Unmanned tractor adaptive path tracking method based on multi-sensor fusion

Abstract

The invention discloses an unmanned tractor adaptive path tracking method based on multi-sensor fusion, which relates to the technical field of unmanned tractor path tracking, and comprises the following steps: carrying out environment and state perception of multi-sensor fusion; constructing a dynamic control window and constraints; adaptive model predictive control calculation is executed; executing the control instruction and performing closed-loop feedback; outputting the control instruction to a steer-by-wire system to realize path tracking; repeating the steps until all the operation paths are completed; according to the self-adaptive path tracking method for the unmanned tractor based on multi-sensor fusion, an enhanced environment map is generated on line, so that the change of a front road surface can be analyzed in advance; by dynamically constructing the preview window and the stability constraint which are tightly coupled with the road adhesion condition, the path deviation can be optimized, and the road adhesion condition can be actively brought into the tradeoff and constraint of a control target, so that a control instruction is automatically softened in soft, wet and slippery low adhesion areas and the like, and the slip risk is inhibited.

Description

Technical Field

[0001] This invention relates to the field of path tracking technology for unmanned tractors, and more specifically to an adaptive path tracking method for unmanned tractors based on multi-sensor fusion. Background Technology

[0002] Automatic path tracking technology for driverless tractors is a core component of precision agriculture. Its core task is to control the tractor to travel stably and accurately along a pre-set work path (such as the centerline of crop rows). Current mainstream high-precision agricultural vehicle automatic navigation systems generally rely on a combination of Global Navigation Satellite System (GNSS) and Inertial Measurement Unit (IMU) for positioning, and employ control algorithms based on kinematic or dynamic models (such as PurePursuit, Stanley, or model predictive control) to generate steering commands. Such systems can achieve satisfactory tracking accuracy on hard surfaces with uniform and good road conditions.

[0003] However, farmland operating environments are significantly complex and non-uniform, with the most prominent factor often overlooked by existing technologies being the spatial variability of soil mechanical properties. Soil moisture content and compaction can vary drastically across different plots and even within the same row, directly leading to dynamic fluctuations in the coefficient of adhesion (i.e., grip) between the tires and the ground. Most existing path tracking methods treat key parameters in the vehicle dynamics model (such as tire lateral stiffness and peak adhesion coefficient) as fixed values, and their controller designs are primarily based on geometric path deviations, failing to incorporate the real-time adhesion capability of the road surface—a fundamental factor directly affecting the vehicle's lateral dynamics and stability—into the closed-loop control system. This "blindness" causes the system to generate overly aggressive control commands based on a fixed model when suddenly transitioning from a high-adhesion zone to a low-adhesion zone, such as from dry, hard soil to wet, soft mud. This can easily lead to excessive wheel slippage or even vehicle instability, severely limiting the robustness, safety, and operational efficiency of unmanned tractors in real, complex farmland conditions. Summary of the Invention

[0004] The purpose of this invention is to provide an adaptive path tracking method for unmanned tractors based on multi-sensor fusion, in order to solve the problem in the prior art where "blindness" causes the system to suddenly enter a low-adhesion zone from a high-adhesion zone, such as from dry hard soil to wet soft mud, and the control commands generated based on a fixed model are easily too aggressive, causing excessive wheel slippage or even vehicle instability.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an adaptive path tracking method for unmanned tractors based on multi-sensor fusion, comprising the following steps:

[0006] S1. Environmental and state perception through multi-sensor fusion: Simultaneously acquire and process data from different sensors to generate an enhanced environmental map containing soil drivability information and geometric path information;

[0007] S2. Constructing dynamic control window and constraints: Based on the distribution of soil adhesion coefficient in the enhanced environmental map and the current motion state of the tractor, determine the aiming distance and vehicle stability constraints of the model predictive control;

[0008] S3. Perform adaptive model predictive control calculation: Based on the output of steps S1 and S2, construct and solve online a control problem with path tracking accuracy, control smoothness and driving stability as comprehensive optimization objectives, and obtain the optimal front wheel steering angle control command;

[0009] S4. Execute control commands and provide closed-loop feedback: Send the optimal front wheel steering angle control command obtained in step S3 to the steer-by-wire system for execution, collect new state and environmental perception data, return to step S1, and form closed-loop control; output the first element of the optimal front wheel steering angle control sequence obtained in step S32 as a control command to the tractor's steer-by-wire system to achieve path tracking.

[0010] S5. Repeat steps S1 to S4 until all job paths are completed.

[0011] Furthermore, step S1 specifically includes the following sub-steps:

[0012] S11. Synchronous data acquisition: Acquire pose and heading data provided by the GNSS / INS integrated navigation system, environmental point cloud and image data obtained by the visual / LiDAR sensor, and real-time soil parameters obtained by the soil physics sensor.

[0013] S12. Generate an environmental semantic layer: Process the environmental point cloud and image data, identify the travel area, crop row center line and obstacle boundary, and generate a basic environmental semantic layer.

[0014] S13. Generate a soil adhesion coefficient distribution layer: Map the real-time soil parameters to soil adhesion coefficients, and generate a spatially continuous soil adhesion coefficient distribution layer based on their geographic coordinates.

[0015] S14. Generate an enhanced environmental map: Spatiotemporally fuse the basic environmental semantic layer with the soil adhesion coefficient distribution layer to form the enhanced environmental map.

[0016] Furthermore, in step S13, the soil parameters include at least soil moisture content and compaction, which are mapped to a dimensionless soil adhesion coefficient used to characterize tire-ground adhesion ability through a preset soil mechanics model.

[0017] Furthermore, step S2 specifically includes the following sub-steps:

[0018] S21. Dynamically determine the aiming distance: Based on the tractor's current longitudinal speed, path curvature, and the soil adhesion coefficient queried from the enhanced environmental map at the aiming point, adaptively adjust the aiming distance of the model predictive control; increase the aiming distance in low adhesion areas and decrease the aiming distance in high adhesion areas.

[0019] S22. Dynamically determine stability constraints: Based on the soil adhesion coefficient at the current and predicted trajectory points, dynamically calculate and set the corresponding lateral acceleration threshold and yaw rate threshold as time-varying state constraints in model predictive control.

[0020] Furthermore, step S3 specifically includes the following sub-steps:

[0021] S31. Construct a parameterized vehicle prediction model: Establish an extended prediction model that includes tractor lateral-yaw dynamics, steering system dynamics, and tire force model; wherein, the tire force model is a nonlinear model based on Pacejka's magic formula.

[0022] S32. Online update of prediction model parameters: Based on the enhanced environmental map output in step S1, obtain the soil adhesion coefficient corresponding to the current and predicted trajectory points in real time, and update the peak road adhesion coefficient parameter in the Pacejka magic formula online.

[0023] S33. Construct and solve the rolling optimization problem: Based on the parameterized prediction model updated in step S32, with the dynamic aiming distance determined in step S2 as the prediction time domain, the slip risk weighted by path tracking deviation, control variable change rate and adhesion coefficient as the optimization objective, and the dynamic stability constraint determined in step S2 as the boundary, perform rolling optimization solution and output the optimal front wheel steering angle control sequence.

[0024] Furthermore, in step S33, the weight of the slip risk penalty term after the adhesion coefficient is negatively correlated with the soil adhesion coefficient of the predicted point, which makes the controller more inclined to generate smooth and conservative control commands in low adhesion areas to suppress slip.

[0025] Furthermore, in step S4, before sending the control command to the steer-by-wire system, the control command is subjected to bandwidth adaptive low-pass filtering and amplitude limiting processing based on the current soil adhesion coefficient.

[0026] Furthermore, step S4 specifically includes the following sub-steps:

[0027] S41. Control command post-processing: Based on the soil adhesion coefficient at the current tractor position obtained from the enhanced environmental map, the optimal front wheel steering angle control command is subjected to adaptive low-pass filtering and rate limiting processing to generate a smooth and safe execution command.

[0028] S42. Instruction Issuance and Execution: The execution instruction is sent to the tractor's steering-by-wire system for execution;

[0029] S43. Status update and cycle switching: Collect the execution feedback data of the steer-by-wire system and the latest vehicle status data, update the system status, and trigger the start of the next control cycle.

[0030] Furthermore, in step S41, the adaptive low-pass filtering and rate limiting processing specifically includes the following steps:

[0031] S411, Parameter Mapping: Based on the current soil adhesion coefficient, query a preset mapping table or use a predefined calculation function to determine the corresponding low-pass filter cutoff frequency and steering angle rate limiting threshold; the mapping relationship or function is configured such that the lower the soil adhesion coefficient, the smaller the values ​​of the cutoff frequency and the limiting threshold.

[0032] S412. Filtering execution: Using the cutoff frequency determined in step S411, construct a first-order or second-order digital low-pass filter to perform real-time filtering calculations on the optimal front wheel steering angle control command to obtain a preliminary smoothing command.

[0033] S413, Limiting Execution: Calculate the rate of change of the preliminary smoothing instruction obtained in step S412. If its absolute value exceeds the limiting threshold determined in step S411, limit the rate of change with the threshold and generate the final execution instruction accordingly.

[0034] Furthermore, in step S43, the state update and cycle switching specifically include the following steps:

[0035] S431, Data Synchronization Acquisition: At the end of the control cycle, the actual steering angle feedback from the steer-by-wire system and the latest attitude and velocity data from the GNSS / INS integrated navigation system are synchronously acquired, and all acquired data are assigned a unified timestamp.

[0036] S432. Calculation of actual state increment: Based on the pose data collected at the beginning and end of the previous control cycle, calculate the actual lateral displacement increment and heading angle increment of the tractor in this cycle; at the same time, based on the actual steering angle feedback, calculate the average actual steering angle in this cycle.

[0037] S433, State Estimator Correction: The actual lateral displacement increment, heading angle increment, and average actual steering angle calculated in step S432 are used as observations and input into the vehicle state observer to correct and update the current lateral speed and yaw rate state estimates of the tractor.

[0038] S434, Periodic Synchronization Trigger: After completing the state estimator correction, a synchronization signal is issued to initiate the data acquisition and calculation of step S1 for the new control cycle.

[0039] Compared with existing technologies, this invention deeply integrates multi-source environmental perception data such as vision and lidar with soil physical parameters to generate an enhanced environmental map online that simultaneously contains geometric semantics and soil drivability information. This allows the path tracking system to analyze the geometric features and mechanical properties of the road surface ahead in advance. By dynamically constructing a pre-aiming window and stability constraints tightly coupled with road surface adhesion conditions, and updating the real-time soil adhesion coefficient as a key parameter in the vehicle dynamics model of model predictive control, the control system possesses adaptive capabilities. It can not only optimize path deviation but also proactively incorporate road surface adhesion conditions into the trade-offs and constraints of the control objectives, thereby achieving a smoother path. The system automatically softens control commands and suppresses the risk of slippage in low-adhesion areas such as soft and slippery surfaces, while maintaining agile tracking response on solid surfaces. It achieves a paradigm upgrade from "geometric path tracking" to "force-path coordinated tracking." Ultimately, while ensuring high-precision tracking, it significantly improves the system's adaptability and stability under different soil conditions, effectively reducing control instability or trajectory deviation caused by sudden changes in ground adhesion. At the same time, through a smooth and conservative adaptive control strategy, it reduces the impact wear and energy consumption of the steering actuator and alleviates soil compaction caused by aggressive steering, thus meeting the multiple requirements of precision agriculture for operational accuracy, adaptability, smoothness, and agronomic friendliness. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0041] Figure 1 This is a schematic diagram of the overall process provided for an embodiment of the present invention. Detailed Implementation

[0042] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0043] As attached Figure 1 As shown:

[0044] Example 1:

[0045] This invention provides an adaptive path tracking method for unmanned tractors based on multi-sensor fusion, comprising the following steps:

[0046] S1. Environmental and state perception through multi-sensor fusion: Simultaneously acquire and process data from different sensors to generate an enhanced environmental map containing soil drivability information and geometric path information;

[0047] S2. Constructing dynamic control windows and constraints: Based on the distribution of soil adhesion coefficients in the enhanced environmental map and the current motion state of the tractor, determine the aiming distance and vehicle stability constraints for model predictive control;

[0048] S3. Perform adaptive model predictive control calculation: Based on the output of steps S1 and S2, construct and solve online a control problem with path tracking accuracy, control smoothness and driving stability as comprehensive optimization objectives, and obtain the optimal front wheel steering angle control command;

[0049] S4. Execute control commands and provide closed-loop feedback: Send the optimal front wheel steering angle control command obtained in step S3 to the steer-by-wire system for execution, collect new state and environmental perception data, and return to step S1 to form closed-loop control.

[0050] S5. Repeat steps S1 to S4 until all job paths are completed.

[0051] Step S1 specifically includes the following sub-steps:

[0052] S11. Synchronous data acquisition: Acquire pose and heading data provided by the GNSS / INS integrated navigation system, environmental point cloud and image data obtained by the visual / LiDAR sensor, and real-time soil parameters obtained by the soil physics sensor.

[0053] S12. Generate environmental semantic layer: Process environmental point cloud and image data, identify travel areas, crop row center lines and obstacle boundaries, and generate a basic environmental semantic layer.

[0054] S13. Generate a soil adhesion coefficient distribution layer: Map real-time soil parameters to soil adhesion coefficients and generate a spatially continuous soil adhesion coefficient distribution layer based on their geographic coordinates.

[0055] S14. Generate an enhanced environmental map: Spatiotemporally fuse the basic environmental semantic layer with the soil adhesion coefficient distribution layer to form an enhanced environmental map.

[0056] Furthermore, in step S13, the soil parameters include at least soil moisture content and compaction, which are mapped to a dimensionless soil adhesion coefficient used to characterize tire-ground adhesion ability through a preset soil mechanics model.

[0057] Furthermore, step S2 specifically includes the following sub-steps:

[0058] S21. Dynamically determine the aiming distance: Based on the tractor's current longitudinal speed, path curvature, and the soil adhesion coefficient queried from the enhanced environmental map at the aiming point, adaptively adjust the aiming distance of the model predictive control; increase the aiming distance in low adhesion areas and decrease the aiming distance in high adhesion areas.

[0059] S22. Dynamically determine stability constraints: Based on the soil adhesion coefficient at the current and predicted trajectory points, dynamically calculate and set the corresponding lateral acceleration threshold and yaw rate threshold as time-varying state constraints in model predictive control.

[0060] First, the system is started and initialized, including calibration and self-testing of all sensors, positioning units, and control units. Once operation begins, the tractor control system enters the main cycle. The first step involves the system synchronously acquiring real-time latitude, longitude, heading angle, and speed data provided by a high-precision GNSS / IMU integrated navigation module. Simultaneously, it acquires images and point cloud data of the field ridges ahead via a front-mounted vision camera and lidar, and obtains soil volumetric moisture content and penetration resistance data of the currently traversed area through near-field soil sensors installed near the wheels. This raw data is aggregated to the central processing unit. The second step involves the data processing module processing the acquired environmental data. After filtering and segmentation, the vision and lidar point cloud data are used to identify clear crop row structures and fit a centerline path as a tracking reference. Simultaneously, soil sensor data is mapped to an estimated soil adhesion coefficient for the area below the tractor's center of gravity using a preset empirical transformation model. Although a detailed global adhesion map has not yet been generated, this real-time adhesion coefficient value is marked on the tractor's current position. At this point, the system obtains a global geometric path and a local, location-related soil adhesion information point. The third step involves the path tracking controller commencing operation. In this embodiment, the controller employs an enhanced pre-aiming tracking algorithm. Its core improvement lies in the fact that the control algorithm reads the local soil adhesion coefficient obtained in the previous step and dynamically adjusts its key parameter—the pre-aiming distance—based on this. Specifically, the algorithm incorporates a simple lookup table: when the adhesion coefficient is higher than a set threshold, a standard shorter pre-aiming distance is used to ensure tracking sensitivity on good road surfaces; when the adhesion coefficient is lower than the threshold, it automatically switches to a longer pre-aiming distance, allowing the controller to "see further" and plan smoother steering actions in advance, thereby avoiding sideslip caused by overly rapid reactions on low-adhesion surfaces. The controller calculates lateral and directional deviations based on the dynamic pre-aiming points and calculates the front wheel steering angle command using a proportional-derivative control law. Finally, this steering angle command is sent to the tractor's steer-by-wire system for execution, driving the vehicle along the path. The system then returns to the first step and enters the next control cycle, forming a closed loop. By introducing real-time adhesion perception and parameter adaptation mechanisms into the traditional geometric tracking algorithm, the problem of overly aggressive control commands on roads with sudden changes in adhesion is initially solved, improving the system's basic adaptability.

[0061] Example 2:

[0062] This embodiment is basically the same as the previous embodiment, except that step S3 specifically includes the following sub-steps:

[0063] S31. Construct a parameterized vehicle prediction model: Establish an extended prediction model that includes tractor lateral-yaw dynamics, steering system dynamics, and tire force model; wherein, the tire force model is a nonlinear model based on Pacejka's magic formula.

[0064] S32. Online update of prediction model parameters: Based on the enhanced environmental map output in step S1, obtain the soil adhesion coefficient corresponding to the current and predicted trajectory points in real time, and update the peak road adhesion coefficient parameter in the Pacejka magic formula online.

[0065] S33. Construct and solve the rolling optimization problem: Based on the parameterized prediction model updated in step S32, take the dynamic aiming distance determined in step S2 as the prediction time domain, take the slip risk after weighting the path tracking deviation, the rate of change of control quantity and the adhesion coefficient as the optimization objective, and take the dynamic stability constraint determined in step S2 as the boundary to perform rolling optimization and output the optimal front wheel steering angle control sequence.

[0066] In step S33, the slip risk penalty term after the adhesion coefficient is weighted is negatively correlated with the soil adhesion coefficient of the predicted point, which makes the controller more inclined to generate smooth and conservative control commands in low adhesion areas to suppress slip.

[0067] It should be noted that in step S4, before sending the control command to the steer-by-wire system, the control command is subjected to bandwidth adaptive low-pass filtering and amplitude limiting processing based on the current soil adhesion coefficient.

[0068] A model predictive control framework was adopted, and soil adhesion information was integrated more deeply into the control model and constraints. After system initialization, the first step of perception and data fusion was significantly enhanced. It not only simultaneously collected positioning, environmental image point clouds, and soil data, but more importantly, it bound the soil parameters acquired in the current and historical cycles to precise geographic coordinates, constructing and continuously updating a raster map of soil adhesion coefficient distribution covering the already traveled area and the short-distance predicted area ahead within the controller. Simultaneously, visual and laser data were processed to generate an environmental semantic layer containing crop rows and obstacle outlines. These two layers were fused in a unified field coordinate system to form an enhanced environmental map that simultaneously represents "where to go" (geometric path) and "road conditions" (adhesion conditions).

[0069] The second step involves the dynamic programming module performing forward-looking calculations based on this enhanced environmental map and the vehicle's current state. It not only dynamically calculates the target distance, but more importantly, it queries the map for the adhesion coefficients of a series of future points along the predicted trajectory. Based on these values, it dynamically calculates the maximum lateral acceleration and yaw rate that the vehicle can safely withstand at each future point, forming a series of time-varying stability constraints. These boundaries are crucial for ensuring the vehicle does not lose stability on low-adhesion surfaces.

[0070] The third step is the core computation of adaptive model predictive control. The controller incorporates a parameterized vehicle dynamics prediction model, which includes a nonlinear tire model based on Pacejka's magic formula. Before solving the optimization problem in each control cycle, the controller updates the key tire model parameter—the peak road adhesion coefficient—at each step in the prediction time domain with the estimated value of the corresponding location retrieved from the map, based on the enhanced environment map. This allows the prediction model to more realistically reflect the vehicle's dynamic response to different road surfaces in the future. Subsequently, the controller performs rolling time-domain optimization under the time-varying stability constraints generated in step two, aiming to minimize path tracking error, control variable changes, and a slip risk penalty term related to the adhesion coefficient, to obtain a sequence of optimal front wheel steering angles. Finally, the first element of this sequence is output as a control command to the steering actuator. This embodiment achieves proactive adaptation of the control system to road adhesion conditions by constructing a deep coupling between the environment model, dynamic constraints, and parameterized prediction model. This not only improves stability but also actively suppresses slippage behavior that may lead to energy waste and soil damage by optimizing the slip penalty term in the objective.

[0071] Example 3:

[0072] This embodiment is basically the same as the previous embodiment, except that step S4 specifically includes the following sub-steps:

[0073] S41. Control command post-processing: Based on the soil adhesion coefficient at the current tractor position obtained from the enhanced environment map, adaptive low-pass filtering and rate limiting processing are performed on the optimal front wheel steering angle control command to generate smooth and safe execution commands.

[0074] S42. Command Issuance and Execution: Send the execution command to the tractor's steering-by-wire system for execution;

[0075] S43. Status Update and Cycle Switching: Collect the execution feedback data of the steer-by-wire system and the latest vehicle status data, update the system status, and trigger the start of the next control cycle.

[0076] Furthermore, in step S41, the adaptive low-pass filtering and rate limiting processing specifically includes the following steps:

[0077] S411, Parameter Mapping: Based on the current soil adhesion coefficient, query the preset mapping table or use the predefined calculation function to determine the corresponding low-pass filter cutoff frequency and steering angle rate limiting threshold; the mapping relationship or function is configured as follows: the lower the soil adhesion coefficient, the smaller the values ​​of the cutoff frequency and limiting threshold.

[0078] S412. Filtering execution: Using the cutoff frequency determined in step S411, construct a first-order or second-order digital low-pass filter to perform real-time filtering calculations on the optimal front wheel steering angle control command to obtain a preliminary smoothing command.

[0079] S413, Limiting Execution: Calculate the rate of change of the preliminary smoothing instruction obtained in step S412. If its absolute value exceeds the limiting threshold determined in step S411, the rate of change is limited by the threshold, and the final execution instruction is generated accordingly.

[0080] Furthermore, in step S43, the state update and cycle switching specifically include the following steps:

[0081] S431, Data Synchronization Acquisition: At the end of the control cycle, the actual steering angle feedback from the steer-by-wire system and the latest attitude and velocity data from the GNSS / INS integrated navigation system are synchronously acquired, and all acquired data are assigned a unified timestamp.

[0082] S432. Calculation of actual state increment: Based on the pose data collected at the beginning and end of the previous control cycle, calculate the actual lateral displacement increment and heading angle increment of the tractor in this cycle; at the same time, based on the actual steering angle feedback, calculate the average actual steering angle in this cycle.

[0083] S433, State Estimator Correction: The actual lateral displacement increment, heading angle increment, and average actual steering angle calculated in step S432 are used as observations and input into the vehicle state observer to correct and update the current lateral speed and yaw rate state estimates of the tractor.

[0084] S434, Periodic Synchronization Trigger: After completing the state estimator correction, a synchronization signal is issued to initiate the data acquisition and calculation of step S1 for the new control cycle.

[0085] This embodiment refines and enhances the accuracy of control command execution and system closed-loop feedback. After completing the same enhanced environment map construction, dynamic constraint calculation, and adaptive model predictive control solution as in Embodiment 2, this embodiment adds a post-processing stage to the output optimal front wheel steering angle control command. This stage is not a simple low-pass filter, but an adaptive process deeply coupled with real-time adhesion conditions. The post-processing module first obtains the soil adhesion coefficient at the tractor's current precise position from the enhanced environment map. Then, based on this coefficient, it dynamically calculates two key execution parameters through a pre-calibrated nonlinear function: the cutoff frequency of the low-pass filter and the steering angular velocity limiting threshold. The logic is: the lower the adhesion coefficient, the lower the cutoff frequency is set to filter out more high-frequency command fluctuations, while the angular velocity limiting threshold is also set smaller to physically limit the severity of steering. The module then uses this dynamic cutoff frequency to filter the original control commands, and applies a dynamic threshold limit to the rate of change of the filtered commands, ultimately generating a smooth and absolutely safe final execution command, which is then sent to the steer-by-wire system. Simultaneously with the execution of the control commands, the system performs high-precision state preparation for the next control cycle. At the end of a control cycle, the system synchronously collects the actual steering angle fed back by the steer-by-wire system, as well as the latest pose data from the GNSS / IMU, and timestamps all data uniformly. Next, the system calculates the actual lateral displacement increment and heading angle increment of the vehicle based on the actual pose changes in the previous cycle, and simultaneously calculates the average actual steering angle for that cycle. These "actually occurring" motion increments are used as the most reliable observations and input into a vehicle state observer to correct and update key state estimates such as the vehicle's lateral velocity and yaw rate. This step effectively utilizes actual motion feedback to correct the accumulated errors of pure model predictions and sensor noise, ensuring the accuracy of the system's self-state perception. Finally, after all states have been updated and synchronization confirmed, the system issues a command to trigger a new cycle of sensor data acquisition and enhanced environmental map update calculation, thereby initiating a new round of optimized control loop. This embodiment achieves refined control across the entire chain from perception, decision-making, execution to feedback by attaching adaptive post-processing to control commands and incremental feedback correction of system states. This significantly improves the system's overall robustness, tracking accuracy, and long-term operational reliability in extremely complex and uncertain farmland environments.

[0086] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. An adaptive path tracking method for unmanned tractors based on multi-sensor fusion, characterized in that, Includes the following steps: S1. Environmental and state perception through multi-sensor fusion: Simultaneously acquire and process data from different sensors to generate an enhanced environmental map containing soil drivability information and geometric path information; S2. Constructing dynamic control window and constraints: Based on the distribution of soil adhesion coefficient in the enhanced environmental map and the current motion state of the tractor, determine the aiming distance and vehicle stability constraints of the model predictive control. S3. Perform adaptive model predictive control calculation: Based on the output of steps S1 and S2, construct and solve online a control problem with path tracking accuracy, control smoothness and driving stability as comprehensive optimization objectives, and obtain the optimal front wheel steering angle control command; S4. Execute control commands and provide closed-loop feedback: Send the optimal front wheel steering angle control command obtained in step S3 to the steer-by-wire system for execution, collect new state and environmental perception data, and return to step S1 to form closed-loop control. S5. Repeat steps S1 to S4 until all job paths are completed.

2. The adaptive path tracking method for unmanned tractors based on multi-sensor fusion according to claim 1, characterized in that, Step S1 specifically includes the following sub-steps: S11. Synchronous data acquisition: Acquire pose and heading data provided by the GNSS / INS integrated navigation system, environmental point cloud and image data obtained by the visual / LiDAR sensor, and real-time soil parameters obtained by the soil physics sensor. S12. Generate an environmental semantic layer: Process the environmental point cloud and image data, identify the travel area, crop row center line and obstacle boundary, and generate a basic environmental semantic layer. S13. Generate a soil adhesion coefficient distribution layer: Map the real-time soil parameters to soil adhesion coefficients, and generate a spatially continuous soil adhesion coefficient distribution layer based on their geographic coordinates. S14. Generate an enhanced environmental map: Spatiotemporally fuse the basic environmental semantic layer with the soil adhesion coefficient distribution layer to form the enhanced environmental map.

3. The adaptive path tracking method for unmanned tractors based on multi-sensor fusion according to claim 2, characterized in that, In step S13, the soil parameters include at least soil moisture content and compaction, which are mapped to a dimensionless soil adhesion coefficient to characterize tire-ground adhesion ability through a preset soil mechanics model.

4. The adaptive path tracking method for unmanned tractors based on multi-sensor fusion according to claim 3, characterized in that, Step S2 specifically includes the following sub-steps: S21. Dynamically determine the aiming distance: Based on the tractor's current longitudinal speed, path curvature, and the soil adhesion coefficient queried from the enhanced environmental map at the aiming point, adaptively adjust the aiming distance of the model predictive control; increase the aiming distance in low adhesion areas and decrease the aiming distance in high adhesion areas. S22. Dynamically determine stability constraints: Based on the soil adhesion coefficient at the current and predicted trajectory points, dynamically calculate and set the corresponding lateral acceleration threshold and yaw rate threshold as time-varying state constraints in model predictive control.

5. The adaptive path tracking method for unmanned tractors based on multi-sensor fusion according to claim 4, characterized in that, Step S3 specifically includes the following sub-steps: S31. Construct a parameterized vehicle prediction model: Establish an extended prediction model that includes tractor lateral-yaw dynamics, steering system dynamics, and tire force model; wherein, the tire force model is a nonlinear model based on Pacejka's magic formula. S32. Online update of prediction model parameters: Based on the enhanced environmental map output in step S1, obtain the soil adhesion coefficient corresponding to the current and predicted trajectory points in real time, and update the peak road adhesion coefficient parameter in the Pacejka magic formula online. S33. Construct and solve the rolling optimization problem: Based on the parameterized prediction model updated in step S32, with the dynamic aiming distance determined in step S2 as the prediction time domain, the slip risk weighted by path tracking deviation, control variable change rate and adhesion coefficient as the optimization objective, and the dynamic stability constraint determined in step S2 as the boundary, perform rolling optimization solution and output the optimal front wheel steering angle control sequence.

6. The adaptive path tracking method for unmanned tractors based on multi-sensor fusion according to claim 5, characterized in that, In step S33, the weight of the slip risk penalty term after the adhesion coefficient is negatively correlated with the soil adhesion coefficient of the prediction point, which makes the controller more inclined to generate smooth and conservative control commands in low adhesion areas to suppress slip.

7. The adaptive path tracking method for unmanned tractors based on multi-sensor fusion according to claim 6, characterized in that, In step S4, before sending the control command to the steer-by-wire system, the control command is subjected to bandwidth adaptive low-pass filtering and amplitude limiting processing based on the current soil adhesion coefficient.

8. The adaptive path tracking method for unmanned tractors based on multi-sensor fusion according to claim 7, characterized in that, Step S4 specifically includes the following sub-steps: S41. Control command post-processing: Based on the soil adhesion coefficient at the current tractor position obtained from the enhanced environmental map, the optimal front wheel steering angle control command is subjected to adaptive low-pass filtering and rate limiting processing to generate a smooth and safe execution command. S42. Instruction Issuance and Execution: The execution instruction is sent to the tractor's steering-by-wire system for execution; S43. Status update and cycle switching: Collect the execution feedback data of the steer-by-wire system and the latest vehicle status data, update the system status, and trigger the start of the next control cycle.

9. The adaptive path tracking method for unmanned tractors based on multi-sensor fusion according to claim 8, characterized in that, In step S41, the adaptive low-pass filtering and rate limiting processing specifically includes the following steps: S411, Parameter Mapping: Based on the current soil adhesion coefficient, query a preset mapping table or use a predefined calculation function to determine the corresponding low-pass filter cutoff frequency and steering angle rate limiting threshold; the mapping relationship or function is configured such that the lower the soil adhesion coefficient, the smaller the values ​​of the cutoff frequency and the limiting threshold. S412. Filtering execution: Using the cutoff frequency determined in step S411, construct a first-order or second-order digital low-pass filter to perform real-time filtering calculations on the optimal front wheel steering angle control command to obtain a preliminary smoothing command. S413, Limiting Execution: Calculate the rate of change of the preliminary smoothing instruction obtained in step S412. If its absolute value exceeds the limiting threshold determined in step S411, limit the rate of change with the threshold and generate the final execution instruction accordingly.

10. The adaptive path tracking method for unmanned tractors based on multi-sensor fusion according to claim 9, characterized in that, In step S43, the state update and cycle switching specifically include the following steps: S431, Data Synchronization Acquisition: At the end of the control cycle, the actual steering angle feedback from the steer-by-wire system and the latest attitude and velocity data from the GNSS / INS integrated navigation system are synchronously acquired, and all acquired data are assigned a unified timestamp. S432. Calculation of actual state increment: Based on the pose data collected at the beginning and end of the previous control cycle, calculate the actual lateral displacement increment and heading angle increment of the tractor in this cycle; at the same time, based on the actual steering angle feedback, calculate the average actual steering angle in this cycle. S433, State Estimator Correction: The actual lateral displacement increment, heading angle increment, and average actual steering angle calculated in step S432 are used as observations and input into the vehicle state observer to correct and update the current lateral speed and yaw rate state estimates of the tractor. S434, Periodic Synchronization Trigger: After completing the state estimator correction, a synchronization signal is issued to initiate the data acquisition and calculation of step S1 for the new control cycle.

Images