip state re-enters a significant wheel slip state and continues for several control cycles; the lateral deviation or directional deviation of the target work zone continues to diverge within a short time window; the steering control quantity or longitudinal control quantity is at the saturation boundary for several consecutive control cycles; and at least one of the following: after the recovery control quantity output, the deviation between the current recovery control state vector and the recovery target set continues to increase or continues to exceed a preset deviation upper limit for several consecutive control cycles. The preset emergency safety distance is preferably 0.30 meters to 2.00 meters. The safety holding control may include maintaining a low speed or zero speed state, maintaining the equipment in a safe position state, maintaining the current stable attitude, and prohibiting entry into a new recovery phase. The manual holding control may include pausing the automatic recovery control process and maintaining the manual takeover mode in effect.

[0080] Through the above processing, step S4 achieves the orderly transformation of the recovery target set into recovery control quantities under conditions of controlled risk and controlled stages, and completes the closed-loop update and anomaly protection of the recovery control process through stability judgment results and safety risk conditions. By using a phased release coordination control and stability judgment closed-loop update mechanism, the recovery target set is transformed into recovery control quantities under conditions of controlled risk and controlled stages, thereby improving the stability and safety of the recovery control process after manual takeover.

[0081] Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An artificial intelligence-based collaborative optimization control method for agricultural machinery efficiency, characterized in that, Includes the following steps: Acquire multi-source operational status data before and after manual takeover, perform recovery control status restoration processing based on multi-source operational status data, and generate recovery control status vector, automatic operation reference status before takeover, geometric relationship of target operation zone and relative relationship of environment; Based on the recovery control state vector, the geometric relationship of the target work zone, and the relative relationship of the environment, a preset takeover scenario attribution process is performed to generate takeover type labels and scenario constraint labels. Based on the takeover type label, scenario constraint label, recovery control state vector, and pre-takeover automatic operation reference state, the preset recovery target collaborative reconstruction processing and preset recovery risk quantification processing are performed to generate a recovery target set and a recovery risk index. Based on the recovery target set, recovery risk index, and scenario constraint labels, a pre-defined phased mitigation coordination control is executed, the recovery control quantity is output, and the recovery control process is updated in a closed loop based on the stability judgment result obtained from the pre-defined stability judgment rules.

2. The agricultural machinery efficiency collaborative optimization control method based on artificial intelligence according to claim 1, characterized in that, Multi-source operational status data includes vehicle motion status data, vehicle posture status data, equipment execution status data, target work zone geometry data, and environmental status data; Vehicle motion state data and vehicle attitude state data are used to generate recovery control state vectors; implement execution state data, vehicle motion state data, and vehicle attitude state data are used to generate automatic operation reference states before takeover; target operation zone geometry data are used to generate target operation zone geometry relationships; and environmental state data are used to generate environmental relative relationships. Multi-source operational status data is also used to characterize the continuous change process of control before and after manual takeover, so as to provide a cross-time period status correlation basis for the recovery of control status.

3. The agricultural machinery efficiency collaborative optimization control method based on artificial intelligence according to claim 1, characterized in that, The recovery process for restoring control status includes: Based on the preset state homogeneity constraint, cross-source correspondence sorting and time-series mapping processing are performed on multi-source operation state data to obtain candidate associated state data; Based on preset control continuity constraints, abnormal fragment removal and reference state backtracking are performed on candidate associated state data to obtain continuous state fragments. Based on preset state credibility fusion rules, consistency verification and restoration fusion processing are performed on continuous state segments to generate recovery control state vectors, automatic operation reference states before takeover, geometric relationships of target operation zones, and relative environmental relationships.

4. The agricultural machinery efficiency collaborative optimization control method based on artificial intelligence according to claim 1, characterized in that, The pre-defined attribution processing for takeover scenarios includes: Based on the recovery control state vector, the geometric relationship of the target work zone, and the relative relationship with the environment, the evolution characteristics of takeover behavior, the deviation characteristics of the target work zone, the environmental disturbance characteristics, and the linkage characteristics of machinery are extracted. Based on the evolutionary characteristics of takeover behavior, deviation characteristics of target work zone, environmental interference characteristics, and equipment linkage characteristics, a pre-trained scenario attribution model is used for inference processing. Based on the preset scenario consistency verification rules, constraint correction is performed on the inference results of the scenario attribution model to generate takeover type labels and scenario constraint labels.

5. The artificial intelligence-based collaborative optimization control method for agricultural machinery efficiency according to claim 4, characterized in that, The type of takeover label includes at least one of the following: turn-around takeover label, temporary obstacle avoidance takeover label, vehicle stuck out of trouble takeover label, and manual fine-tuning takeover label; The scenario constraint labels include at least one of the following: recovery speed upper limit constraint label, recovery heading convergence priority label, equipment recovery access label, recovery order constraint label, and safety maintenance constraint label; Scene constraint labels are used to define the target generation boundaries in the collaborative reconstruction process of recovery targets and the stage execution boundaries in the phased release coordination control.

6. The artificial intelligence-based collaborative optimization control method for agricultural machinery efficiency according to claim 1, characterized in that, The preset recovery target collaborative reconstruction process includes: Based on the takeover type label, scenario constraint label, recovery control state vector, and pre-takeover automatic operation reference state, a recovery phase division process is performed to obtain a recovery phase sequence corresponding to the current takeover scenario. Based on the recovery phase sequence, recovery target association mapping processing and target consistency correction processing are performed to generate a recovery target set; The recovery target set includes at least the recovery target speed, the recovery target steering angle, the recovery target equipment position status, and the recovery sequence label, and the recovery target speed, the recovery target steering angle, and the recovery target equipment position status constitute a mutually coupled recovery target combination under the constraint of the recovery sequence label.

7. The agricultural machinery efficiency collaborative optimization control method based on artificial intelligence according to claim 1, characterized in that, The preset recovery risk quantification process includes: Based on the recovery control state vector, environmental relative relationship and automatic operation reference state before takeover, determine the trajectory deviation evolution risk, environmental interference risk and equipment recovery mismatch risk; Based on a pre-defined hierarchical collaborative quantification method, joint characterization processing is performed on trajectory deviation evolution risk, environmental interference risk, and equipment recovery mismatch risk to generate a recovery risk index.

8. The agricultural machinery efficiency collaborative optimization control method based on artificial intelligence according to claim 6, characterized in that, The pre-defined recovery target collaborative reconstruction process also includes: Based on the recovery risk index, adaptive constraint correction processing of risk level is performed, and convergence enhancement correction, amplitude suppression correction, stage delay correction or safety maintenance correction are applied to at least one recovery objective in the recovery objective set. The risk level adaptive constraint correction process includes at least one of the following: performing risk suppression correction on the recovery target speed, performing convergence enhancement correction on the recovery target steering angle, performing access delay correction on the recovery target equipment position status, and performing order rearrangement correction on the recovery order label.

9. The artificial intelligence-based collaborative optimization control method for agricultural machinery efficiency according to claim 1, characterized in that, The pre-defined phased slow-release coordination control includes: Based on the recovery target set, the recovery control variables are determined, including steering control variables, longitudinal control variables, and tool control variables. Based on the recovery sequence label and the scenario constraint label, enable allocation and slow-release output adjustment are performed on the steering control quantity, longitudinal control quantity and tool control quantity during the execution phase. When the controlled object in the preceding stage meets the preset transition stability conditions, the controlled object in the subsequent stage is triggered to enter the effective recovery stage, thereby forming a recovery control chain in which the steering control quantity, longitudinal control quantity, and machine control quantity are sequentially controlled, amplitude-limited, and mutually coordinated.

10. The artificial intelligence-based collaborative optimization control method for agricultural machinery efficiency according to claim 9, characterized in that, The closed-loop update and recovery control process based on the stability determination result obtained from the preset stability determination rules includes: Based on the preset stability determination rules, a joint determination is performed on the trajectory convergence state, velocity convergence state, and equipment readiness state of the current recovery control process to obtain the stability determination result. When the stability assessment result meets the preset recovery completion conditions, switch to the normal automatic operation control mode; When the stability determination result does not meet the preset recovery completion conditions and the preset security risk conditions are not triggered, the preset takeover scenario attribution processing, preset recovery target collaborative reconstruction processing, preset recovery risk quantification processing and preset phased mitigation coordination control are re-executed based on the updated recovery control state vector. When a preset security risk condition is triggered, the preset security hold control or manual hold control is executed.

Images