Irrigation robot inter-row navigation and obstacle avoidance method and system

Abstract

The invention provides an irrigation robot inter-row navigation and obstacle avoidance method and system, and belongs to the technical field of high-precision navigation. The invention aims to solve the problem of path planning and execution caused by dynamic obstacles, sudden animal interference and environment change. The method comprises the following steps: acquiring a current robot pose; judging the motion trail of the dynamic obstacle; generating an execution path of the irrigation robot; generating a standby navigation sequence; generating an execution path of the irrigation robot; generating a prediction result of the motion trail of the burst animal; and a safe irrigation execution instruction is generated. The system comprises a path planning module, a low-power-consumption processor, a navigation instruction generation module, a path adjustment module, a safety distance verification module, a trajectory smoothing module, a real-time feedback module, a feature extraction module, an image processing accelerator, a trajectory prediction analysis module, a path analysis module, an obstacle avoidance decision module and an irrigation sequence optimization module. A distance adjustment module and a track verification module.

Description

Technical Field

[0001] This invention relates to a method and system for row navigation and obstacle avoidance of an irrigation robot, belonging to the field of high-precision navigation technology. Background Technology

[0002] In modern agriculture, intelligent irrigation robots are considered a crucial technological means to improve farmland productivity and resource utilization. Especially in complex, unstructured environments, such as farmland with its complex terrain, diverse obstacles, and unstable signals, achieving precise navigation and safe obstacle avoidance is key to driving agricultural automation. Efficient irrigation through robots not only optimizes water resource allocation but also reduces the cost and error of manual operations. However, current navigation and obstacle avoidance technologies face significant technical challenges in the dynamic and non-standardized environment of farmland, limiting the widespread application of irrigation robots in actual production. Existing methods have obvious limitations when dealing with complex farmland environments.

[0003] Traditional navigation technologies largely rely on Global Navigation Satellite System (GNSS) signals, but in farmland, these signals are often unstable due to terrain obstruction or vegetation interference, making it difficult to meet the demands for high-precision positioning. Meanwhile, obstacle avoidance methods based on two-dimensional vision are ineffective at distinguishing between crops and hard obstacles, especially under conditions of significant lighting variations. The vision system may misjudge crops as obstacles or identify hard objects to be avoided as objects that can be run over. These methods are slow to react to rapidly changing obstacles in dynamic environments, failing to meet real-time requirements and hindering the robot's efficient operation in complex farmland. The core technical challenge lies in how to achieve low-latency, accurate navigation and safe obstacle avoidance using only onboard computing units.

[0004] First, the dynamic nature of farmland environments demands that robots complete environmental perception and decision-making within less than 100 milliseconds. The limited computing power of onboard computing units makes real-time processing of complex algorithms difficult, especially when navigation and obstacle avoidance tasks need to be handled simultaneously, making the allocation of computing resources a bottleneck. This low-latency requirement further amplifies the second technical challenge: how to accurately distinguish between crops and hard obstacles in scenarios with changing lighting and dynamic obstacles. For example, in farmland, irrigation robots may need to identify whether the obstacle ahead is a low-lying crop that can be crushed or a stone or person that needs to be avoided while moving rapidly. Because changes in lighting can interfere with the data from visual sensors, the robot may misjudge and collide with obstacles or incorrectly detour around crops, affecting irrigation efficiency. Therefore, how to achieve precise inter-row navigation and safe obstacle avoidance with a latency of less than 100 milliseconds using only onboard computing units in complex farmland environments with unstable GNSS signals and frequent dynamic obstacles has become a key issue for breakthroughs in intelligent irrigation robot technology. Summary of the Invention

[0005] This invention addresses the path planning and execution problems caused by dynamic obstacles, sudden animal interference, and environmental changes, and proposes a method and system for inter-row navigation and obstacle avoidance for irrigation robots.

[0006] The technical solution adopted by the present invention to solve the above problems is: a method for inter-row navigation and obstacle avoidance of an irrigation robot, comprising: Step 1: Collect real-time field data, generate a preliminary environmental map, use a sensor fusion engine to integrate the preliminary environmental map, determine the current pose of the irrigation robot and obtain the real-time environmental map; Step 2: Extract local visual data based on the current pose of the irrigation robot, preprocess the local visual distance data, and use the pose calculation component to detect the preprocessed local visual data to obtain the motion trajectory of the dynamic obstacle. Step 3: If the trajectory of the dynamic obstacle overlaps with the path of the irrigation robot, an alternative obstacle avoidance path is generated through the path generation module and the navigation command is adjusted. The adjusted navigation command is corrected based on the environmental data in the real-time environmental map to generate the execution path of the irrigation robot. Step 4: If the execution path of the irrigation robot does not meet the low-latency response requirement, the path is updated through a fast iterative optimization algorithm to obtain a safe irrigation execution sequence. In the safe irrigation execution sequence, the dynamic obstacle trajectory is predicted and the collision probability is calculated. If the collision probability exceeds the threshold, the path is adjusted and a backup navigation sequence is generated. Step 5: Integrate the backup navigation sequence and real-time environmental map data and input them into the low-power processor to obtain the obstacle classification results. Use the navigation instruction generation module to verify the safe distance maintenance of the obstacle classification and recognition results to obtain the corrected execution path of the irrigation robot. Step 6: Extract the real-time environmental feature vector of the corrected execution path of the irrigation robot, compensate for the muddy ground interference in the real-time environmental feature vector, obtain the compensated visual feature data, and use the pose calculation component to perform dynamic trajectory prediction on the compensated visual feature data to obtain the prediction result of the movement trajectory of the sudden animal. Step 7: If the predicted trajectory of the sudden animal overlaps with the corrected execution path of the irrigation robot, then generate a set of alternative obstacle avoidance paths between the rows of the field based on the obstacle avoidance decision logic and determine the final navigation sequence between the rows of the field. Step 8: Convert the final field row navigation sequence into navigation instructions that the equipment can execute, and obtain safe irrigation execution instructions.

[0007] Furthermore, the pose calculation component includes: A combination of low-power processors, image processing accelerators, and sensors; The sensor array is used to collect real-time data on the field environment; Image processing accelerators are used to compensate for muddy ground interference in environmental feature vectors; Inertial measurement units are used to collect inertial measurement unit data of the field environment; The pose calculation component is used to extract visual feature data and detect the extracted visual feature data.

[0008] Furthermore, step 1 specifically includes: Real-time data is collected from the field environment using a combination of sensors to obtain raw environmental data, which includes lidar point clouds, camera images, and inertial measurement unit data. The lidar point clouds are then denoised and registered, and the camera images are fused for feature extraction to generate a preliminary environmental map. Missing values ​​are detected in the preliminary environmental map. If there are missing data in the preliminary environmental map, the missing areas are filled in by interpolation to obtain a complete preliminary environmental map. The sensor fusion engine integrates the completed preliminary environmental map, and the inertial measurement unit data is fused to compensate for the GNSS signal to obtain an optimized environmental map. The current pose of the irrigation robot is determined by fusing the optimized environmental map and inertial measurement unit data using the Kalman filter algorithm. If the current pose deviation of the irrigation robot exceeds a first preset threshold, the pose is optimized using the particle filter algorithm to obtain a high-precision pose. The optimized environmental map is then updated based on the high-precision pose to obtain a real-time environmental map.

[0009] Furthermore, step 2 specifically includes: Based on the current pose of the irrigation robot, local visual data is extracted from the real-time data collected by the sensor combination. The local visual data is then denoised to obtain the first visual data. An image processing accelerator is used to correct for illumination changes in the first visual data to obtain the second visual data. The second visual data is used to extract features by a pose calculation component, and the structure of crop rows in the field is detected to obtain the crop distribution. If there are non-crop row areas in the crop distribution, the location of hard obstacles is determined by an edge detection algorithm. Based on the location of the hard obstacle, the second visual data is analyzed using an optical flow algorithm to obtain the motion trajectory of the dynamic obstacle.

[0010] Furthermore, step 3 specifically includes: Step 3.1: If the speed of the dynamic obstacle exceeds the second preset threshold, the Kalman filter algorithm is used to predict the future position of the dynamic obstacle, and the navigation path of the irrigation robot is updated according to the predicted position of the obstacle to generate dynamic navigation instructions. Step 3.2: If the trajectory of the dynamic obstacle overlaps with the updated navigation path of the irrigation robot, the alternative obstacle avoidance path is calculated through the path planning module to obtain a set of alternative paths; if they do not overlap, the updated navigation path of the irrigation robot is directly converted into a navigation command that the device can execute to obtain a safe irrigation execution command. Step 3.3: Use real-time data collected by the sensor combination to perform environmental constraint analysis on the candidate path set to obtain the filtered path set; Step 3.4: Integrate the filtered path set with the motion trajectory of the dynamic obstacle through the fusion algorithm to obtain the optimal obstacle avoidance path; if the optimal obstacle avoidance path meets the preset passage conditions, adjust the navigation path of the irrigation robot according to the optimal obstacle avoidance path; if it does not meet the conditions, repeat steps 3.1-3.3 until the preset passage conditions are met. Step 3.5: Use the original environmental data to perform real-time correction on the navigation path of the irrigation robot to obtain the corrected navigation instructions. Update the motion control parameters of the irrigation robot with the corrected navigation instructions to determine the final field row navigation instructions. Generate drive signals based on the final row navigation instructions to obtain the execution path of the irrigation robot.

[0011] Furthermore, step 4 specifically includes: If the execution path of the irrigation robot coincides with the movement trajectory of the dynamic obstacle, the path planning module calculates the set of alternative paths to obtain a preliminary path set. Environmental constraint analysis is performed on the initial path set using real-time sensor data to select an optimized path set that meets the passage conditions. The potential interference in the optimized path set is evaluated using a dynamic trajectory prediction algorithm to determine the optimal obstacle avoidance path. By integrating the optimal obstacle avoidance path with the collected real-time data, an adjusted irrigation sequence is generated. A fast iterative optimization algorithm is used to verify the low-latency response of the adjusted irrigation sequence to obtain a safe execution sequence. The drive signal is generated based on the safe execution sequence, the final irrigation control command is determined, and the final irrigation control command is dynamically corrected by real-time data collected by the sensor combination to obtain the backup navigation sequence.

[0012] Furthermore, step 5 specifically includes: The backup navigation sequence and real-time environmental map data are integrated and input into a low-power processor for feature extraction and classification to obtain obstacle classification results; The obstacle classification results are input into the navigation instruction generation module to obtain the initial navigation instruction sequence. If the initial navigation instruction sequence overlaps with the obstacle positions in the obstacle classification results, the alternative paths are recalculated through the path adjustment module to obtain the adjusted navigation instruction sequence. The adjusted navigation command sequence is analyzed for distance constraints by the safety distance verification module to obtain a navigation command sequence that meets the safety distance requirement. The trajectory smoothing module is then used to optimize the navigation command sequence that meets the safety distance requirement to obtain a smooth inter-line path planning. The smooth inter-row path planning is dynamically corrected by a real-time feedback module to obtain the corrected execution path of the irrigation robot.

[0013] Furthermore, step 6 specifically includes: The real-time environmental feature vector of the irrigation robot's execution path is extracted by the feature extraction module. The muddy ground interference in the environmental feature vector is compensated by the image processing accelerator to obtain the compensated visual feature data. A pose calculation component is used to perform dynamic pose estimation on the compensated visual feature data to obtain dynamic pose parameters. The trajectory prediction and analysis module is then used to predict the motion trajectory based on the dynamic pose parameters to obtain the predicted motion trajectory data. Based on the predicted motion trajectory data, the corresponding sensor images are acquired. The YOLOv5s model is used to detect animal targets in the images and extract the bounding boxes of the animal targets. The animal movement is tracked by Kalman filtering, and the movement trajectory of sudden animals is predicted.

[0014] Furthermore, step 7 specifically includes: If the predicted trajectory of the sudden animal overlaps with the corrected execution path of the irrigation robot, the trajectory features of the overlapping area are extracted through the path analysis module to obtain the feature data of the overlapping area. The obstacle avoidance decision module processes the feature data of overlapping areas to generate alternative obstacle avoidance paths between rows of fields, thus obtaining a set of alternative paths. A path fusion algorithm is used to integrate the candidate path set with the environmental map data to obtain fused path data. The irrigation sequence optimization module then adjusts the sequence of the fused path data to obtain an optimized irrigation execution sequence. If the optimized irrigation execution sequence does not meet the preset obstacle safety distance, the path offset is calculated through the distance adjustment module to obtain the adjusted safe path data. The trajectory verification module was used to perform a feasibility analysis on the adjusted safe path data to obtain the final field row navigation sequence.

[0015] A navigation and obstacle avoidance system for irrigation robots includes: The path planning module is used to plan the execution path of the irrigation robot, and inputs alternative obstacle avoidance paths and a preliminary path; The navigation instruction generation module is used to generate navigation instructions for the irrigation robot. The path adjustment module is used to adjust the execution path of the irrigation robot; The safety distance verification module is used to perform distance constraint analysis on the navigation commands of the irrigation robot; The trajectory smoothing module is used to optimize navigation command sequences that meet safety distance requirements; A real-time feedback module is used to dynamically correct smooth inter-line path planning; The feature extraction module is used to extract the real-time environmental feature vector of the execution path of the irrigation robot; The trajectory prediction and analysis module is used to predict the motion trajectory of the irrigation robot based on its dynamic position parameters. The path analysis module is used to extract trajectory features of the overlapping area between the execution path of the irrigation robot and the movement trajectory of the sudden animal. The obstacle avoidance decision module is used to analyze the trajectory characteristics of overlapping areas and generate alternative obstacle avoidance paths between field rows; The irrigation sequence optimization module is used to adjust the sequence of path data; The distance adjustment module is used to adjust the path offset of the irrigation execution sequence; The trajectory verification module is used to perform feasibility analysis on safe path data; An image processing accelerator is used to compensate for muddy ground interference in environmental feature vectors. A low-power processor is used to perform feature extraction and classification processing on integrated data to output obstacle classification results.

[0016] The beneficial effects of this invention are: 1. This invention utilizes multi-sensor fusion to collect LiDAR point clouds, camera images, and inertial measurement unit data to generate and optimize a preliminary environmental map, compensating for GNSS signal instability to accurately determine the robot's pose. To address interference from lighting and mud, an image processing accelerator is used to enhance visual data, combined with a pose calculation component to achieve crop row detection and dynamic obstacle trajectory prediction. When obstacles or sudden animal trajectories overlap with the path, this invention generates alternative paths through obstacle avoidance decision logic and a path generation module, fusing real-time environmental data for rapid iterative optimization to form a low-latency inter-row path.

[0017] 2. This invention generates a safe irrigation sequence for inter-row irrigation in fields by using dynamic trajectory prediction and low-power processor verification of safe distances. Its core technological advantage lies in ensuring that the robot completes irrigation tasks efficiently, accurately, and safely through multi-sensor fusion and real-time optimization, significantly improving its autonomous navigation and operational capabilities in complex agricultural environments. Attached Figure Description

[0018] Figure 1This is a flowchart illustrating a method for row navigation and obstacle avoidance in an irrigation robot. Figure 2 A flowchart for obtaining the execution path of the irrigation robot; Figure 3 Flowchart for obtaining instructions to ensure safe irrigation; Figure 4 This is a structural block diagram of an irrigation robot's inter-row navigation and obstacle avoidance system. Detailed Implementation

[0019] Combination Figure 1-3 This implementation method is described as follows: Figure 1 As shown, the steps of the irrigation robot's inter-row navigation and obstacle avoidance method according to this embodiment include: S101: Real-time data is collected from the field environment through multi-sensor fusion, including LiDAR point cloud and camera images. After processing, a preliminary environmental map is obtained. The sensor fusion engine is used to integrate the preliminary environmental map, fuse inertial measurement unit data to compensate for GNSS signal instability, determine the current robot pose and generate a real-time environmental map. In this embodiment, the pose calculation component includes a low-power processor, an image processing accelerator, an inertial measurement unit (IMU), and a sensor assembly. The sensor assembly includes a LiDAR, an RGB camera, and a depth camera. The LiDAR is a Velodyne VLP-16 with 16 lines and a 360-degree horizontal field of view, used to acquire point cloud data. The RGB camera is a FLIRBlackflyS with a resolution of 1280x720 pixels, used for high-frame-rate image acquisition. The depth camera is an Intel RealSense D435 with a depth resolution of 1280x720, used to capture local depth images and generate point clouds. There is no specific model requirement for the IMU.

[0020] In this embodiment, real-time data is collected from the field environment using a combination of sensors to obtain LiDAR point clouds and camera images, resulting in raw environmental data. Data preprocessing methods are used to reduce noise and register the LiDAR point clouds, and feature extraction is performed by fusing the camera images to generate a preliminary environmental map. If data is missing in the preliminary environmental map, interpolation methods are used to fill in the missing areas, resulting in a complete preliminary environmental map. The preliminary environmental map is then integrated using a sensor fusion engine, fusing inertial measurement unit (IMU) data and correcting for GNSS signal instability to obtain an optimized environmental map. A Kalman filter algorithm is used to fuse the optimized environmental map and IMU data to determine the robot's current pose. If the robot's pose deviation exceeds a preset threshold, a particle filter algorithm is used to optimize the pose, resulting in a high-precision pose estimate. Based on the high-precision pose estimate, the optimized environmental map is updated, generating a real-time dynamic environmental map.

[0021] S102: Extract local region features based on the current robot pose, enhance the image processing accelerator to address the interference of lighting changes, obtain clear visual data, and use the pose calculation component to perform crop row detection on the clear visual data, distinguish between crops and hard obstacles, and obtain the motion trajectory of dynamic obstacles. The robot's current pose is acquired, and local visual data is extracted from sensor data. Preprocessing methods are used to denoise the visual data, resulting in first visual data. An image processing accelerator is used to correct for illumination changes in the first visual data, yielding second visual data. A pose calculation component extracts features from the second visual data, detecting crop row structure and obtaining crop row distribution. Based on the crop row distribution, if non-crop row areas are detected, an edge detection algorithm is used to identify hard obstacles and determine their locations. For the hard obstacle locations, an optical flow algorithm is used to analyze the second visual data to determine the motion trajectory of dynamic obstacles, obtaining motion trajectory data. Based on the motion trajectory data, if the speed of a dynamic obstacle exceeds a preset threshold, a Kalman filter algorithm is used to predict the future position of the dynamic obstacle, obtaining predicted position data. Based on the predicted position data, the robot's navigation path is updated, and dynamic navigation commands are generated.

[0022] S103: If the trajectory of a dynamic obstacle overlaps with the robot's path, an alternative obstacle avoidance path is generated based on the path generation module, and an adjusted navigation command is obtained. The obstacle avoidance decision logic is used to fuse the adjusted navigation command with real-time environmental perception data to determine the execution path of the irrigation robot. like Figure 2 As shown, obtaining the execution path of the irrigation robot includes: If the trajectory of a dynamic obstacle overlaps with the robot's path, the path planning module calculates alternative obstacle avoidance paths, resulting in a set of alternative paths. Real-time sensor data is used to perform environmental constraint analysis on the alternative path set, resulting in a filtered path set. A fusion algorithm integrates the filtered path set with the dynamic obstacle trajectory to determine the optimal obstacle avoidance path. If the optimal obstacle avoidance path meets preset passage conditions, an adjusted navigation path is generated based on the optimal obstacle avoidance path. Environmental perception data is used to correct the adjusted navigation path in real time, resulting in corrected navigation commands. The robot's motion control parameters are updated using the corrected navigation commands to determine the final inter-row navigation commands. Drive signals are generated based on the final inter-row navigation commands to obtain the robot's execution path.

[0023] S104: Determine whether the final inter-row path meets the low-latency response requirement, update the path through a fast iterative optimization algorithm, obtain a safe irrigation execution sequence, optimize the irrigation sequence by applying dynamic trajectory prediction to assess potential interference, and generate a backup navigation sequence. If the trajectory of a dynamic obstacle overlaps with the irrigation path, a set of alternative paths is calculated using the path planning module to obtain a preliminary path set. Environmental constraint analysis is performed on the preliminary path set using real-time sensor data to select an optimized path set that meets the passage conditions. A dynamic trajectory prediction algorithm is used to evaluate potential interference in the optimized path set to determine the optimal obstacle avoidance path. The optimal obstacle avoidance path and real-time environmental perception data are fused to generate an adjusted irrigation sequence. A fast iterative optimization algorithm is used to perform low-latency response verification on the adjusted irrigation sequence to obtain a safe execution sequence. A drive signal is generated based on the safe execution sequence to determine the final irrigation control command. The final irrigation control command is dynamically corrected using real-time sensor data to obtain a backup navigation sequence.

[0024] S105: Acquire backup sequence and environmental map update data, merge them into decision input, process the decision input through a low-power processor, determine the obstacle classification and recognition result, use navigation command output to verify the safe distance maintenance of the obstacle classification and recognition result, and obtain the execution path of the irrigation robot. The system acquires updated environmental map data and backup sequence data, integrates the data through a sensor fusion module to obtain decision input data. A low-power processor performs feature extraction and classification on the decision input data to obtain obstacle classification results. A navigation command generation module generates an initial navigation command sequence based on the obstacle classification results. If the navigation command sequence overlaps with the obstacle locations in the obstacle classification results, a path adjustment module calculates alternative paths to obtain an adjusted navigation command sequence. A safe distance verification module performs distance constraint analysis on the adjusted navigation command sequence to obtain a verified path that meets the safe distance requirements. Based on the verified path, a trajectory smoothing module optimizes the path to obtain a smooth inter-line path plan. A real-time feedback module dynamically corrects the smooth inter-line path plan to obtain the final execution path.

[0025] S106: Extract real-time data features based on the execution path of the irrigation robot, apply an image processing accelerator to compensate for interference from muddy ground, obtain compensated visual data, and use a pose calculation component to perform dynamic trajectory prediction on the compensated visual data to determine the movement trajectory of sudden animal interference. The process acquires the execution path data of the irrigation robot, processes the real-time environmental features through a feature extraction module to obtain an environmental feature vector, and compensates for muddy ground interference in the environmental feature vector using an image processing accelerator to obtain compensated visual feature data. A pose calculation component is used to perform dynamic pose estimation on the compensated visual feature data to obtain dynamic pose parameters. A trajectory prediction and analysis module is used to predict the motion trajectory based on the dynamic pose parameters to obtain predicted motion trajectory data. Based on the predicted motion trajectory data, corresponding sensor-acquired images are acquired, and a YOLOv5s model is used to detect animal targets in the images and extract their bounding boxes. Kalman filtering is then used to track animal movement and predict the trajectory of sudden animal movements.

[0026] S107: If the movement trajectory of a sudden animal disturbance overlaps with the verified inter-row path, then an alternative obstacle avoidance path is generated for the inter-row path based on the obstacle avoidance decision logic, and the adjusted navigation instructions for the inter-row path are obtained. The adjusted navigation instructions for the inter-row path are integrated with the updated environmental map to determine the final inter-row path. The final inter-row path is then updated through irrigation sequence optimization to obtain a safe inter-row irrigation execution sequence. like Figure 3 As shown, the steps for obtaining a safe inter-row irrigation execution sequence are as follows: If the movement trajectory data of sudden animal interference overlaps with the verified path, the path analysis module extracts the trajectory features of the overlapping area to obtain overlapping area feature data. The obstacle avoidance decision module processes the overlapping area feature data to generate alternative obstacle avoidance paths between field rows, resulting in a set of alternative paths. A path fusion algorithm integrates the alternative path set with the environmental map data to obtain fused path data. The irrigation sequence optimization module adjusts the sequence of the fused path data to obtain an optimized irrigation execution sequence. If the optimized irrigation execution sequence does not meet the preset obstacle safety distance, the distance adjustment module calculates the path offset to obtain adjusted safe path data. The trajectory verification module performs a feasibility analysis on the adjusted safe path data to obtain the final field row navigation sequence. The instruction generation module converts the final field row navigation sequence into executable navigation instructions for the equipment, obtaining safe irrigation execution instructions.

[0027] Furthermore, this embodiment also proposes an inter-row navigation and obstacle avoidance system for irrigation robots, such as... Figure 4 As shown, it includes: The path planning module 201 is used to plan the execution path of the irrigation robot, and inputs alternative obstacle avoidance paths and a preliminary path; The low-power processor 202 is used to perform feature extraction and classification processing on the integrated data and output obstacle classification results.

[0028] The navigation instruction generation module 203 is used to generate navigation instructions for the irrigation robot; The path adjustment module 204 is used to adjust the execution path of the irrigation robot; The safety distance verification module 205 is used to perform distance constraint analysis on the navigation commands of the irrigation robot; The trajectory smoothing module 206 is used to optimize the navigation command sequence that meets the safety distance requirement; The real-time feedback module 207 is used to dynamically correct the smooth inter-row path planning; The feature extraction module 208 is used to extract the real-time environmental feature vector of the execution path of the irrigation robot; Image processing accelerator 301 is used to compensate for muddy ground interference in environmental feature vectors; The trajectory prediction and analysis module 302 is used to predict the motion trajectory of the irrigation robot based on its dynamic position parameters. The path analysis module 303 is used to extract the trajectory features of the overlapping area between the execution path of the irrigation robot and the movement trajectory of the sudden animal. The obstacle avoidance decision module 304 is used to analyze the trajectory characteristics of overlapping areas and generate alternative obstacle avoidance paths between field rows; The irrigation sequence optimization module 305 is used to adjust the sequence of path data; Distance adjustment module 306 is used to adjust the path offset of the irrigation execution sequence; The trajectory verification module 307 is used to perform feasibility analysis on the safe path data; Example This embodiment is a specific description of the present invention. Taking a cornfield as an example, this embodiment further illustrates the row navigation and obstacle avoidance method for irrigation robots proposed in this invention, including: S1: Obtain the current robot pose; When collecting real-time data from the cornfield environment, this embodiment uses a Velodyne VLP-16 LiDAR to generate point cloud data at a frequency of 20 frames per second. The point cloud resolution is 0.2 degrees, covering a 360-degree horizontal field of view and a vertical field of view of -15 degrees to +15 degrees, generating approximately 30,000 points per frame. Simultaneously, a FLIRBlackflyS camera is used to acquire 1280x720 pixel RGB images at 30 frames per second. The point cloud data is processed using the VoxelGrid filter from the PCL library, with a voxel size set to 0.05 meters, reducing the point cloud density to approximately 5,000 points per frame to reduce computational load while preserving the cornfield topography and plant outlines. Camera images are corrected for distortion using OpenCV, employing a camera intrinsic parameter matrix (fx=800, fy=800, cx=640, cy=360) to ensure image alignment with the point cloud. The initial environmental map construction employs the RTAB-Map algorithm, registering point clouds with images. Key points are extracted from the images using SIFT features, and a local 3D map with a resolution of 0.1 meters is generated by combining the point cloud with the ICP algorithm (maximum iterations 30, distance threshold 0.1 meters). The sensor fusion engine uses an extended Kalman filter (EKF) to integrate data, inputting LiDAR point clouds, image features, and acceleration (range ±4g, resolution 0.01g) and angular velocity (range ±500 degrees / second, resolution 0.1 degrees / second) from the inertial measurement unit (IMU). The EKF state vector includes robot position (x, y, z), velocity, and attitude (quaternions). The process noise covariance matrix Q is set to 0.01, the measurement noise covariance R is set to 0.1, and the fusion frequency is 100Hz. To address the instability of GNSS signals (positioning accuracy ±2 meters, update rate 1 Hz), IMU data is preprocessed using a Madgwick filter (gain β=0.1) to compensate for attitude drift when GNSS signals are lost. After fusion, the robot's pose error is controlled within 0.05 meters. The final pose is aligned with the map through coordinate transformation (homogeneous transformation matrix, translation error <0.05 meters, rotation error <0.5 degrees) to ensure navigation accuracy in the cornfield. The entire process is implemented using ROS, with modules communicating via topic for smooth data transmission. The logical chain, from data acquisition to map building and fusion optimization, forms a closed loop, meeting real-time navigation requirements.

[0029] S2: Determine the trajectory of a dynamic obstacle; Based on the current robot pose (x, y, z, pose quaternions), when extracting local features from the cornfield environment, a depth camera (such as Intel RealSense D435, depth resolution 1280x720, frame rate 30 fps, field of view 85.2°x58°) can be used to capture depth images of the local area, generating point cloud data (approximately 921,600 points / frame). The depth range is limited (0.5 m to 3 m) using the PassThrough filter in the PCL library, preserving corn plant and ground features and reducing the point cloud density to approximately 100,000 points / frame. To address interference from varying lighting conditions, an NVIDIA Jetson TX2 image processing accelerator is used to run the CLAHE algorithm (contrast-limited adaptive histogram equalization, cropping limit 2.0, grid size 8x8) to enhance the RGB image (resolution 1920x1080) acquired by the depth camera. The brightness range (0-255) is adjusted to improve the edge sharpness of plants in low-light areas, resulting in an approximately 30% increase in output image contrast. Subsequently, using the pose calculation component and enhanced RGB images, crop rows were detected using Hough transform (angular resolution 1°, minimum line segment length 0.2m), and straight line features were extracted (slope error <0.05) to distinguish crop rows from hard obstacles (such as stones, width >0.3m, height >0.1m). To determine the trajectory of dynamic obstacles (such as moving agricultural machinery), a YOLOv5 model (input image size 640x640, confidence threshold 0.6) was used to detect dynamic targets. Combined with Kalman filtering (state vector includes position x,y and velocity vx,vy, process noise covariance 0.01, measurement noise covariance 0.1, update frequency 50Hz), the target motion was tracked, and the predicted trajectory deviation was controlled within 0.1m. The entire process is implemented using ROS2. Data is transmitted through topics (such as / camera / depth, / obstacle / trajectory). Feature extraction, image enhancement, crop row detection, and dynamic obstacle tracking are completed sequentially to form a closed loop, ensuring that the robot can navigate based on crop rows and avoid obstacles.

[0030] S3: Generate the execution path for the irrigation robot; In agricultural robot navigation, when the trajectory of a dynamic obstacle (such as moving agricultural machinery with a speed range of 0.5-2 m / s) overlaps with the robot path (based on corn rows with a row spacing of 0.7 m), the path generation module first uses the A* algorithm (grid resolution 0.1 m, heuristic function is Euclidean distance, weight 1.0) to generate alternative obstacle avoidance paths. Inputting the current robot pose (x, y, z, quaternion pose, accuracy 0.01 m) and the predicted obstacle trajectory (position error 0.05 m, speed error 0.1 m / s), the algorithm searches for neighboring grids starting from the current point, generating three alternative paths (maximum path length 10 m, radius of curvature > 0.5 m). The path with a minimum distance to the obstacle greater than 0.8 m is selected, with a computation time of approximately 0.02 seconds. Next, the navigation command adjustment module generates smooth navigation commands based on the selected path using cubic spline interpolation (control point interval 0.2 meters, smoothing factor 0.3), including velocity (0-1.5 m / s) and angular velocity (-0.5 to 0.5 radians / s), ensuring smooth robot movement along the corn rows. The obstacle avoidance decision logic fusion module receives navigation commands and real-time environmental perception data (LiDAR point cloud, range 10 meters, angular resolution 0.5°, frequency 20 Hz), and clusters the point cloud using the DBSCAN algorithm (neighborhood radius 0.2 meters, minimum number of points 5) to identify dynamic obstacles (diameter > 0.3 meters). If the distance between the obstacle and the path is less than 0.5 meters, the logic module uses weighted fusion (navigation command weight 0.6, perception data weight 0.4), adjusts the robot speed (reduced to 0.3 m / s), and replans the local path (update frequency 10 Hz). The final inter-row path is published via the ROS2 topic ( / navigation / path), containing the position sequence (accuracy 0.01 meters) and velocity commands, ensuring the robot navigates around obstacles and maintains its navigation along the corn row (slope error <0.1). The entire process is run through an embedded controller (such as a Raspberry Pi 4 with 4GB of memory), with data transmission latency controlled within 0.01 seconds, forming a closed-loop control.

[0031] S4: Generate a backup navigation sequence; In agricultural robot irrigation scenarios, a real-time path verification algorithm is used to determine whether the final row-to-row path meets the low-latency response requirement. This algorithm, based on an embedded controller (Jetson Nano, 8GB memory), processes sensor data (LiDAR, 15-meter scanning range, 0.3° angular resolution, 30Hz frequency) to calculate path delay. The algorithm takes the current path point sequence (0.02-meter accuracy) as input and combines it with real-time location data subscribed to from the ROS2 topic ( / irrigation / path). Through timestamp difference analysis (sampling interval 0.005 seconds), it determines whether the delay is less than 0.015 seconds. If the delay exceeds this limit, a fast iterative optimization algorithm is triggered. Using the RRT* algorithm (500 sampling points, 0.15-meter step size, maximum 1000 iterations), a new path (maximum length 8 meters, radius of curvature > 0.6 meters) is generated within 0.03 seconds, ensuring that the path points are aligned with the irrigation area (row spacing 0.65 meters) with an error < 0.08. Next, for the irrigation execution sequence, dynamic obstacle trajectories are predicted based on Kalman filtering (velocity range 0.3-1.8 m / s, position error 0.04 m), generating a sequence containing position (x, y, z, accuracy 0.01 m) and irrigation rate (0-2 L / s). Potential interference assessment is performed through Monte Carlo simulation (1000 samples, time window 0.5 sec), analyzing the collision probability (threshold < 0.1) between obstacles (diameter > 0.25 m) and the path. If the probability exceeds the threshold, a path adjustment mechanism is activated, based on gradient descent optimization (learning rate 0.01, 50 iterations), generating a backup sequence, adjusting the irrigation angle (0-45°) and robot speed (0.2-1.2 m / s) to ensure irrigation coverage > 95%. The final path and sequence are published via ROS2 (update frequency 15 Hz), with controller latency controlled within 0.008 seconds, achieving closed-loop irrigation control.

[0032] S5: Generate the execution path for the irrigation robot; In agricultural robot navigation scenarios, backup sequences and updated environmental map data are first acquired through multi-source sensor fusion. An RGB-D camera (1280x720 resolution, 20Hz frame rate, depth range 0.5-10 meters) and an ultrasonic sensor (0.2-5 meters detection distance, 10Hz frequency) are used to collect environmental information, generating point cloud data (0.05-meter accuracy). The point cloud is downsampled using VoxelGrid filtering (0.03-meter voxel size) and combined with a SLAM algorithm (Cartographer, localization accuracy 0.02 meters) to update the environmental map, achieving a map resolution of 0.1 meters. Backup sequences are extracted from a historical path database (storing 100 paths, 5-10 meters in length), and fused with the map data through timestamp matching (error <0.01 seconds) to generate decision inputs containing position (x, y, z, accuracy 0.01 meters) and velocity (0.5-1.5 m / s). Next, the decision input is processed by a low-power processor (Raspberry Pi 4, 4GB memory), and a lightweight convolutional neural network (MobileNetV3, 2.5M parameters, 0.02 seconds inference time) is used for obstacle classification to identify obstacle types (static obstacles such as rocks, dynamic obstacles such as animals, classification confidence > 0.9). The classification results are combined with Euclidean distance calculation (distance accuracy 0.01 meters) to determine whether the distance between the obstacle and the robot is less than the safety threshold of 1.2 meters. Navigation commands are generated using the A* algorithm (grid resolution 0.1 meters, maximum search depth 1000 nodes), and the output is a sequence of commands containing velocity (0.3-1.0 m / s) and direction (angular accuracy 5°). The command sequence is verified by a collision detection algorithm (based on a circular collider, robot radius 0.4 meters) to calculate the nearest distance to the obstacle (threshold > 1.0 meter). If the threshold is not met, path replanning is triggered, using the DWA algorithm (prediction time window 0.3 seconds, maximum angular velocity 30° / s) to adjust the path points (curvature radius > 0.8 meters). The final verified inter-row path was published via ROS2 (topic / nav / path, frequency 10Hz), ensuring that the robot navigates along a farmland path with a row spacing of 0.7 meters with an error of <0.06 meters, achieving safe and efficient path planning.

[0033] S6: Predicted results of the movement trajectory of the sudden animal; In agricultural robot navigation scenarios, when extracting real-time data features from validated row paths, a high-frequency LiDAR (sampling frequency 50Hz, angular resolution 0.2°, distance accuracy 0.02m) is used to scan the farmland environment, generating two-dimensional point cloud data (approximately 1000 points / frame). Principal component analysis (PCA) is used to extract texture features from the point cloud (feature dimension 10, retaining 95% variance after dimensionality reduction). Combined with time series analysis (sliding window 0.5 seconds), terrain smoothness is calculated (standard deviation <0.05m is considered smooth). To address muddy ground interference, an image processing accelerator (NVIDIA Jetson Nano, 1.5 TFLOPS computing power) is used to process RGB camera data (resolution 1920x1080, frame rate 30Hz). A bilateral filtering algorithm (spatial sigma 5 pixels, color sigma 0.1) is used to remove noise caused by mud, generating compensated visual data (brightness deviation <10%). The compensated visual data is used to calculate the robot's current pose (position error <0.03 m) through a pose calculation component (based on the EKF algorithm, fused with IMU data, angular velocity accuracy 0.1° / s). Combined with optical flow (feature point matching quantity 200, error <0.01 pixels), the dynamic trajectory is predicted (prediction time window 0.4 seconds). To determine the trajectory of sudden animal interference, a YOLOv5s model (7.2M parameters, inference time 0.015 seconds) is used to detect animal targets in the image (confidence >0.85), and the target bounding box is extracted (size accuracy 0.02 m). Kalman filtering (state vector dimension 6, process noise 0.01) is used to track animal movement and predict its trajectory in the next 0.5 seconds (velocity range 0-2 m / s, direction error <5°). If the distance between the intersection of the predicted trajectory and the robot path is less than the safety threshold of 1.5 meters, the path adjustment logic is triggered (based on the RRT algorithm, with 500 sampling points and a step size of 0.2 meters) to ensure that the robot avoids dynamic interference and maintains a row spacing of 0.8 meters in the farmland (error < 0.05 meters). All data is published through the ROS2 topic ( / nav / trajectory, frequency 20Hz) to ensure real-time performance.

[0034] S7: Generate safe irrigation execution commands; In the corn row navigation scenario, when the movement trajectory of a sudden animal disturbance overlaps with the validated row path, a candidate obstacle avoidance path is first generated using a deep reinforcement learning algorithm (DQN, state space dimension 12, action space dimension 4, learning rate 0.001). A local environment map of the corn rows is constructed using LiDAR data (sampling frequency 20Hz, angular resolution 0.5°, distance accuracy 0.05m), with a resolution of 0.1m / grid, generating a 100x100 grid. DQN calculates the obstacle avoidance path based on the current robot position (localization accuracy 0.04m) and animal trajectory prediction (velocity range 0-1.5m / s, prediction window 0.3s), prioritizing a path length less than 5m and a minimum distance greater than 1.2m from the animal trajectory. Next, the candidate paths are fused with the global path, and an adjusted navigation command is generated using the A* algorithm (heuristic function based on Euclidean distance, weight 1.2). The command update frequency is 10Hz, maintaining a row spacing of 0.75m (error <0.06m). Subsequently, navigation commands and environmental map updates were integrated, and Bayesian filtering (probabilistic grid update cycle of 0.2 seconds) was used to dynamically adjust the obstacle distribution in the map (confidence threshold of 0.9) to generate the final corn row path, ensuring that the distance between path points was less than 0.3 meters. Finally, an irrigation sequence optimization algorithm (based on a genetic algorithm, population size of 50, iterations of 100) was used to adjust the irrigation priority of path points. Combined with soil moisture sensor data (accuracy of 0.02%, sampling frequency of 1Hz), a safe irrigation execution sequence was generated with an irrigation time window of 5 seconds and a coverage rate of over 95%. All data was published through the ROS2 topic ( / irrigation / path, frequency 15Hz) to ensure real-time performance.

[0035] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A method for row navigation and obstacle avoidance of an irrigation robot, characterized in that, include: Step 1: Collect real-time field data, generate a preliminary environmental map, use a sensor fusion engine to integrate the preliminary environmental map, determine the current pose of the irrigation robot and obtain the real-time environmental map; Step 2: Extract local visual data based on the current pose of the irrigation robot, preprocess the local visual distance data, and use the pose calculation component to detect the preprocessed local visual data to obtain the motion trajectory of the dynamic obstacle. Step 3: If the trajectory of the dynamic obstacle overlaps with the path of the irrigation robot, an alternative obstacle avoidance path is generated through the path generation module and the navigation command is adjusted. The adjusted navigation command is corrected based on the environmental data in the real-time environmental map to generate the execution path of the irrigation robot. Step 4: If the execution path of the irrigation robot does not meet the low-latency response requirement, the path is updated through a fast iterative optimization algorithm to obtain a safe irrigation execution sequence. In the safe irrigation execution sequence, the dynamic obstacle trajectory is predicted and the collision probability is calculated. If the collision probability exceeds the threshold, the path is adjusted and a backup navigation sequence is generated. Step 5: Integrate the backup navigation sequence and real-time environmental map data and input them into the low-power processor to obtain the obstacle classification results. Use the navigation instruction generation module to verify the safe distance maintenance of the obstacle classification and recognition results to obtain the corrected execution path of the irrigation robot. Step 6: Extract the real-time environmental feature vector of the corrected execution path of the irrigation robot, compensate for the muddy ground interference in the real-time environmental feature vector, obtain the compensated visual feature data, and use the pose calculation component to perform dynamic trajectory prediction on the compensated visual feature data to obtain the prediction result of the movement trajectory of the sudden animal. Step 7: If the predicted trajectory of the sudden animal overlaps with the corrected execution path of the irrigation robot, then generate a set of alternative obstacle avoidance paths between the rows of the field based on the obstacle avoidance decision logic and determine the final navigation sequence between the rows of the field. Step 8: Convert the final field row navigation sequence into navigation instructions that the equipment can execute, and obtain safe irrigation execution instructions.

2. The irrigation robot's row navigation and obstacle avoidance method according to claim 1, characterized in that, The pose calculation component includes: Low-power processor, image processing accelerator, sensor combination and inertial measurement unit; The sensor array is used to collect real-time data on the field environment; The image processing accelerator is used to compensate for muddy ground interference in the environmental feature vector. The inertial measurement unit is used to collect inertial measurement unit data of the field environment; The pose calculation component is used to extract visual feature data and detect the extracted visual feature data.

3. The method for row navigation and obstacle avoidance of an irrigation robot according to claim 1, characterized in that, Step 1 specifically includes: Real-time data is collected from the field environment using a combination of sensors to obtain raw environmental data, which includes lidar point clouds, camera images, and inertial measurement unit data. The lidar point clouds are then denoised and registered, and the camera images are fused for feature extraction to generate a preliminary environmental map. Missing values ​​are detected in the preliminary environmental map. If there are missing data in the preliminary environmental map, the missing areas are filled in by interpolation to obtain a complete preliminary environmental map. The sensor fusion engine integrates the completed preliminary environmental map, and the inertial measurement unit data is fused to compensate for the GNSS signal to obtain an optimized environmental map. The current pose of the irrigation robot is determined by fusing the optimized environmental map and inertial measurement unit data using the Kalman filter algorithm. If the current pose deviation of the irrigation robot exceeds a first preset threshold, the pose is optimized using the particle filter algorithm to obtain a high-precision pose. The optimized environmental map is then updated based on the high-precision pose to obtain a real-time environmental map.

4. The method for row navigation and obstacle avoidance of an irrigation robot according to claim 1, characterized in that, Step 2 specifically includes: Based on the current pose of the irrigation robot, local visual data is extracted from the real-time data collected by the sensor combination. The local visual data is then denoised to obtain the first visual data. An image processing accelerator is used to correct for illumination changes in the first visual data to obtain the second visual data. The second visual data is used to extract features by a pose calculation component, and the structure of crop rows in the field is detected to obtain the crop distribution. If there are non-crop row areas in the crop distribution, the location of hard obstacles is determined by an edge detection algorithm. Based on the location of the hard obstacle, the second visual data is analyzed using an optical flow algorithm to obtain the motion trajectory of the dynamic obstacle.

5. The method for row navigation and obstacle avoidance of an irrigation robot according to claim 1, characterized in that, Step 3 specifically includes: Step 3.1: If the speed of the dynamic obstacle exceeds the second preset threshold, the Kalman filter algorithm is used to predict the future position of the dynamic obstacle, and the navigation path of the irrigation robot is updated according to the predicted position of the obstacle to generate dynamic navigation instructions. Step 3.2: If the trajectory of the dynamic obstacle overlaps with the updated navigation path of the irrigation robot, the alternative obstacle avoidance path is calculated through the path planning module to obtain a set of alternative paths; if they do not overlap, the updated navigation path of the irrigation robot is directly converted into a navigation command that the device can execute to obtain a safe irrigation execution command. Step 3.3: Use real-time data collected by the sensor combination to perform environmental constraint analysis on the candidate path set to obtain the filtered path set; Step 3.4: Integrate the filtered path set with the motion trajectory of the dynamic obstacle through the fusion algorithm to obtain the optimal obstacle avoidance path; if the optimal obstacle avoidance path meets the preset passage conditions, adjust the navigation path of the irrigation robot according to the optimal obstacle avoidance path; if it does not meet the conditions, repeat steps 3.1-3.3 until the preset passage conditions are met. Step 3.5: Use the original environmental data to perform real-time correction on the navigation path of the irrigation robot to obtain the corrected navigation instructions. Update the motion control parameters of the irrigation robot with the corrected navigation instructions to determine the final field row navigation instructions. Generate drive signals based on the final row navigation instructions to obtain the execution path of the irrigation robot.

6. The method for row navigation and obstacle avoidance of an irrigation robot according to claim 1, characterized in that, Step 4 specifically includes: If the execution path of the irrigation robot coincides with the movement trajectory of the dynamic obstacle, the path planning module calculates the set of alternative paths to obtain a preliminary path set. Environmental constraint analysis is performed on the initial path set using real-time sensor data to select an optimized path set that meets the passage conditions. The potential interference in the optimized path set is evaluated using a dynamic trajectory prediction algorithm to determine the optimal obstacle avoidance path. By integrating the optimal obstacle avoidance path with the collected real-time data, an adjusted irrigation sequence is generated. A fast iterative optimization algorithm is used to verify the low-latency response of the adjusted irrigation sequence to obtain a safe execution sequence. Drive signals are generated based on the safe execution sequence to determine the final irrigation control command. The final irrigation control command is dynamically corrected using real-time data collected by a combination of sensors to obtain a backup navigation sequence.

7. The irrigation robot's row navigation and obstacle avoidance method according to claim 1, characterized in that, Step 5 specifically includes: The backup navigation sequence and real-time environmental map data are integrated and input into a low-power processor for feature extraction and classification to obtain obstacle classification results; The obstacle classification results are input into the navigation instruction generation module to obtain the initial navigation instruction sequence. If the initial navigation instruction sequence overlaps with the obstacle positions in the obstacle classification results, the alternative paths are recalculated through the path adjustment module to obtain the adjusted navigation instruction sequence. The adjusted navigation command sequence is analyzed for distance constraints by the safety distance verification module to obtain a navigation command sequence that meets the safety distance requirement. The trajectory smoothing module is then used to optimize the navigation command sequence that meets the safety distance requirement to obtain a smooth inter-line path planning. The smooth inter-row path planning is dynamically corrected by a real-time feedback module to obtain the corrected execution path of the irrigation robot.

8. The method for row navigation and obstacle avoidance of an irrigation robot according to claim 1, characterized in that, Step 6 specifically includes: The real-time environmental feature vector of the irrigation robot's execution path is extracted by the feature extraction module. The muddy ground interference in the environmental feature vector is compensated by the image processing accelerator to obtain the compensated visual feature data. A pose calculation component is used to perform dynamic pose estimation on the compensated visual feature data to obtain dynamic pose parameters. The trajectory prediction and analysis module is then used to predict the motion trajectory based on the dynamic pose parameters to obtain the predicted motion trajectory data. Based on the predicted motion trajectory data, the corresponding sensor images are acquired. The YOLOv5s model is used to detect animal targets in the images and extract the bounding boxes of the animal targets. The animal movement is tracked by Kalman filtering, and the movement trajectory of sudden animals is predicted.

9. A method for row navigation and obstacle avoidance of an irrigation robot according to claim 1, characterized in that, Step 7 specifically includes: If the predicted trajectory of the sudden animal overlaps with the corrected execution path of the irrigation robot, the trajectory features of the overlapping area are extracted through the path analysis module to obtain the feature data of the overlapping area. The obstacle avoidance decision module processes the feature data of overlapping areas to generate alternative obstacle avoidance paths between rows of fields, thus obtaining a set of alternative paths. A path fusion algorithm is used to integrate the candidate path set with the environmental map data to obtain fused path data. The irrigation sequence optimization module then adjusts the sequence of the fused path data to obtain an optimized irrigation execution sequence. If the optimized irrigation execution sequence does not meet the preset obstacle safety distance, the path offset is calculated through the distance adjustment module to obtain the adjusted safe path data. The trajectory verification module was used to perform a feasibility analysis on the adjusted safe path data to obtain the final field row navigation sequence.

10. A navigation and obstacle avoidance system for an irrigation robot, applied to the navigation and obstacle avoidance method for an irrigation robot as described in any one of claims 1-9, characterized in that, include: The path planning module is used to plan the execution path of the irrigation robot, and inputs alternative obstacle avoidance paths and a preliminary path; The navigation instruction generation module is used to generate navigation instructions for the irrigation robot. The path adjustment module is used to adjust the execution path of the irrigation robot; The safety distance verification module is used to perform distance constraint analysis on the navigation commands of the irrigation robot; The trajectory smoothing module is used to optimize navigation command sequences that meet safety distance requirements; A real-time feedback module is used to dynamically correct smooth inter-line path planning; The feature extraction module is used to extract the real-time environmental feature vector of the execution path of the irrigation robot; The trajectory prediction and analysis module is used to predict the motion trajectory of the irrigation robot based on its dynamic position parameters. The path analysis module is used to extract trajectory features of the overlapping area between the execution path of the irrigation robot and the movement trajectory of the sudden animal. The obstacle avoidance decision module is used to analyze the trajectory characteristics of overlapping areas and generate alternative obstacle avoidance paths between field rows; The irrigation sequence optimization module is used to adjust the sequence of path data; The distance adjustment module is used to adjust the path offset of the irrigation execution sequence; The trajectory verification module is used to perform feasibility analysis on safe path data; An image processing accelerator is used to compensate for muddy ground interference in environmental feature vectors. A low-power processor is used to perform feature extraction and classification processing on integrated data to output obstacle classification results.

Images