Unmanned vehicle multi-modal motion planning escape method and device for complex terrain, equipment and medium

Abstract

The application provides a complex terrain-oriented unmanned vehicle multi-modal motion planning escape method, device, equipment and medium. The method comprises the following steps: acquiring unmanned vehicle bottom contour scanning point cloud data and collecting current torque feedback sequence data of each wheel; mapping the bottom contour scanning point cloud data to the current torque feedback sequence data to generate an environment contact stress distribution atlas; extracting the sinking depth scalar and the slip characteristic parameter of each wheel; determining the chassis three-dimensional physical lock boundary according to the sinking depth scalar and the slip characteristic parameter, configuring a multi-modal form reconstruction strategy containing a vehicle body pitch adjustment sequence and an independent wheel end pulse output sequence; issuing the vehicle body pitch adjustment sequence to the suspension controller and issuing the independent wheel end pulse output sequence to the hub motor controller according to the priority time sequence, so that the unmanned vehicle performs a peristaltic gait action to cross the three-dimensional physical lock boundary. The application improves the escape success rate.

Description

Technical Field

[0001] This application relates to the fields of autonomous driving and intelligent control technology, and more specifically, to a method, device, electronic device, and computer-readable storage medium for multimodal motion planning and obstacle avoidance of unmanned vehicles in complex terrain. Background Technology

[0002] With the rapid development of artificial intelligence and autonomous driving technologies, unmanned inspection equipment has been widely used in areas such as perimeter security in industrial parks, safety monitoring in production workshops, and protection of critical infrastructure. These unmanned inspection vehicles, equipped with autonomous navigation and environmental perception capabilities, typically need to be capable of all-weather operation, conducting unmanned inspections in complex outdoor environments. In actual outdoor inspection scenarios, unmanned vehicles often face complex terrains such as mud, sand, or rugged, uneven surfaces lacking structured roads, which places extremely high demands on their motion planning and terrain adaptability.

[0003] In existing autonomous vehicle motion planning schemes for complex terrain, a feedback adjustment mechanism based on a single wheel slip ratio is typically adopted. This scheme first uses onboard sensors to collect the rotational speed and actual vehicle speed of each wheel in real time to calculate the current wheel slip ratio. Then, the control unit compares and analyzes the data against a set slip ratio safety threshold to determine whether the vehicle is in a slipping or stuck state. Finally, for wheels in a slipping state, the output torque limit of the corresponding hub motor is adjusted to distribute local traction force.

[0004] However, this conventional obstacle avoidance control scheme has obvious technical defects. Due to the highly non-uniform surface physical and mechanical properties of complex outdoor terrain, relying solely on a single wheel-end slip ratio or torque adjustment cannot fully perceive the three-dimensional spatial interference between the vehicle chassis and the terrain, as well as the overall mechanical contact state. At the same time, when facing complex terrain with deep subsidence or severe local lock-up, the conventional scheme lacks global multimodal collaborative planning of vehicle attitude and wheel-end drive, and cannot generate an effective overall obstacle-crossing action to break the mechanical deadlock, causing the unmanned vehicle to easily get stuck in a continuous state in complex terrain. Summary of the Invention

[0005] This application provides a method, apparatus, electronic device, and computer-readable storage medium for multimodal motion planning and obstacle avoidance of unmanned vehicles in complex terrain, so as to at least alleviate the above-mentioned technical problems.

[0006] A multimodal motion planning method for autonomous vehicles to overcome difficulties in complex terrain includes the following steps: Acquire the point cloud data of the vehicle's undercarriage contour and collect the current torque feedback sequence data of each wheel of the vehicle; Mapping the vehicle underbody contour scan point cloud data onto the current torque feedback sequence data generates an environmental contact stress distribution map. The settlement depth scalar and slip characteristic parameters of each wheel were extracted from the environmental contact stress distribution map. The three-dimensional physical lock-up boundary of the chassis is determined based on the settlement depth scalar and slip characteristic parameters, and a multi-modal morphological reconstruction strategy corresponding to the three-dimensional physical lock-up boundary is configured. The multi-modal morphological reconstruction strategy includes the vehicle body pitch adjustment sequence and the independent wheel end pulse output sequence. According to the preset priority sequence, the vehicle pitch adjustment sequence is sent to the suspension controller, and the independent wheel-end pulse output sequence is sent to the wheel hub motor controller, so that the unmanned vehicle can perform a crawling gait action to cross the three-dimensional physical lock-up boundary.

[0007] Optionally, the step of mapping the vehicle underbody contour scan point cloud data onto the current torque feedback sequence data to generate an environmental contact stress distribution map includes: Identify 3D coordinate nodes in the scanned point cloud data of the vehicle's undercarriage contour; Obtain the hardware acquisition timestamp information from the current torque feedback sequence data; Based on the hardware acquisition timestamp information, torque load attributes are configured for the three-dimensional coordinate nodes, resulting in multiple enhanced three-dimensional coordinate nodes; Multiple enhanced 3D coordinate nodes are spatially topologically stitched together to construct an environmental contact stress distribution map.

[0008] Optionally, the steps of configuring torque load attributes for the three-dimensional coordinate nodes based on hardware acquisition timestamp information to obtain enhanced three-dimensional coordinate nodes include: Retrieve the laser echo reception time of the three-dimensional coordinate nodes; Compare the laser echo reception time with the hardware acquisition timestamp information; When the difference between the laser echo reception time and the hardware acquisition timestamp information is within the preset tolerance range, the corresponding instantaneous torque value is extracted. The instantaneous torque value is written as a node attribute into the corresponding three-dimensional coordinate node to obtain the enhanced three-dimensional coordinate node.

[0009] Optionally, the step of extracting the settlement depth scalar of each wheel from the environmental contact stress distribution map includes: Locate the first pixel coordinate representing the axle rotation center in the environmental contact stress distribution map; Locate the second pixel coordinate representing the actual terrain contact profile in the environmental contact stress distribution map; Calculate the vertical Euclidean distance from the first pixel coordinate to the second pixel coordinate; The settlement depth scalar is determined by subtracting the pre-stored standard tire radius constant from the vertical Euclidean distance.

[0010] Optionally, the steps for determining the three-dimensional physical lock-up boundary of the chassis based on the settlement depth scalar and slip characteristic parameters include: Determine whether the settlement depth scalar exceeds the preset chassis ground clearance safety threshold; If the settlement depth scalar exceeds the chassis ground clearance safety threshold, the corresponding wheel will be marked as a failed and trapped node. Obtain the three-dimensional position coordinates of all failed and trapped nodes; Construct the outer envelope geometric convex hull based on all three-dimensional position coordinates; The three-dimensional physical locked boundary of the chassis is determined based on the outer envelope geometric convex hull.

[0011] Optionally, the steps for configuring a multimodal morphological reconstruction strategy corresponding to the 3D physically locked boundary include: Calculate the projected area of ​​the outer geometric convex hull onto the horizontal plane; Calculate the percentage of the vehicle's projected area that is trapped relative to the total projected area of ​​the unmanned vehicle chassis; Based on the trapped ratio coefficient, a matching address is performed in the preset strategy mapping data table to determine the initial set of escape actions; Collect the current torque output margin of normally operating wheels that are not marked as failed or trapped nodes; The action amplitude parameters in the initial set of escape actions are pruned and restricted based on the current torque output margin in order to configure a multimodal morphological reconstruction strategy.

[0012] Optionally, the multimodal morphology reconstruction strategy includes the following methods for generating the vehicle body pitch adjustment sequence: Analyze the quadrant positions of the failed and trapped nodes in the chassis coordinate system; Determine the target suspension support strut that needs to be lifted based on its quadrant position; Generate a set of hydraulic stroke expansion commands for the target suspension support strut according to a preset step-increment rule; The vehicle pitch adjustment sequence is generated based on the hydraulic stroke expansion command set.

[0013] Optionally, the multimodal morphology reconstruction strategy includes the following methods for generating independent wheel-end pulse output sequences: Obtain the estimated value of the maximum ground adhesion force of the wheel corresponding to the edge of the three-dimensional physical lock boundary; Set a peak torque target that is greater than the estimated maximum ground adhesion value; A continuous torque control signal waveform with sinusoidal fluctuations is generated based on the target torque peak. Discretize and slice the continuous torque control signal waveform in the time dimension to generate independent wheel-end pulse output sequences.

[0014] Optionally, the steps to enable the autonomous vehicle to perform a crawling gait to cross a three-dimensional physical lockout boundary include: Execute a vehicle pitch adjustment sequence to make the unmanned vehicle body generate a longitudinal tilt angle and maintain the longitudinal tilt angle; Send independent wheel-end pulse output sequences to drive the locally trapped wheels to generate periodic reciprocating tangential vibrations, and monitor the change in the unmanned vehicle's displacement to get out of trouble in real time; When the change in displacement after escaping the obstacle exceeds the preset escape distance threshold, the transmission of the vehicle pitch adjustment sequence is stopped and the output of the independent wheel-end pulse output sequence is terminated.

[0015] A multimodal motion planning and obstacle avoidance device for unmanned vehicles tackling complex terrain includes: The data acquisition module is used to acquire the point cloud data of the vehicle's undercarriage contour scan and to collect the current torque feedback sequence data of each wheel of the unmanned vehicle. The map generation module is used to map the vehicle underside contour scan point cloud data onto the current torque feedback sequence data to generate an environmental contact stress distribution map. The feature extraction module is used to extract the settlement depth scalar and slip characteristic parameters of each wheel from the environmental contact stress distribution map. The strategy configuration module is used to determine the three-dimensional physical lock-up boundary of the chassis based on the settlement depth scalar and slip characteristic parameters, and to configure the multi-modal morphological reconstruction strategy corresponding to the three-dimensional physical lock-up boundary. The multi-modal morphological reconstruction strategy includes the vehicle body pitch adjustment sequence and the independent wheel end pulse output sequence. The motion execution module is used to send the vehicle pitch adjustment sequence to the suspension controller according to the preset priority sequence, and to send the independent wheel end pulse output sequence to the wheel hub motor controller, so that the unmanned vehicle can perform a creeping gait action to cross the three-dimensional physical lock-up boundary.

[0016] An electronic device includes: a memory, a processor, and an unmanned vehicle multimodal motion planning and escape program for complex terrain stored in the memory and executable on the processor, the unmanned vehicle multimodal motion planning and escape program for complex terrain being configured to implement the steps of any of the unmanned vehicle multimodal motion planning and escape methods for complex terrain described above.

[0017] A computer-readable storage medium stores a multimodal motion planning and escaping program for unmanned vehicles in complex terrain. When executed by a processor, the multimodal motion planning and escaping program for unmanned vehicles in complex terrain implements the steps of any of the above-described multimodal motion planning and escaping methods for unmanned vehicles in complex terrain.

[0018] The technical advantages of the technical solution provided in this application are: This application presents a multimodal motion planning and traction escape method for unmanned vehicles (UAVs) in complex terrain. Addressing the technical shortcomings of traditional control schemes, such as the inability to comprehensively perceive the three-dimensional spatial interference between the vehicle chassis and the terrain, and the tendency to become trapped in a persistent state, this method acquires the vehicle's underbody contour scanning point cloud data and collects the current wheel torque feedback sequence data. The point cloud data is mapped to the torque data to generate an environmental contact stress distribution map, from which a settlement depth scalar and slip characteristic parameters are extracted. Based on these two parameters, the three-dimensional physical lock-up boundary of the chassis is determined, effectively solving the problem of insufficient perception dimension caused by traditional schemes relying solely on wheel-end slip rate. Compared to traditional mechanisms that only monitor wheel speed, this application deeply integrates the three-dimensional spatial contour topography with real-time dynamic torque feedback, enabling a more dimensional and concrete representation of the mechanical interaction between the chassis and the complex terrain surface. It provides a high degree of completeness in defining the three-dimensional physical lock-up boundary of the traction source, offering a more reliable boundary basis for subsequent traction escape actions.

[0019] Based on the determined three-dimensional physical lock-up boundary, this application configures a multimodal morphological reconstruction strategy that includes a vehicle pitch adjustment sequence and an independent wheel-end pulse output sequence. These strategies are then distributed to the suspension controller and wheel hub motor controller according to their priority sequence, enabling the autonomous vehicle to perform a creeping gait to overcome the physical lock-up boundary. This solves the problem of traditional solutions lacking global multimodal collaborative planning and failing to effectively break mechanical lock-up. Traditional solutions, when faced with severe local lock-up, often only apply single wheel torque restrictions, failing to change the geometric interference state between the chassis and the ground. This application, however, changes the vehicle's spatial attitude (vehicle pitch adjustment) through the suspension controller, while simultaneously coordinating with the wheel hub motor controller to output pulse vibrations (independent wheel-end pulses), forming a multimodal creeping gait that coordinates spatial structural decoupling and transient breaking of contact surface friction. This combination fundamentally breaks down the chassis's jamming effect from the physical contact level, resulting in higher extrication efficiency compared to traditional single speed control strategies. It also effectively enables the autonomous vehicle to operate continuously in unstructured road terrain. Attached Figure Description

[0020] Figure 1 This application provides an embodiment of a multimodal motion planning and obstacle avoidance scenario for unmanned vehicles in complex terrain. Figure 2 This application provides an embodiment of a multimodal motion planning and obstacle avoidance method for unmanned vehicles in complex terrain. Figure 3 This application provides an embodiment of a multimodal motion planning and obstacle avoidance device for unmanned vehicles tackling complex terrain. Figure 4 An electronic device is described in an embodiment of this application; Figure 5 This is a computer-readable storage medium according to an embodiment of the present application. Detailed Implementation

[0021] like Figure 1 The image shown illustrates a multimodal motion planning and obstacle avoidance scenario for unmanned vehicles navigating complex terrain, as described in an embodiment of this application. Figure 2 As shown in the figure, this application provides an embodiment of a multimodal motion planning and obstacle avoidance method for unmanned vehicles in complex terrain. The method includes the following steps: Acquire the point cloud data of the undercarriage contour of the unmanned vehicle, and collect the current torque feedback sequence data of each wheel of the unmanned vehicle; The vehicle underside contour scan point cloud data is mapped onto the current torque feedback sequence data to generate an environmental contact stress distribution map. The settlement depth scalar and slip characteristic parameters of each wheel are extracted from the environmental contact stress distribution map. The three-dimensional physical lock-up boundary of the chassis is determined based on the settlement depth scalar and the slip characteristic parameters, and a multi-modal morphological reconstruction strategy corresponding to the three-dimensional physical lock-up boundary is configured. The multi-modal morphological reconstruction strategy includes a vehicle body pitch adjustment sequence and an independent wheel-end pulse output sequence. According to the preset priority sequence, the vehicle pitch adjustment sequence is sent to the suspension controller, and the independent wheel-end pulse output sequence is sent to the wheel hub motor controller, so that the unmanned vehicle performs a creeping gait action to cross the three-dimensional physical lock-up boundary.

[0022] Optionally, the step of mapping the vehicle underside contour scan point cloud data onto the current torque feedback sequence data to generate an environmental contact stress distribution map includes: Identify the three-dimensional coordinate nodes in the scanned point cloud data of the vehicle underside contour; Obtain the hardware acquisition timestamp information from the current torque feedback sequence data; Based on the hardware acquisition timestamp information, torque load attributes are configured for the three-dimensional coordinate nodes to obtain multiple enhanced three-dimensional coordinate nodes; Multiple enhanced three-dimensional coordinate nodes are spatially topologically stitched together to construct the environmental contact stress distribution map.

[0023] Preferably, the specific implementation process of "mapping the vehicle bottom contour scanning point cloud data onto the current torque feedback sequence data to generate an environmental contact stress distribution map" in this application is as follows: In unmanned inspection scenarios such as perimeter security, production workshop safety monitoring, and protection of important infrastructure, when the unmanned vehicle travels along a preset inspection route or an autonomously planned inspection route, it first uses a lidar, vision sensor, and multi-sensor obstacle avoidance component installed under the chassis or in front of the chassis to continuously scan the relative spatial contour between the vehicle bottom and the terrain to obtain vehicle bottom contour scanning point cloud data. The vehicle bottom contour scanning point cloud data is not aimed at identifying ordinary ground obstacles, but is used to express the three-dimensional spatial relationship of the passable gap under the chassis, the terrain undulations near the wheels, the protruding obstacles under the chassis, and the terrain contact profile. At the same time, the unmanned vehicle reads the current torque feedback sequence data from the hub motor controllers corresponding to each wheel. The current torque feedback sequence data is used to express the torque fluctuation state formed by the contact of each wheel with mud, sand, gravel, ditches, or rugged ground during the same inspection period. The vehicle undercarriage contour scanning point cloud data provides a three-dimensional spatial contact basis, and the current torque feedback sequence data provides a wheel-end torque load basis. Subsequently, the vehicle undercarriage contour scanning point cloud data and the current torque feedback sequence data are mapped according to the time correspondence and spatial correspondence, so that the environmental contact stress distribution map can simultaneously reflect the terrain geometric interference and wheel-end torque load changes, rather than relying solely on wheel speed or slip rate for extrication judgment.

[0024] Preferably, when identifying the three-dimensional coordinate nodes in the vehicle undercarriage contour scan point cloud data, the vehicle undercarriage contour scan point cloud data is first processed by point cloud frame reading, so that each frame of the vehicle undercarriage contour scan point cloud data corresponds to a scanning moment during the unmanned vehicle's inspection and driving process; then, the spatial points in the vehicle undercarriage contour scan point cloud data are normalized according to the unmanned vehicle chassis coordinate system, so that the front-rear position, left-right position, and up-down height of the spatial points all fall into the same chassis coordinate system. After completing the coordinate normalization process, the spatial points are judged for neighborhood continuity. Spatial points that can continuously represent the terrain profile under the vehicle undercarriage, the contact surface near the wheels, or the protruding obstacles under the chassis are retained as three-dimensional coordinate nodes; spatial points that are outside the effective scanning range of the vehicle undercarriage or whose distance from adjacent spatial points changes too abruptly are removed from the vehicle undercarriage contour scan point cloud data as invalid spatial points. The effective scanning range of the vehicle underside is determined by the unmanned vehicle chassis projection area, the wheel envelope area, and the passable gap under the chassis. The invalid spatial points are no longer involved in the subsequent configuration of torque load attributes. The resulting three-dimensional coordinate nodes not only retain the spatial morphology of the complex terrain under the vehicle but also enable the subsequent configuration of torque load attributes to be carried out around the spatial position related to the chassis's trapped state.

[0025] Preferably, after identifying the three-dimensional coordinate nodes, a node spatial description record is established for each three-dimensional coordinate node. This record includes the front-rear coordinates, left-right coordinates, and up-down coordinates of the three-dimensional coordinate node in the chassis coordinate system, the laser echo reception time, and the identifier of the corresponding scan frame. The front-rear coordinates are used to determine whether the three-dimensional coordinate node corresponds to terrain near the front axle, terrain near the rear axle, or terrain in the center of the chassis. The left-right coordinates are used to determine whether the three-dimensional coordinate node corresponds to the area near the left wheel, the area near the right wheel, or the central area of ​​the chassis. The up-down coordinates are used to express the height relationship of terrain protrusions or depressions relative to the chassis. The laser echo reception time and the corresponding scan frame identifier are used together to correspond with the hardware acquisition timestamp information. The node spatial description record is continuously read in subsequent spatial influence area division, hardware acquisition timestamp information retrieval, and torque load attribute configuration, ensuring that the three-dimensional coordinate nodes are not isolated spatial points but can establish a traceable mapping relationship with the current torque feedback sequence data in both time and space dimensions.

[0026] Preferably, when acquiring the hardware acquisition timestamp information in the current torque feedback sequence data, the wheel-end torque feedback record is first read from the hub motor controller corresponding to each wheel. The wheel-end torque feedback record includes the wheel identifier, torque sample value, sampling time, and sampling period record. Then, the sampling time is converted into a unified time base consistent with the laser echo reception time to form the hardware acquisition timestamp information. The hardware acquisition timestamp information is not a separate time marker, but is used to define the acquisition sequence and acquisition interval of each torque sample value in the current torque feedback sequence data. Since there is a temporal correspondence between the wheel torque change and the contact state of the terrain under the vehicle when the unmanned vehicle is inspecting at low speed in complex terrain or crawling in a stuck state, converting the sampling time into a unified time base can reduce the mapping misalignment caused by differences in sensor sampling frequency, hub motor controller feedback delay, and point cloud scanning frame rate. After the wheel-end torque feedback record is converted into hardware acquisition timestamp information, it continues to participate in the processing of the wheel-end torque time sequence, so that the hardware acquisition timestamp information and the torque sample value maintain a homologous correspondence.

[0027] Preferably, to enable the hardware acquisition timestamp information to participate in the attribute configuration of the three-dimensional coordinate nodes, the current torque feedback sequence data is processed by wheel-end organization. The torque sampling values ​​under the same wheel identifier, arranged in order of sampling time, are organized into a wheel-end torque time series corresponding to that wheel identifier, and the corresponding hardware acquisition timestamp information is retained in each wheel-end torque time series. Subsequently, based on the positional relationship of each wheel in the unmanned vehicle chassis structure relative to the chassis coordinate system, the three-dimensional coordinate nodes are divided into spatial influence regions adjacent to each wheel. These spatial influence regions do not change the coordinate values ​​of the three-dimensional coordinate nodes but are used to express which wheel identifier has a spatial affiliation relationship with the wheel's adjacent contact surface corresponding to the three-dimensional coordinate node. After completing the spatial influence region division, the three-dimensional coordinate node simultaneously has a node spatial description record and a wheel-end affiliation relationship. The wheel-end affiliation relationship is used to point the three-dimensional coordinate node to the corresponding wheel-end torque time series. Subsequently, the hardware acquisition timestamp information can be used to extract torque sampling values ​​matching the three-dimensional coordinate node from the corresponding wheel-end torque time series.

[0028] Preferably, when configuring torque load attributes for the three-dimensional coordinate nodes based on the hardware acquisition timestamp information, the node space description record of each three-dimensional coordinate node is first read. The laser echo reception time and wheel-end affiliation are extracted from the node space description record. Then, the hardware acquisition timestamp information adjacent to the laser echo reception time is retrieved from the wheel-end torque time sequence corresponding to the wheel-end affiliation. When the laser echo reception time falls between adjacent hardware acquisition timestamps, a time proximity interpolation method is used to determine the instantaneous torque value corresponding to the three-dimensional coordinate node. When the time difference between the laser echo reception time and a certain hardware acquisition timestamp is within a preset tolerance range, the torque sample value corresponding to that hardware acquisition timestamp is directly extracted as the instantaneous torque value. The time proximity interpolation method assigns time weights to adjacent torque sample values ​​based on the time interval between adjacent hardware acquisition timestamps and the laser echo reception time to form the instantaneous torque value corresponding to the laser echo reception time. The instantaneous torque value is then written as a torque load attribute into the corresponding three-dimensional coordinate node, transforming the three-dimensional coordinate node from a point cloud node that simply expresses spatial shape into an enhanced three-dimensional coordinate node that simultaneously expresses spatial position, wheel end affiliation, and wheel end torque loading state.

[0029] Preferably, the preset tolerance interval is configured based on both the point cloud scanning cycle record and the sampling cycle record. During configuration, a shorter sampling cycle is used as the basic time resolution, and an allowable time deviation range is set according to the vehicle displacement change when the unmanned vehicle travels at low speeds in complex terrain. The preset tolerance interval is used to limit the effective matching range between the laser echo reception time and the hardware acquisition timestamp information, reducing the erroneous binding of point cloud scanning results that are too far apart with torque sampling values. If a 3D coordinate node cannot find hardware acquisition timestamp information within the preset tolerance interval in the corresponding wheel-end torque time series, then the 3D coordinate node is marked as a 3D coordinate node to be compensated, and the torque load attribute compensation of the 3D coordinate node to be compensated is performed using enhanced 3D coordinate nodes with continuous spatial positions in adjacent scanning frames. After torque load attribute compensation, the 3D coordinate node to be compensated continues to participate in spatial topology stitching as an enhanced 3D coordinate node, thereby reducing the breakage of the environmental contact stress distribution pattern caused by short-term sampling gaps.

[0030] Preferably, the torque load attribute in the enhanced 3D coordinate nodes is not directly equivalent to terrain contact stress, but rather serves as a wheel-end torque load mapping quantity characterizing wheel-end torque load changes in subsequent map construction. Specifically, for multiple enhanced 3D coordinate nodes in the same spatial influence area, terrain height change records are first generated according to the vertical coordinates, and then wheel-end torque load change records are generated by combining the torque load attribute. Subsequently, the terrain height change records are correlated with the wheel-end torque load change records, so that terrain protrusions, terrain depressions, and wheel-adjacent contact surfaces can establish a contact load correspondence with torque increase, torque fluctuation, or torque mutation. The contact load correspondence continues to participate in the construction of the environmental contact stress distribution map, enabling the subsequent environmental contact stress distribution map to express the technical chain of wheel-end torque load changes caused by terrain topography changes, which differs from the conventional processing method of only using point cloud data as an obstacle avoidance map or only using the current torque feedback sequence data as a single drive feedback.

[0031] Preferably, when spatially stitching together multiple enhanced 3D coordinate nodes, firstly, the multiple enhanced 3D coordinate nodes are positioned in a two-dimensional grid according to the front-back and left-right coordinates in the chassis coordinate system, so that each enhanced 3D coordinate node is assigned to a corresponding chassis projection grid cell; then, within each chassis projection grid cell, the grid cell load characterization value is determined based on the up-down coordinates and torque load attributes of the enhanced 3D coordinate nodes. The grid cell load characterization value is formed by multiple enhanced 3D coordinate nodes within the same chassis projection grid cell and is used to express the degree of terrain contact undulation and wheel-end torque load at the chassis projection position. Through this stitching method, multiple discrete enhanced 3D coordinate nodes are organized into a chassis contact topology with spatial continuity. The chassis contact topology continues to serve as the spatial topological basis for the environmental contact stress distribution map and maintains the correspondence between the enhanced 3D coordinate nodes, chassis projection grid cells, and grid cell load characterization values ​​in subsequent continuity corrections.

[0032] Preferably, after forming the chassis contact topology, the load characterization values ​​of adjacent chassis projection grid cells are continuously corrected to suppress local spikes caused by uneven point cloud density or short-term wheel bounce. This continuous correction is not a simple averaging of all grid cell load characterization values, but rather first determines whether adjacent chassis projection grid cells have a continuous terrain profile relationship, and then performs neighborhood smoothing on the load characterization values ​​of grid cells with a continuous terrain profile relationship. For grid cell load characterization values ​​located at the edges of ditches, protruding rocks, or wheel settlement, their edge abrupt changes are retained, allowing the chassis contact topology to continue expressing the spatial boundaries causing chassis jamming. The chassis contact topology after continuous correction is further converted into an environmental contact stress distribution map. The spatial topological basis in the environmental contact stress distribution map is reflected in the topological connection relationship between each chassis projection grid cell, ensuring that the environmental contact stress distribution map has both spatial continuity and retains the edge abrupt changes related to the trapped state in complex terrain.

[0033] Preferably, the environmental contact stress distribution map includes the chassis projection position, terrain height status, torque load attributes, and wheel end attribution. The chassis projection position is derived from the front-to-back and left-to-right coordinates of the enhanced 3D coordinate node in the chassis coordinate system. The terrain height status is derived from the up-to-down coordinates of the enhanced 3D coordinate node. The torque load attributes are derived from the instantaneous torque value in the current torque feedback sequence data that matches the hardware acquisition timestamp information. The wheel end attribution is derived from the spatial influence area to which the 3D coordinate node belongs. Based on the above structure, each map unit in the environmental contact stress distribution map can be traced back to the corresponding enhanced 3D coordinate node, the corresponding hardware acquisition timestamp information, the corresponding current torque feedback sequence data, and the corresponding wheel end attribution. Therefore, when extracting the settlement depth scalar and slip characteristic parameters from the environmental contact stress distribution map, the terrain spatial morphology and wheel end torque load changes can be used simultaneously for judgment. The map unit, as the data unit bearing the chassis projection position, terrain height status, torque load attributes, and wheel end attribution in the environmental contact stress distribution map, continues to participate in the subsequent extraction of the settlement depth scalar and slip characteristic parameters.

[0034] Preferably, in an exemplary scenario of complex terrain inspection, when the unmanned vehicle travels to a muddy, sunken area, multiple 3D coordinate nodes in the vehicle's undercarriage contour scan point cloud data will present a concave terrain profile in the vicinity of the corresponding wheel. This vicinity, after spatial influence region division, forms a spatial influence region associated with the corresponding wheel identifier. The wheel-end torque time series corresponding to this spatial influence region in the current torque feedback sequence data will show torque increase and intensified fluctuations. At this time, after configuring torque load attributes for the 3D coordinate nodes based on the hardware acquisition timestamp information, the enhanced 3D coordinate nodes within the concave terrain profile will be assigned higher torque load attributes. After spatial topology stitching of multiple enhanced 3D coordinate nodes, the environmental contact stress distribution map will form a continuous high-load map region in the vicinity of the wheel. This high-load map region is formed by the continuous arrangement of multiple map units with higher torque load attributes. This high-load map region continues to participate in the extraction of settlement depth scalar and slip characteristic parameters, providing a basis from both spatial geometry and wheel-end torque load changes when subsequently determining the 3D physical lock-up boundary of the chassis.

[0035] Optionally, the step of configuring torque load attributes for the three-dimensional coordinate nodes based on the hardware-acquired timestamp information to obtain enhanced three-dimensional coordinate nodes includes: Retrieve the laser echo reception time of the three-dimensional coordinate nodes; Compare the laser echo reception time with the hardware acquisition timestamp information; When the difference between the laser echo reception time and the hardware acquisition timestamp information is within a preset tolerance range, the corresponding instantaneous torque value is extracted. The instantaneous torque value is written as a node attribute into the corresponding three-dimensional coordinate node to obtain the enhanced three-dimensional coordinate node.

[0036] Preferably, the specific implementation process of configuring torque load attributes for the three-dimensional coordinate nodes based on the hardware acquisition timestamp information is as follows: When the unmanned vehicle is in an outdoor inspection scenario such as perimeter security of a park, safety monitoring of a production workshop, or protection of important infrastructure, the established node spatial description record and the wheel-end torque time sequence that has been organized are read first; the node spatial description record comes from the three-dimensional coordinate nodes in the vehicle bottom contour scan point cloud data, and the node spatial description record includes the front-rear coordinates, left-right coordinates, and up-down coordinates of the three-dimensional coordinate nodes in the chassis coordinate system, the laser echo reception time, and the scan frame identifier to which they belong; the wheel-end torque time sequence comes from the current torque feedback sequence data, and the wheel-end torque time sequence includes multiple torque sample values ​​corresponding to the same wheel identifier and hardware acquisition timestamp information corresponding to each torque sample value. By first reading the node space description record and the wheel end torque time series, subsequent retrieval of the laser echo reception time, comparison of the laser echo reception time with the hardware acquisition timestamp information, extraction of the instantaneous torque value, and writing of the torque load attribute all have a clear data source, and establish a mapping basis based on time correspondence and wheel end affiliation between the three-dimensional coordinate nodes and the current torque feedback sequence data.

[0037] Preferably, when retrieving the laser echo reception time of the three-dimensional coordinate node, multiple three-dimensional coordinate nodes are first read sequentially according to their respective scan frame identifiers, and the corresponding node spatial description record is called when each three-dimensional coordinate node is read sequentially; then, the laser echo reception time is extracted from the node spatial description record and written into the node time retrieval record corresponding to the three-dimensional coordinate node. The node time retrieval record includes the three-dimensional coordinate node, its respective scan frame identifier, the laser echo reception time, and the wheel end affiliation relationship. The wheel end affiliation relationship is determined by the spatial influence area to which the three-dimensional coordinate node belongs. The spatial influence area is determined by the relationship between the front-rear coordinates and left-right coordinates of the three-dimensional coordinate node in the chassis coordinate system and the chassis mounting position of each wheel. Thus, the node time retrieval record binds the laser echo reception time and the wheel end affiliation relationship to the same three-dimensional coordinate node, so that when comparing the laser echo reception time with the hardware acquisition timestamp information, the retrieval can be performed directly within the wheel end torque time series pointed to by the wheel end affiliation relationship.

[0038] Preferably, after forming the node time retrieval record, based on the wheel-end attribution relationship in the node time retrieval record, a wheel-end torque time sequence corresponding to the wheel-end attribution relationship is selected from multiple wheel-end torque time sequences, and a time retrieval window is established in the selected wheel-end torque time sequence. The time retrieval window is centered on the laser echo reception time and bounded by a preset tolerance interval jointly determined by the point cloud scanning cycle record and the sampling cycle record; the point cloud scanning cycle record originates from the continuous scanning process of the vehicle undercarriage contour scanning point cloud data, and the sampling cycle record originates from the wheel-end sampling process of the current torque feedback sequence data. Through the time retrieval window, the hardware acquisition timestamp information participating in the subsequent comparison is limited to a range that has a time correspondence with the laser echo reception time, so that the three-dimensional coordinate nodes between different scanning frames will not cross the time retrieval window to read torque sampling values ​​of different time periods.

[0039] Preferably, when comparing the laser echo reception time with the hardware acquisition timestamp information, candidate hardware acquisition timestamp information is first read from the time retrieval window. This candidate hardware acquisition timestamp information originates from the hardware acquisition timestamp information falling within the time retrieval window in the wheel-end torque time series. Then, the time interval between the laser echo reception time and each candidate hardware acquisition timestamp is calculated, and the time interval is written into a time difference record. The time difference record includes a three-dimensional coordinate node, the laser echo reception time, the candidate hardware acquisition timestamp information, the torque sample value, and the time interval. This time difference record is not separately stored intermediate data, but is used to determine whether the time interval is within the preset tolerance range, thereby determining whether the three-dimensional coordinate node can extract the instantaneous torque value from the wheel-end torque time series corresponding to the wheel-end attribution relationship.

[0040] Preferably, when the time interval in the time difference record is within the preset tolerance range, the corresponding candidate hardware acquisition timestamp information is used as the valid hardware acquisition timestamp information, and the valid hardware acquisition timestamp information, the laser echo reception time, the three-dimensional coordinate node, and the torque sampling value are written into the valid time matching record. If the same three-dimensional coordinate node corresponds to multiple valid hardware acquisition timestamp information within the time retrieval window, the multiple valid hardware acquisition timestamp information are sorted in ascending order of the time interval, and the valid hardware acquisition timestamp information with the smaller time interval is selected to participate in the extraction of the instantaneous torque value. Through the valid time matching record, a traceable time correspondence can be established between the source of the instantaneous torque value extraction and the laser echo reception time of the three-dimensional coordinate node, and the valid hardware acquisition timestamp information, the torque sampling value, and the three-dimensional coordinate node maintain a common source correspondence in the subsequent torque load attribute configuration process.

[0041] Preferably, when the difference between the laser echo reception time and the hardware acquisition timestamp information is within a preset tolerance range, the specific process for extracting the corresponding instantaneous torque value is as follows: first, read the torque sample value corresponding to the effective hardware acquisition timestamp information according to the effective time matching record; then, write the torque sample value into the instantaneous torque value record according to the wheel end affiliation relationship to form the instantaneous torque value corresponding to the three-dimensional coordinate node. The instantaneous torque value record includes the three-dimensional coordinate node, wheel end affiliation relationship, effective hardware acquisition timestamp information, torque sample value, and instantaneous torque value. The instantaneous torque value record continues to participate in the configuration of the torque load attribute. Since the instantaneous torque value originates from the torque sample value in the current torque feedback sequence data, and the torque sample value has established a matching relationship with the laser echo reception time through the effective hardware acquisition timestamp information, the instantaneous torque value can express the wheel end torque load state of the wheel adjacent contact surface corresponding to the three-dimensional coordinate node near the scanning time.

[0042] Preferably, if the laser echo reception time is between two adjacent valid hardware acquisition timestamps, and both adjacent valid hardware acquisition timestamps are within the preset tolerance range, then the torque sampling values ​​corresponding to the two adjacent valid hardware acquisition timestamps are read respectively. Based on the time interval between the laser echo reception time and the two adjacent valid hardware acquisition timestamps, time proximity interpolation is performed on the two torque sampling values ​​to form the instantaneous torque value corresponding to the three-dimensional coordinate node. The time proximity interpolation does not change the current torque feedback sequence data itself, but rather forms the instantaneous torque value corresponding to the laser echo reception time in the instantaneous torque value record. The instantaneous torque value is then used to configure the torque load attribute, ensuring that even when the point cloud scanning time and the torque sampling time do not completely coincide, the three-dimensional coordinate node can still obtain the wheel-end torque load state corresponding to the laser echo reception time.

[0043] Preferably, if no hardware acquisition timestamp information within the preset tolerance range is found in the time retrieval window, the corresponding 3D coordinate node is marked as a 3D coordinate node to be compensated, and enhanced 3D coordinate nodes with spatial continuity with the 3D coordinate node to be compensated are retrieved in the previous and next scan frames. The spatial continuity is determined based on the proximity of the 3D coordinate node to be compensated and the enhanced 3D coordinate nodes in adjacent scan frames in the chassis coordinate system in the forward / backward, left / right, and up / down directions. When the enhanced 3D coordinate nodes in adjacent scan frames are in the same spatial influence area and the terrain height is continuous, the torque load attribute of the enhanced 3D coordinate nodes in adjacent scan frames is read, and torque load attribute compensation is performed on the 3D coordinate node to be compensated. After torque load attribute compensation, the 3D coordinate node to be compensated continues to participate in spatial topology stitching as an enhanced 3D coordinate node, so that short-term sampling loss will not directly cause data breakpoints in the subsequent environmental contact stress distribution map at the corresponding chassis projection position.

[0044] Preferably, when writing the instantaneous torque value as a node attribute into the corresponding 3D coordinate node, the node spatial description record, node time retrieval record, effective time matching record, and instantaneous torque value record of the 3D coordinate node are read first. Then, the instantaneous torque value in the instantaneous torque value record is written as a torque load attribute into the node attribute field of the 3D coordinate node, while simultaneously retaining the wheel end affiliation, the effective hardware acquisition timestamp information, and the corresponding scan frame identifier. After writing is completed, the 3D coordinate node is transformed into an enhanced 3D coordinate node, which includes front-back coordinates, left-right coordinates, up-down coordinates, laser echo reception time, corresponding scan frame identifier, wheel end affiliation, effective hardware acquisition timestamp information, and torque load attribute. Through this writing method, the enhanced 3D coordinate node simultaneously carries spatial position, laser echo reception time, effective hardware acquisition timestamp information, and wheel end torque load status, and can serve as a basic node for subsequent spatial topology stitching.

[0045] Preferably, after the torque load attribute is written, the enhanced 3D coordinate node undergoes attribute writing verification processing. This verification process reads the laser echo reception time, valid hardware acquisition timestamp information, wheel end affiliation, and torque load attribute from the enhanced 3D coordinate node. It then determines whether the valid hardware acquisition timestamp information originates from the wheel end torque time series corresponding to the wheel end affiliation, and whether the torque load attribute originates from the torque sampling value corresponding to the valid hardware acquisition timestamp information. If the enhanced 3D coordinate node passes the attribute writing verification processing, it is written into the enhanced 3D coordinate node sequence. If the enhanced 3D coordinate node fails the attribute writing verification processing, the process returns to reading the node time retrieval record, and the retrieval of candidate hardware acquisition timestamp information and the generation of time difference records are re-executed within the time retrieval window. The enhanced 3D coordinate node sequence continues to participate in the spatial topology stitching of multiple enhanced 3D coordinate nodes, enabling the subsequent construction of the environmental contact stress distribution map to read the enhanced 3D coordinate nodes that have already completed time correspondence and wheel end affiliation.

[0046] Preferably, when the unmanned vehicle traverses muddy ground, sandy edges, or gravelly ditches, the three-dimensional coordinate nodes in the vehicle underbody contour scan point cloud data will form a continuous undulating terrain profile near the wheel contact surface. The wheel-end torque time series in the current torque feedback sequence data will exhibit torque sampling value fluctuations corresponding to the wheel-end torque loading state during the same inspection period. At this time, by retrieving the laser echo reception time of the three-dimensional coordinate nodes, comparing the laser echo reception time with the hardware acquisition timestamp information, extracting the corresponding instantaneous torque value, and writing the instantaneous torque value as a torque load attribute into the corresponding three-dimensional coordinate node, the three-dimensional coordinate nodes within the continuous undulating terrain profile can be transformed into enhanced three-dimensional coordinate nodes with wheel-end torque loading states. Multiple enhanced three-dimensional coordinate nodes can collectively express the correspondence between the wheel-adjacent contact surface and the wheel-end torque loading state in subsequent spatial topology stitching, thereby providing a foundational node for further calculation of the environmental contact stress distribution map.

[0047] Preferably, in all-weather complex terrain inspection scenarios, the vehicle undercarriage contour scanning point cloud data collected by lidar, vision sensors, and multi-sensor obstacle avoidance components are affected by changes in illumination, differences in ground reflection, and local occlusion. The current torque feedback sequence data is also affected by short-term wheel bounce and abrupt changes in terrain contact. These changes in illumination, differences in ground reflection, local occlusion, short-term wheel bounce, and abrupt changes in terrain contact collectively form a fluctuating acquisition state. To adapt the enhanced 3D coordinate nodes to this fluctuating acquisition state, when configuring the torque load attributes, an arbitrary torque sample value is not directly written to an arbitrary 3D coordinate node. Instead, the node time retrieval record generation, time retrieval window establishment, time difference record generation, effective time matching record generation, instantaneous torque value record generation, and attribute writing verification processing are performed sequentially. The above processing enables the torque load attribute to be formed along the technical chain of "three-dimensional coordinate nodes in the vehicle bottom contour scanning point cloud data, wheel end torque time series in the current torque feedback sequence data, hardware acquisition timestamp information, instantaneous torque value, and enhanced three-dimensional coordinate nodes", thereby enabling the subsequent environmental contact stress distribution map to express the correlation between the chassis spatial morphology and the wheel end torque loading state under complex terrain.

[0048] Optionally, the step of extracting the settlement depth scalar of each wheel from the environmental contact stress distribution map includes: Locate the first pixel coordinate representing the axle rotation center in the environmental contact stress distribution map; Locate the second pixel coordinate representing the actual terrain contact profile in the environmental contact stress distribution map. Calculate the vertical Euclidean distance from the first pixel coordinate to the second pixel coordinate; The settlement depth scalar is determined by subtracting the pre-stored standard tire radius constant from the vertical Euclidean distance.

[0049] Preferably, the specific implementation process for extracting the settlement depth scalar of each wheel from the environmental contact stress distribution map is as follows: When the unmanned vehicle enters muddy, sandy, gravelly ditch, or rugged terrain along the inspection route, the previously constructed environmental contact stress distribution map is read first. The environmental contact stress distribution map includes the chassis projection position, terrain height status, torque load attribute, and wheel end attribution relationship. The chassis projection position is used to express the front-rear and left-right positions of the map unit in the chassis coordinate system. The terrain height status is used to express the height relationship between the actual terrain contact profile corresponding to the map unit and the chassis. The torque load attribute is used to express the wheel end torque load state associated with the map unit. The wheel end attribution relationship is used to express the spatial correspondence between the map unit and the corresponding wheel. By reading the environmental contact stress distribution map, the terrain height status and wheel end torque load status of the area adjacent to the wheel can be obtained simultaneously in the same data representation. This allows the subsequent positioning of the first pixel coordinate representing the axle rotation center and the second pixel coordinate representing the actual terrain contact profile to no longer rely solely on ordinary image edge positioning, but rather to define the positioning range jointly by the chassis projection position, the terrain height status, the torque load attribute, and the wheel end attribution relationship.

[0050] Preferably, before locating the first pixel coordinate representing the axle rotation center in the environmental contact stress distribution map, the unmanned vehicle chassis parameter record and wheel installation position record are read first. The unmanned vehicle chassis parameter record is derived from the chassis structure data configured before the unmanned vehicle's inspection operation. The chassis structure data is used to record the structural positional relationship between the unmanned vehicle chassis and each wheel, and is used to form the unmanned vehicle chassis parameter record and the wheel installation position record. The wheel installation position record is derived from the installation position relationship of each wheel relative to the chassis coordinate system. The unmanned vehicle chassis parameter record includes the coordinate origin position of the chassis coordinate system, the chassis front-rear direction reference, the chassis left-right direction reference, and the chassis up-down direction reference. The wheel installation position record includes the wheel identification of each wheel, the front-rear direction coordinate, left-right direction coordinate, and up-down direction coordinate of the axle rotation center relative to the chassis coordinate system. Subsequently, based on the map scale mapping record of the environmental contact stress distribution map, the front-rear, left-right, and up-down coordinates of the axle rotation center relative to the chassis coordinate system are transformed into the pixel scale space of the environmental contact stress distribution map to form first pixel coordinate candidate values ​​corresponding to each wheel. The map scale mapping record originates from the scale correspondence between the chassis coordinate system and the pixel scale space during the construction of the environmental contact stress distribution map. The first pixel coordinate candidate values ​​continue to participate in the first pixel coordinate positioning, so that the first pixel coordinates have a structural and scale source from the unmanned vehicle chassis parameter record, the wheel installation position record, and the map scale mapping record.

[0051] Preferably, when locating the first pixel coordinates, the candidate value of the first pixel coordinates is not directly used as the final first pixel coordinates. Instead, the wheel end attribution relationship associated with the corresponding wheel is read, and a wheel-adjacent map region consistent with the wheel end attribution relationship is retrieved in the environmental contact stress distribution map. The wheel-adjacent map region is composed of multiple map units with the same wheel end attribution relationship, and the map units in the wheel-adjacent map region simultaneously carry the chassis projection position, terrain height status, and torque load attributes. Subsequently, the first pixel coordinate candidate value is spatially overlapped with the wheel-adjacent map region. If the first pixel coordinate candidate value falls into the wheel-adjacent map region or is adjacent to the central region of the wheel-adjacent map region, the first pixel coordinate candidate value is used as the first pixel coordinate representing the axle rotation center. If there is an offset between the first pixel coordinate candidate value and the wheel-adjacent map region, the position of the first pixel coordinate candidate value is corrected according to the chassis projection position of the wheel-adjacent map region to form the first pixel coordinate representing the axle rotation center. Therefore, the first pixel coordinates are both derived from the wheel installation position record and constrained by the wheel end attribution relationship and the wheel adjacent map region in the environmental contact stress distribution map. This can reduce the problem of using a fixed position after the axle rotation center projection shifts due to changes in vehicle posture under complex terrain.

[0052] Preferably, when locating the second pixel coordinate representing the actual terrain contact profile in the environmental contact stress distribution map, the first pixel coordinate is first used as the longitudinal reference position, and a vertical search line is established in the environmental contact stress distribution map along the chassis vertical direction reference. The vertical search line passes through the wheel-adjacent map area where the first pixel coordinate is located and passes through multiple map units related to the corresponding wheel contact. Subsequently, the terrain height state and torque load attributes of each map unit are read along the vertical search line, and the candidate position of the terrain contact profile is determined according to the continuous change of the terrain height state. The candidate position of the terrain contact profile is used to express the pixel position that can represent the actual terrain contact profile in the map area below the wheel or adjacent to the wheel. Since the candidate position of the terrain contact profile comes from the map unit in the environmental contact stress distribution map, and the map unit comes from the spatial topology stitching of the enhanced three-dimensional coordinate nodes, the candidate position of the terrain contact profile can maintain spatial consistency with the enhanced three-dimensional coordinate nodes in the preceding vehicle undercarriage contour scan point cloud data and continue to participate in the determination of the second pixel coordinate.

[0053] Preferably, when determining the second pixel coordinates, the candidate positions of the terrain contact profile are further processed by torque load attribute filtering, so that the second pixel coordinates not only represent the map position with the lowest geometric height or closest to the outer edge of the wheel, but also the map position corresponding to the actual contact load of the wheel. Specifically, the torque load attribute change state is first read from the vertical search line and its adjacent map units, and then the candidate positions of the terrain contact profile where the terrain height state is continuously concave or convex and the torque load attribute change state is in the wheel end torque load enhancement state are determined as the second pixel coordinates representing the actual terrain contact profile. If there are multiple candidate positions of terrain contact profiles in the same wheel adjacent map area, the candidate position of terrain contact profile that has the same wheel end affiliation relationship as the first pixel coordinates and corresponds more to the torque load attribute change state is preferentially selected as the second pixel coordinates. Through the torque load attribute filtering process, the second pixel coordinates are derived from the combined limitation of the terrain height state and the torque load attribute change state, which is different from the processing method of selecting the lowest point from the ordinary ground contour. It is more suitable for the actual terrain contact profile positioning of unmanned vehicles when local settlement occurs in mud, sand or gravel ditches.

[0054] Preferably, before calculating the vertical Euclidean distance from the first pixel coordinate to the second pixel coordinate, the scale mapping record of the environmental contact stress distribution map is first read. This scale mapping record expresses the correspondence between the actual length scale in the chassis coordinate system and the pixel scale space in the environmental contact stress distribution map. Then, the first pixel coordinate and the second pixel coordinate are placed under the same scale mapping record for coordinate scale unification, ensuring that the first pixel coordinate and the second pixel coordinate are at the same pixel scale or the same actual length scale. After coordinate scale unification, the vertical coordinate difference between the first pixel coordinate and the second pixel coordinate is read along the chassis vertical reference, and the vertical Euclidean distance is determined based on this vertical coordinate difference. Since the vertical Euclidean distance is calculated under the same scale mapping record, when the vertical Euclidean distance is subsequently used to perform difference processing with the standard tire radius constant, both are in a consistent scale expression, avoiding the problem of inconsistent dimensions caused by directly subtracting the pixel distance from the actual length value.

[0055] Preferably, the standard tire radius constant is pre-stored in the unmanned vehicle tire parameter record, which is established before the unmanned vehicle inspection task is configured based on the wheel model, tire outer diameter parameters, and wheel hub installation parameters. The unmanned vehicle tire parameter record includes the standard tire radius constant corresponding to each wheel identifier, the scale type of the standard tire radius constant, and the scale conversion relationship with the map scale mapping record. The standard tire radius constant is not estimated temporarily during the settlement judgment stage, but is pre-stored along with the chassis structure data before the unmanned vehicle performs the inspection operation. When the environmental contact stress distribution map is expressed in pixel scale, the map scale mapping record is read and the standard tire radius constant in the unmanned vehicle tire parameter record is converted into a map scale standard tire radius constant consistent with the environmental contact stress distribution map. When the vertical Euclidean distance has been converted to the actual length scale, the standard tire radius constant corresponding to the actual length scale in the unmanned vehicle tire parameter record is read. Through the cooperation of the unmanned vehicle tire parameter record and the map scale mapping record, the standard tire radius constant can participate in the determination of the settlement depth scalar at the same scale as the vertical Euclidean distance.

[0056] Preferably, when determining the settlement depth scalar by subtracting the pre-stored standard tire radius constant from the vertical Euclidean distance, the standard tire radius constant under the corresponding wheel identifier is first read according to the wheel end affiliation relationship. Then, the difference between the vertical Euclidean distance corresponding to the first pixel coordinate and the second pixel coordinate and the standard tire radius constant under the wheel identifier is processed to obtain the settlement candidate value of the corresponding wheel. If the settlement candidate value is greater than the preset non-settlement tolerance deviation, the settlement candidate value is used as the settlement depth scalar of the wheel; if the settlement candidate value is within the non-settlement tolerance deviation, the settlement depth scalar of the wheel is recorded as a state value of no effective settlement. The non-settlement tolerance is pre-configured based on the scanning noise range of the vehicle bottom contour scanning point cloud data, the map scale mapping record of the environmental contact stress distribution map, and the normal tire compression deformation range. The non-settlement tolerance is used to distinguish between normal tire contact compression and wheel settlement under complex terrain. The state value of not forming effective settlement and the settlement depth scalar are written together into the subsequent wheel settlement characterization record, so that the settlement depth scalar can express the degree of wheel sinking beyond normal contact deformation.

[0057] Preferably, when multiple second pixel coordinates exist within the same wheel proximity map region, the vertical Euclidean distance of each second pixel coordinate relative to the same first pixel coordinate is calculated, and the difference between each vertical Euclidean distance and the corresponding standard tire radius constant is processed to form multiple settlement candidate values. Subsequently, based on the torque load attribute change state corresponding to each second pixel coordinate, the multiple settlement candidate values ​​undergo consistency screening. When the second pixel coordinates corresponding to multiple settlement candidate values ​​are all located within the same continuous undulating terrain profile, and the torque load attribute change state corresponding to multiple settlement candidate values ​​shows a continuous increasing load trend, the multiple settlement candidate values ​​are fused according to the spatial continuity relationship within the wheel proximity map region to form a settlement depth scalar for that wheel. The consistency screening process uses the torque load attribute change state and the spatial continuity relationship to jointly determine whether multiple settlement candidate values ​​originate from the same wheel settlement region. The settlement depth scalar is formed by combining multiple settlement candidate values ​​with the torque load attribute change state; therefore, the settlement depth scalar can reduce the situation where settlement judgment is triggered only by a single abnormal pixel location or a single discrete terrain point.

[0058] Preferably, after the settlement depth scalar is determined, it is further correlated with the wheel end attribution, first pixel coordinates, second pixel coordinates, vertical Euclidean distance, and standard tire radius constant in the environmental contact stress distribution map to form a wheel settlement characterization record. The wheel settlement characterization record includes wheel identification, first pixel coordinates, second pixel coordinates, vertical Euclidean distance, standard tire radius constant, non-settlement tolerance, state value indicating no effective settlement, and the settlement depth scalar. The wheel settlement characterization record continues to participate in the subsequent determination of the chassis's three-dimensional physical lock-up boundary based on the settlement depth scalar and the slip characteristic parameters, ensuring that the settlement depth scalar is not merely a numerical result but can be traced back to a specific wheel-adjacent map region and a specific actual terrain contact profile in the environmental contact stress distribution map. Through the wheel settlement characterization record, when subsequently determining whether the settlement depth scalar exceeds a pre-set chassis ground clearance safety threshold, the settlement depth scalar, first pixel coordinates, second pixel coordinates, and wheel-adjacent map region related to the corresponding wheel can be directly read.

[0059] Preferably, when the unmanned vehicle passes over the edge of the sand, the environmental contact stress distribution map under a certain wheel will form a continuously concave terrain height state in the wheel's adjacent map region. Simultaneously, the torque load attribute corresponding to that wheel will show an enhanced change corresponding to the wheel-end torque loading state. At this time, based on the wheel's installation position record, the first pixel coordinate representing the axle rotation center is located, and the second pixel coordinate representing the actual terrain contact profile is located in the wheel's adjacent map region along the chassis's vertical reference. Subsequently, the vertical Euclidean distance between the first and second pixel coordinates is calculated based on the map scale mapping record, and the standard tire radius constant corresponding to the wheel identifier is read and processed for difference to obtain the wheel's settlement depth scalar. This settlement depth scalar is further written into the wheel settlement characterization record and subsequently participates in the determination of the three-dimensional physical locked boundary along with slip characteristic parameters, ensuring that the wheel sinking caused by the sand edge maintains a traceable correspondence between the wheel's adjacent map region, the first pixel coordinate, the second pixel coordinate, the vertical Euclidean distance, and the wheel-end torque loading state.

[0060] Preferably, in all-weather complex terrain inspection scenarios, the vehicle undercarriage contour scanning point cloud data collected by lidar, vision sensors, and multi-sensor obstacle avoidance components will experience local map fluctuations due to changes in illumination, differences in ground reflection, or partial occlusion. These local map fluctuations will manifest as short-term jumps in the terrain height state of some map units in the environmental contact stress distribution map. To reduce the impact of these local map fluctuations on the settlement depth scalar, when locating the second pixel coordinates, the terrain height state of a single map unit is not used as the sole basis. Instead, the second pixel coordinates are determined by combining the continuity of the terrain height state of adjacent map units within the wheel's adjacent map region and the change in torque load attributes. Subsequently, the vertical Euclidean distance obtained based on the second pixel coordinates is further processed by subtracting it from the standard tire radius constant, and normal tire contact compression and wheel settlement are distinguished by non-settlement tolerance deviation. The resulting settlement depth scalar can maintain consistency with the spatial continuity in the environmental contact stress distribution map and provide input basis corresponding to the complex terrain contact state for subsequent three-dimensional physical lock-up boundary judgment of the chassis.

[0061] Optionally, the step of determining the three-dimensional physical lock-up boundary of the chassis based on the settlement depth scalar and the slip characteristic parameters includes: Determine whether the settlement depth scalar exceeds the preset chassis ground clearance safety threshold; If the settlement depth scalar exceeds the chassis ground clearance safety threshold, the corresponding wheel is marked as a failed and trapped node. Obtain the three-dimensional position coordinates of all the failed and trapped nodes; Construct the outer envelope geometric convex hull based on all the aforementioned three-dimensional position coordinates; The three-dimensional physical lock-up boundary of the chassis is determined based on the outer envelope geometric convex hull.

[0062] Preferably, the specific implementation process of determining the three-dimensional physical locking boundary of the chassis based on the settlement depth scalar and the slip characteristic parameters is as follows: When the unmanned vehicle is inspecting muddy, sandy, gravelly, or uneven ground, the wheel settlement characterization record, environmental contact stress distribution map, and slip characteristic parameters generated in the previous step are read first. The wheel settlement characterization record includes wheel identification, first pixel coordinates, second pixel coordinates, vertical Euclidean distance, standard tire radius constant, non-settlement allowable deviation, and settlement depth scalar. The environmental contact stress distribution map includes chassis projection position, terrain height status, torque load attributes, and wheel end attribution relationship. The slip characteristic parameters are used to express the tangential slip state of the corresponding wheel during contact with complex terrain. By simultaneously reading the wheel settlement characterization record, the environmental contact stress distribution map, and the slip characteristic parameters, when subsequently determining whether the settlement depth scalar exceeds the preset chassis ground clearance safety threshold, it is possible to simultaneously obtain the wheel sinking degree, the wheel end torque load state in the wheel adjacent map area, and the corresponding wheel tangential slip state, so that the marking of the failed and trapped node has data sources from the settlement depth scalar, the torque load attribute, and the slip characteristic parameters.

[0063] Preferably, the chassis ground clearance safety threshold is preset before the unmanned vehicle inspection task is configured and stored in the chassis passage parameter record. The chassis passage parameter record is derived from the unmanned vehicle chassis parameter record, wheel mounting position record, unmanned vehicle tire parameter record, and suspension travel parameter record. The unmanned vehicle chassis parameter record provides the height position of the chassis lower surface relative to the chassis coordinate system, the wheel mounting position record provides the positional relationship of the axle rotation center of each wheel relative to the chassis coordinate system, the unmanned vehicle tire parameter record provides the standard tire radius constant, and the suspension travel parameter record provides the range of chassis height variation under normal inspection posture. Based on the unmanned vehicle chassis parameter record, the wheel mounting position record, the unmanned vehicle tire parameter record, and the suspension travel parameter record, the critical settlement amount that allows the unmanned vehicle to maintain a passable clearance under the chassis when passing through terrain protrusions or partial wheel depressions under normal inspection posture is first determined, and then the critical settlement amount is written into the chassis passage parameter record as the chassis ground clearance safety threshold. The ground clearance safety threshold is formed by the unmanned vehicle chassis parameter records, the wheel installation position records, the unmanned vehicle tire parameter records, and the suspension travel parameter records, and is subsequently compared with the settlement depth scalar on the same scale.

[0064] Preferably, before determining whether the settlement depth scalar exceeds a pre-set chassis ground clearance safety threshold, the atlas scale mapping record in the wheel settlement characterization record and the chassis ground clearance safety threshold in the chassis traffic parameter record are read first, and the settlement depth scalar and the chassis ground clearance safety threshold are expressed at the same scale. If the settlement depth scalar originates from the pixel scale space of the environmental contact stress distribution map, the settlement depth scalar is converted into an actual length scale consistent with the chassis ground clearance safety threshold according to the atlas scale mapping record; if the settlement depth scalar is already at the actual length scale, the chassis ground clearance safety threshold is directly read for comparison. After scale unification is achieved through the atlas scale mapping record, the settlement depth scalar and the chassis ground clearance safety threshold are then compared to reduce the judgment deviation caused by direct comparison of values ​​at different scales. The result of the size comparison is then used to form a settlement exceedance determination record.

[0065] Preferably, when determining whether the settlement depth scalar exceeds a preset chassis ground clearance safety threshold, the corresponding wheel settlement characterization record is read one by one according to the wheel identification, and the settlement depth scalar, non-settlement allowable deviation, first pixel coordinate, and second pixel coordinate are read from the wheel settlement characterization record; then, the settlement depth scalar is compared with the chassis ground clearance safety threshold, and the non-settlement allowable deviation is used to determine whether the settlement depth scalar has exceeded the normal tire contact compression range. If the settlement depth scalar exceeds the chassis ground clearance safety threshold, and the settlement depth scalar has exceeded the normal tire contact compression range corresponding to the non-settlement allowable deviation, a settlement boundary violation determination record is formed; the settlement boundary violation determination record includes wheel identification, settlement depth scalar, chassis ground clearance safety threshold, non-settlement allowable deviation, first pixel coordinate, and second pixel coordinate. The settlement boundary violation determination record continues to participate in the failure and entrapment node marking, so that there is a direct data connection relationship between the settlement depth scalar and the failure and entrapment node.

[0066] Preferably, if the settlement depth scalar exceeds the chassis ground clearance safety threshold, the corresponding wheel is marked as a failed and trapped node. During marking, not only the wheel identifier is saved, but the wheel end attribution relationship in the settlement boundary violation determination record, the slippage characteristic parameters, and the environmental contact stress distribution map is read and written into the failed and trapped node marking record. The failed and trapped node marking record includes the wheel identifier, settlement depth scalar, chassis ground clearance safety threshold, slippage characteristic parameters, wheel end attribution relationship, first pixel coordinates, and second pixel coordinates. If the settlement depth scalar of a wheel exceeds the chassis ground clearance safety threshold, and the corresponding slip characteristic parameter indicates that the wheel is in a continuous tangential slip state within the adjacent map region, then the failure-trapped node marker record corresponding to that wheel is written into the failure-trapped node marker record that prioritizes participation in boundary construction. If the settlement depth scalar of a wheel exceeds the chassis ground clearance safety threshold but the slip characteristic parameter does not indicate a continuous tangential slip state, then the failure-trapped node marker record corresponding to that wheel is still retained and participates in the calculation as the failure-trapped node marker record corresponding to the settlement boundary node during subsequent outer envelope geometric convex hull construction. Through this marking method, the slip characteristic parameter does not replace the boundary judgment of the settlement depth scalar, but rather, together with the settlement depth scalar, determines the participation of failure-trapped nodes in the outer envelope geometric convex hull construction and the determination of three-dimensional physical locked boundaries.

[0067] Preferably, when obtaining the three-dimensional position coordinates of all the failed and trapped nodes, the wheel identifier, first pixel coordinate, second pixel coordinate, and wheel end attribution relationship are first read from the failed and trapped node marking record. Then, the first pixel coordinate and second pixel coordinate are transformed into the chassis coordinate system according to the map scale mapping record to form the three-dimensional position coordinates corresponding to the failed and trapped node. The three-dimensional position coordinates include front-back coordinates, left-right coordinates, and up-down coordinates; wherein, the front-back coordinates and left-right coordinates are derived from the correspondence between the first pixel coordinate and the second pixel coordinate in the chassis projection position, and the up-down coordinates are derived from the terrain height state corresponding to the second pixel coordinate and the settlement depth scalar of the failed and trapped node. The three-dimensional position coordinates are then written into the failed and trapped node position record, which includes the failed and trapped node, three-dimensional position coordinates, settlement depth scalar, slip characteristic parameters, and wheel end attribution relationship, so that the three-dimensional position coordinates, settlement depth scalar, slip characteristic parameters, and wheel end attribution relationship of all failed and trapped nodes can be directly read when constructing the outer envelope geometric convex hull.

[0068] Preferably, after forming the location record of the failed and trapped node, the three-dimensional position coordinates of all failed and trapped nodes are first spatially consistent. This spatial consistency process includes unifying all three-dimensional position coordinates to the chassis coordinate system, merging repeated three-dimensional position coordinates within the adjacent map region of the same wheel, and retaining the three-dimensional position coordinates corresponding to a larger settlement depth scalar and a continuous tangential sliding state as the boundary representative points of the adjacent map region of that wheel. For multiple three-dimensional position coordinates that are close in the front-rear and left-right directions but differ significantly in the up-down direction, the corresponding terrain height state and settlement depth scalar are read to retain the three-dimensional position coordinates that can express the chassis jamming height relationship. After spatial consistency processing, the three-dimensional position coordinates of all failed and trapped nodes form a trapped spatial point record. This trapped spatial point record includes the spatially consistent three-dimensional position coordinates, the corresponding failed and trapped node, the settlement depth scalar, the slip characteristic parameters, and the wheel end attribution relationship. The trapped spatial point record continues to participate in the construction of the outer envelope geometric convex hull, ensuring that the outer envelope geometric convex hull is not directly generated by repeated points, discrete noise points, or points with inconsistent scales.

[0069] Preferably, when constructing the outer envelope geometric convex hull based on all the three-dimensional position coordinates, the three-dimensional position coordinates in the trapped spatial point record are first read, and then the three-dimensional position coordinates are subjected to planar projection processing according to the chassis front-rear direction reference and chassis left-right direction reference to form trapped plane projection points; then, according to the orientation relationship of the trapped plane projection points around the origin of the chassis coordinate system, the trapped plane projection points are sorted by outer edge to form outer edge sorting results; then, trapped plane projection points located inside the outer edge connection relationship are removed from the outer edge sorting results to form outer projection boundary points. The outer projection boundary points are used to express the outer coverage range of the failed trapped node on the chassis horizontal plane. After the outer projection boundary points are determined, the outer projection boundary points are re-associated with the vertical coordinates in the corresponding three-dimensional position coordinates to form three-dimensional outer boundary points. The three-dimensional outer boundary points continue to participate in the construction of the outer envelope geometric convex hull, so that the outer envelope geometric convex hull has both horizontal coverage range and vertical settlement range.

[0070] Preferably, the outer envelope geometric convex hull is formed by the three-dimensional outer boundary points. During construction, outer edge connecting edges are generated according to the spatial adjacency relationship between adjacent three-dimensional outer boundary points, and an outer envelope surface is formed according to the height change relationship between the outer edge connecting edges and the chassis vertical reference. The outer edge connecting edges are used to express the boundary connection relationship of multiple failed and trapped nodes in the chassis circumferential direction, and the outer envelope surface is used to express the settlement coverage range of multiple failed and trapped nodes in three-dimensional space. If the number of three-dimensional position coordinates in the trapped spatial point record is small, the three-dimensional position coordinates of a single or two adjacent failed and trapped nodes are expanded by combining the chassis projection range of the wheel adjacent map area to which the failed and trapped node belongs, so as to form a three-dimensional outer boundary point that can cover the wheel adjacent map area; the boundary expansion processing is carried out based on the wheel installation position record, the standard tire radius constant, and the wheel adjacent map area, without introducing an expansion area unrelated to the unmanned vehicle chassis structure. The resulting outer envelope geometric convex hull includes three-dimensional outer boundary points, outer edge connecting edges, and outer envelope surfaces. The outer envelope geometric convex hull can express the outer boundaries of the failed trapped node and its adjacent trapped contact area in the circumferential and height directions of the chassis, and continues to participate in the determination of the three-dimensional physical lock-up boundary.

[0071] Preferably, when determining the three-dimensional physical lock-up boundary of the chassis based on the outer envelope geometric convex hull, the three-dimensional outer boundary points, outer edge connecting edges, and outer edge envelope surfaces in the outer envelope geometric convex hull are first read, and then the map units in the environmental contact stress distribution map that intersect or are adjacent to the outer envelope geometric convex hull are read. Subsequently, the terrain height state, torque load attributes, and wheel end affiliation of these map units are correlated with the outer envelope geometric convex hull, and map units located near the outer edge of the outer envelope geometric convex hull that have an enhanced wheel end torque load state or a continuous tangential sliding state are selected to form lock-up boundary candidate map units. The lock-up boundary candidate map units are used to express the contact area near the outer edge of the outer envelope geometric convex hull that forms spatial interference with the chassis or wheel, an enhanced wheel end torque load state, or a continuous tangential sliding state. Through the lock-up boundary candidate map units, the outer envelope geometric convex hull is not just a geometric outline, but continues to establish a correspondence with the terrain height state, torque load attributes, and slip characteristic parameters in the environmental contact stress distribution map.

[0072] Preferably, when determining the three-dimensional physical locked boundary, the candidate map units of the locked boundary are subjected to boundary continuity processing. This boundary continuity processing first connects adjacent candidate map units of the locked boundary according to the chassis projection position, and then determines whether they belong to the same trapped contact area based on the changes in terrain height and torque load attributes between adjacent candidate map units. For candidate map units of the locked boundary belonging to the same trapped contact area, they are continuously arranged along the outer edge of the outer envelope of the outer envelope geometric convex hull to form a continuous segment of the locked boundary. For a single candidate map unit of the locked boundary that lacks spatial continuity with adjacent candidate map units, its corresponding settlement depth scalar and slip characteristic parameters are read for verification. Candidate map units of the locked boundary that still exhibit a trapped contact state after verification are retained in the continuous segment of the locked boundary. The continuous segment of the locked boundary continues to participate in the formation of the three-dimensional physical locked boundary, enabling the three-dimensional physical locked boundary to express the continuous trapped contact area between the chassis and the complex terrain.

[0073] Preferably, the three-dimensional physical lock-up boundary is jointly determined by the outer envelope geometric convex hull and the lock-up boundary continuous segment. The outer envelope geometric convex hull is used to define the three-dimensional outer envelope range of the failed and trapped node in the chassis coordinate system. The lock-up boundary continuous segment is used to define the trapped contact range near the outer edge of the outer envelope geometric convex hull, which corresponds to the terrain height state, torque load attribute, and slip characteristic parameters. After determining the three-dimensional physical lock-up boundary, the three-dimensional physical lock-up boundary is written into the physical lock-up boundary record. The physical lock-up boundary record includes the failed and trapped node, three-dimensional position coordinates, outer envelope geometric convex hull, lock-up boundary continuous segment, settlement depth scalar, slip characteristic parameters, and wheel end affiliation. The physical lock-up boundary record continues to participate in the subsequent configuration of the multimodal morphological reconstruction strategy corresponding to the three-dimensional physical lock-up boundary, so that the subsequent vehicle pitch adjustment sequence and independent wheel end pulse output sequence can be generated around the spatial range defined by the three-dimensional physical lock-up boundary.

[0074] Preferably, when the unmanned vehicle passes through a muddy, sunken area, if the settlement depth scalar corresponding to the front wheel exceeds the chassis ground clearance safety threshold, and the slip characteristic parameter corresponding to the front wheel indicates that the front wheel is in a continuous tangential slipping state, then the front wheel is marked as a failed and trapped node, and the first pixel coordinate, second pixel coordinate, settlement depth scalar, and slip characteristic parameter corresponding to the front wheel are written into the failed and trapped node marking record. Subsequently, according to the map scale mapping record, the first pixel coordinate and second pixel coordinate corresponding to the front wheel are converted into three-dimensional position coordinates, and the three-dimensional position coordinates are written into the failed and trapped node position record. If the rear wheel or the other wheel also forms a failed and trapped node, then the three-dimensional position coordinates of all failed and trapped nodes jointly participate in forming a trapped spatial point record, and the trapped spatial point record continues to participate in the construction of the outer envelope geometric convex hull; the outer envelope geometric convex hull, combined with the locked boundary candidate map unit in the environmental contact stress distribution map, forms a three-dimensional physical locked boundary. This process allows the wheel settlement caused by muddy sinking areas, the wheel-end torque load state, and the tangential slip state to jointly contribute to determining the range of chassis space trapped.

[0075] Preferably, in all-weather complex terrain inspection scenarios, the vehicle undercarriage contour scanning point cloud data collected by lidar, vision sensors, and multi-sensor obstacle avoidance components will experience local map fluctuations due to differences in ground reflection, partial occlusion, or short-term wheel bounce. These local map fluctuations will manifest as short-term jumps in the terrain height state and torque load attribute change state of some map units in the environmental contact stress distribution map. To reduce the impact of local map fluctuations on the three-dimensional physical lock-up boundary, the determination of the failed and trapped node is not based directly on the terrain height state of a single map unit, but rather on the settlement depth scalar, slip characteristic parameters, and chassis ground clearance safety threshold recorded in the wheel settlement characterization record. When constructing the outer envelope geometric convex hull, not all discrete map units are used directly, but the three-dimensional position coordinates are first spatially consistent through the failed and trapped node position records and trapped spatial point records. After the above processing, the outer envelope geometric convex hull is further combined with the continuous segments of the locked boundary to determine the three-dimensional physical locked boundary, so that the three-dimensional physical locked boundary can correspond to the continuous trapped contact area under complex terrain, and provide spatial boundary basis for subsequent multimodal morphological reconstruction strategies.

[0076] Optionally, the step of configuring the multimodal morphological reconstruction strategy corresponding to the three-dimensional physically locked boundary includes: Calculate the projected area of ​​the outer envelope geometric convex hull on the horizontal plane; Calculate the entrapment ratio coefficient of the projected area relative to the total projected area of ​​the unmanned vehicle chassis; Based on the trapped ratio coefficient, a matching address is performed in a preset strategy mapping data table to determine the initial set of escape actions; Collect the current torque output margin of normally operating wheels that are not marked as the failed and trapped node; The action amplitude parameters in the initial set of escape actions are pruned and restricted based on the current torque output margin in order to configure the multimodal morphological reconstruction strategy.

[0077] Preferably, the specific implementation process of configuring the multimodal morphological reconstruction strategy corresponding to the three-dimensional physical lock-up boundary is as follows: After determining the three-dimensional physical lock-up boundary, first read the previously formed physical lock-up boundary record, outer envelope geometric convex hull, lock-up boundary continuous segment, failed trapped node marker record, and failed trapped node position record; the physical lock-up boundary record includes the failed trapped node, three-dimensional position coordinates, outer envelope geometric convex hull, lock-up boundary continuous segment, settlement depth scalar, slip characteristic parameter, and wheel end attribution relationship; the outer envelope geometric convex hull includes three-dimensional outer boundary points, outer edge connecting edges, and outer edge envelope surface; the lock-up boundary continuous segment is used to express the continuous trapped contact area near the outer edge of the outer envelope geometric convex hull; the failed trapped node marker record is used to express the wheel identifier, settlement depth scalar, slip characteristic parameter, and wheel end attribution relationship of the node marked as a failed trapped node; the failed trapped node position record is used to express the three-dimensional position coordinates of the failed trapped node in the chassis coordinate system. By reading the physical locked boundary record, the outer envelope geometric convex hull, the locked boundary continuous segment, the failed and trapped node marker record, and the failed and trapped node position record, when calculating the projected area of ​​the outer envelope geometric convex hull on the horizontal plane, the three-dimensional position coordinates that have been standardized in scale and spatial consistency can be directly used, and the projected area of ​​the outer envelope geometric convex hull on the horizontal plane can maintain the same data source as the three-dimensional physical locked boundary, the failed and trapped node, and the locked boundary continuous segment.

[0078] Preferably, when calculating the projected area of ​​the outer envelope geometric convex hull on the horizontal plane, the three-dimensional outer boundary points in the outer envelope geometric convex hull are first read, and each three-dimensional outer boundary point is projected onto the horizontal plane of the chassis according to the chassis front-rear direction reference and the chassis left-right direction reference to form horizontal projection boundary points. Then, according to the spatial adjacency relationship defined by the outer edge connecting edge, the horizontal projection boundary points are sequentially connected to form a horizontal projection envelope region; the horizontal projection envelope region is used to express the coverage range of the failed trapped node and its adjacent trapped contact area on the horizontal plane of the chassis. After forming the horizontal projection envelope region, the map scale mapping record is read, the chassis projection position in the horizontal projection envelope region is scaled uniformly, and the area of ​​the horizontal projection envelope region is calculated under the same scale expression to obtain the projected area of ​​the outer envelope geometric convex hull on the horizontal plane. The projected area of ​​the outer envelope geometric convex hull on the horizontal plane continues to participate in the calculation of the trapped ratio coefficient, so that the outer envelope geometric convex hull is converted from a three-dimensional spatial boundary into horizontal coverage range data that can participate in strategy mapping data table matching and addressing.

[0079] Preferably, before calculating the entrapment ratio coefficient of the projected area of ​​the outer envelope geometric convex hull on the horizontal plane relative to the total projected area of ​​the unmanned vehicle chassis, the chassis outer contour projection record in the unmanned vehicle chassis parameter record is first read. This chassis outer contour projection record originates from chassis structure data configured before the unmanned vehicle inspection task is configured. The chassis outer contour projection record includes the chassis's front-rear boundaries, left-right boundaries, and the projection boundary of the chassis's lower surface. Based on the chassis outer contour projection record, the total projected area of ​​the unmanned vehicle chassis is formed on the horizontal plane, and the total projected area of ​​the unmanned vehicle chassis is calculated under the same scale expression corresponding to the atlas scale mapping record. Subsequently, the projected area of ​​the outer envelope geometric convex hull on the horizontal plane is proportionally converted to the total projected area of ​​the unmanned vehicle chassis to form the entrapment ratio coefficient. This entrapment ratio coefficient is used to express the degree of occupancy of the three-dimensional physical locked boundary within the total projected area of ​​the unmanned vehicle chassis. The entrapment ratio coefficient continues to serve as input data for policy mapping data table matching and addressing.

[0080] Preferably, the strategy mapping data table is pre-established before the unmanned vehicle inspection task is configured and stored along with the unmanned vehicle's complex terrain escape parameters. The strategy mapping data table originates from unmanned vehicle chassis parameter records, wheel mounting position records, unmanned vehicle tire parameter records, suspension travel parameter records, hub motor output parameter records, and inspection terrain type records. The inspection terrain type records are used to express complex terrain categories such as mud, sand, gravel ditches, or rugged ground. The hub motor output parameter records are used to express the allowable torque output range for each wheel under normal inspection and escape conditions. The suspension travel parameter records are used to express the available suspension travel range during vehicle pitch adjustment. The strategy mapping data table includes a trapped proportion coefficient range, a failed trapped node distribution type, a slip characteristic parameter level, a three-dimensional physical lock-up boundary space type, an initial escape action set, an upper limit for action amplitude parameters, a vehicle pitch adjustment sequence template, and an independent wheel-end pulse output sequence template. Through the strategy mapping data table, subsequent matching and addressing can simultaneously read the trapped ratio coefficient, the failed trapped node marker record, the sliding feature parameter, and the physical deadlock boundary record, so that the target matching record can maintain a correspondence with the coverage area of ​​the three-dimensional physical deadlock boundary, the distribution state of the failed trapped node, and the tangential sliding state expressed by the sliding feature parameter.

[0081] Preferably, when performing matching and addressing in a preset strategy mapping data table based on the trapped ratio coefficient, the trapped ratio coefficient, the failed trapped node marker record, the slip characteristic parameter, and the physical lock-up boundary record are first read. Then, the trapped ratio coefficient is matched with the trapped ratio coefficient interval in the strategy mapping data table, and the wheel identifier and wheel end affiliation in the failed trapped node marker record are matched with the failed trapped node distribution type in the strategy mapping data table. After completing interval matching and position matching, the slip characteristic parameter is matched with the slip characteristic parameter level in the strategy mapping data table, and the three-dimensional physical lock-up boundary in the physical lock-up boundary record is matched with the three-dimensional physical lock-up boundary space type in the strategy mapping data table. The interval matching, position matching, state matching, and boundary matching together form the matching and addressing result. The matching and addressing result is used to determine the target matching record in the strategy mapping data table, and the target matching record is used to determine the initial escape action set.

[0082] Preferably, when determining the initial set of escaping actions, the initial set of escaping actions, the vehicle pitch adjustment sequence template, the independent wheel-end pulse output sequence template, and the upper limit of the action amplitude parameters are read from the target matching record. The spatial range of the three-dimensional physical lock-up boundary, the wheel-end affiliation relationship of the failed and trapped node, and the slip characteristic parameters are written into the initial set of escaping actions. The initial set of escaping actions includes suspension lifting actions, suspension lowering actions, vehicle pitch holding actions, wheel-end pulse drive actions, and wheel-end brief stop actions. The suspension lifting actions, suspension lowering actions, and vehicle pitch holding actions are used to generate the vehicle pitch adjustment sequence, and the wheel-end pulse drive actions and wheel-end brief stop actions are used to generate the independent wheel-end pulse output sequence. The initial set of escaping actions is not directly sent to the suspension controller or wheel hub motor controller. Instead, after subsequently reading the current torque output margin of the normally operating wheels, it continues to be subject to action amplitude parameter trimming restrictions, enabling the initial set of escaping actions to form trimmed action amplitude parameters based on the current torque output margin of the normally operating wheels.

[0083] Preferably, when collecting the current torque output margin of normally operating wheels not marked as the failed and trapped node, the failed and trapped node marking record is first read, and the wheel identifiers not appearing in the failed and trapped node marking record are identified as normally operating wheel identifiers. Then, based on the normally operating wheel identifier, the current torque sample value of the corresponding normally operating wheel is read from the current torque feedback sequence data, and the allowable torque output range of the corresponding normally operating wheel is read from the hub motor output parameter record. The difference between the available upper limit value in the allowable torque output range and the current torque sample value is processed to form the current torque output margin of the corresponding normally operating wheel. The current torque output margin, along with the normally operating wheel identifier, wheel end affiliation, current torque sample value, and allowable torque output range, is written into the normally operating wheel margin record. The normally operating wheel margin record continues to participate in the action amplitude parameter clipping and restriction in the initial escape action set.

[0084] Preferably, after forming the normal operating wheel margin record, the current torque output margin of each normal operating wheel is sorted by wheel end, so that the current torque output margin of each normal operating wheel corresponds to the front-rear position, left-right position, and wheel end affiliation of the normal operating wheel in the chassis coordinate system. Then, the distribution positions of the failed and trapped nodes in the physical lock-up boundary record are read, and the traction participation direction of each normal operating wheel relative to the three-dimensional physical lock-up boundary is determined. If a normal operating wheel is located outside the three-dimensional physical lock-up boundary and is in a relative traction direction with a failed and trapped node, its current torque output margin is written into the priority participation margin field of the normal operating wheel margin record. If a normal operating wheel is located inside the three-dimensional physical lock-up boundary or close to a continuous segment of the lock-up boundary, its current torque output margin is written into the auxiliary participation margin field of the normal operating wheel margin record. Through this wheel end sorting, the normal operating wheel margin record can establish a spatial directional correspondence between the current torque output margin and the wheel end pulse drive actions in the initial escape action set.

[0085] Preferably, when limiting the action amplitude parameters in the initial set of escaping actions based on the current torque output margin, the upper limit of the action amplitude parameters in the initial set of escaping actions is first read, then the current torque output margin in the normal operating wheel margin record is read, and the current torque output margin is matched with the wheel-end pulse drive action according to the normal operating wheel identifier. If the action amplitude parameter corresponding to the wheel-end pulse drive action exceeds the current torque output margin of the normal operating wheel, the action amplitude parameter is trimmed to a range not exceeding the current torque output margin to form the trimmed wheel-end action amplitude parameter; if the action amplitude parameter corresponding to the wheel-end pulse drive action does not exceed the current torque output margin of the normal operating wheel, the action amplitude parameter is retained and written into the trimmed wheel-end action amplitude parameter. The trimmed wheel-end action amplitude parameter continues to participate in the generation of independent wheel-end pulse output sequences, so that the independent wheel-end pulse output sequences maintain a correspondence with the current torque output margin of the normal operating wheel.

[0086] Preferably, when limiting the amplitude parameters of the suspension lifting, suspension lowering, and vehicle pitch holding actions in the initial set of escape actions, the suspension travel parameter record, physical lock-up boundary record, and failed-trapped node position record are read. The suspension travel parameter record provides the available travel range of each suspension support strut, the physical lock-up boundary record provides the spatial range of the three-dimensional physical lock-up boundary, and the failed-trapped node position record provides the three-dimensional position coordinates of the failed-trapped node. Subsequently, based on the forward / backward and left / right positions of the failed-trapped node in the chassis coordinate system, the suspension support struts that need to participate in vehicle pitch adjustment are determined, and the amplitude parameters of the suspension lifting, suspension lowering, and vehicle pitch holding actions are trimmed to the range allowed by the suspension travel parameter record to form the trimmed suspension amplitude parameters. The trimmed suspension amplitude parameters continue to participate in the generation of the vehicle pitch adjustment sequence, enabling the vehicle pitch adjustment sequence to adjust its attitude around the spatial range of the three-dimensional physical lock-up boundary.

[0087] Preferably, after forming the trimmed rear wheel end motion amplitude parameters and the trimmed rear suspension motion amplitude parameters, the initial escaping action set is reconstructed and organized to form a candidate morphology reconstruction action record. The candidate morphology reconstruction action record includes a three-dimensional physical lock-up boundary, a entrapment ratio coefficient, a target matching record, the initial escaping action set, the trimmed rear wheel end motion amplitude parameters, the trimmed rear suspension motion amplitude parameters, the normal operating wheel margin record, and a failed entrapment node marker record. The candidate morphology reconstruction action record continues to participate in the configuration of the multimodal morphology reconstruction strategy, ensuring that the multimodal morphology reconstruction strategy is not simply statically read from the strategy mapping data table, but is jointly limited by the spatial range of the three-dimensional physical lock-up boundary, the entrapment ratio coefficient, the distribution state of failed entrapment nodes, and the current torque output margin.

[0088] Preferably, when configuring the multimodal morphological reconstruction strategy, the trimmed suspension action amplitude parameters are first read from the candidate morphological reconstruction action record. Based on the front-rear and left-right coverage ranges of the three-dimensional physical lock-up boundary in the chassis coordinate system, the sequential relationship of suspension actions in the vehicle pitch adjustment sequence is determined. If the three-dimensional physical lock-up boundary is mainly distributed on the front side of the chassis, the sequential relationship of suspension actions preferentially includes the front suspension lifting action and the rear suspension lowering action; if the three-dimensional physical lock-up boundary is mainly distributed on the rear side of the chassis, the sequential relationship of suspension actions preferentially includes the rear suspension lifting action and the front suspension lowering action; if the three-dimensional physical lock-up boundary is distributed on one side of the chassis, the sequential relationship of suspension actions includes the travel adjustment action of the corresponding side suspension support strut. A vehicle pitch adjustment sequence is generated based on the sequential relationship of suspension actions and the trimmed suspension action amplitude parameters, and this sequence is further written into the multimodal morphological reconstruction strategy to change the geometric interference relationship between the chassis and complex terrain.

[0089] Preferably, when configuring the multimodal morphological reconstruction strategy, the trimmed wheel-end action amplitude parameters are also read from the candidate morphological reconstruction action records, and the wheel-end action sequence in the independent wheel-end pulse output sequence is determined based on the normal operating wheel margin record and the failure-trapped node marking record. For normal operating wheels that are not marked as failure-trapped nodes and have a high current torque output margin, the wheel-end action sequence prioritizes configuring wheel-end pulse drive actions; for normal operating wheels that are close to the three-dimensional physical lock-up boundary but are not marked as failure-trapped nodes, the wheel-end action sequence configures smaller amplitude wheel-end pulse drive actions or brief wheel-end stop actions; for wheels that have been marked as failure-trapped nodes, the wheel-end action sequence configures the wheel-end action rhythm required for reciprocating tangential vibration based on the corresponding slip characteristic parameters. An independent wheel-end pulse output sequence is generated based on the wheel-end action sequence and the trimmed wheel-end action amplitude parameters, and the independent wheel-end pulse output sequence is further written into the multimodal morphological reconstruction strategy to enable the wheel-end drive actions to unfold around the edge region of the three-dimensional physical lock-up boundary while the chassis attitude is changed by the vehicle body pitch adjustment sequence.

[0090] Preferably, after the multimodal morphological reconstruction strategy is configured, the multimodal morphological reconstruction strategy is written into the morphological reconstruction strategy record; the morphological reconstruction strategy record includes three-dimensional physical lock-up boundary, trapped ratio coefficient, target matching record, vehicle body pitch adjustment sequence, independent wheel end pulse output sequence, trimmed wheel end motion amplitude parameters, trimmed suspension motion amplitude parameters, and normal operating wheel margin record.

[0091] Preferably, when the unmanned vehicle passes the edge of the sand and the three-dimensional physical lock boundary mainly covers the front area of ​​the chassis, the projected area of ​​the outer envelope geometric convex hull on the horizontal plane will be concentrated in the horizontal projected envelope area of ​​the front of the chassis. The projected area of ​​the outer envelope geometric convex hull on the horizontal plane is proportionally converted to the total projected area of ​​the unmanned vehicle chassis to form a entrapment ratio coefficient. Subsequently, based on the entrapment ratio coefficient, the failure entrapment node marker record, and the slip characteristic parameters, matching and addressing are performed in the strategy mapping data table to obtain an initial set of extrication actions including the front suspension lifting action and the rear normal operating wheel wheel end pulse drive action. After further reading the current torque output margin of the rear normal operating wheel, the wheel end pulse drive action and suspension lifting action in the initial set of extrication actions are subject to action amplitude parameter clipping and restriction to form a corresponding multimodal morphological reconstruction strategy. This multimodal morphological reconstruction strategy enables the vehicle pitch adjustment sequence and the independent wheel end pulse output sequence to both unfold around the front three-dimensional physical lock boundary, and enables the vehicle pitch adjustment sequence and the independent wheel end pulse output sequence to continue to be written into the morphological reconstruction strategy record.

[0092] Preferably, in all-weather complex terrain inspection scenarios, the data collected by lidar, visual sensors, and multi-sensor obstacle avoidance components will cause local fluctuations in the environmental contact stress distribution map, and the current torque feedback sequence data will also experience local torque fluctuations due to short-term wheel bounce. To reduce the impact of local map fluctuations and local torque fluctuations on the multimodal morphology reconstruction strategy, the projected area of ​​the outer envelope geometric convex hull on the horizontal plane is used when calculating the entrapment ratio coefficient, instead of using the area of ​​a single map unit. When performing matching addressing, the entrapment ratio coefficient, the failed entrapment node marker record, and the slip characteristic parameter are read simultaneously, instead of relying solely on a single settlement value. When limiting the action amplitude parameter, the remaining balance record of the normally operating wheel is read, instead of directly using the upper limit of the action amplitude parameter in the strategy mapping data table. The morphology reconstruction strategy record formed by the above processing can maintain a correspondence with the three-dimensional physical locked boundary, the current torque output balance of the normally operating wheel, and the contact state of the complex terrain, and provide data basis for the subsequent execution of creeping gait actions by the unmanned vehicle.

[0093] Optionally, the multimodal morphology reconstruction strategy includes the following methods for generating the vehicle body pitch adjustment sequence: Analyze the quadrant positions of the failed and trapped nodes in the chassis coordinate system; The target suspension support strut that needs to be lifted is determined based on the quadrant position. A set of hydraulic stroke expansion commands for the target suspension support strut is generated according to a preset step-increment rule; The vehicle pitch adjustment sequence is generated based on the hydraulic stroke expansion command set.

[0094] Preferably, the specific implementation process of the vehicle body pitch adjustment sequence generation method included in the multimodal morphology reconstruction strategy is as follows: After forming the morphology reconstruction strategy record, the three-dimensional physical lock-up boundary, the failure and entrapment node marker record, the failure and entrapment node position record, the trimmed suspension action amplitude parameters, and the suspension travel parameter record in the morphology reconstruction strategy record are read first; the failure and entrapment node marker record is used to provide the wheel identifier, settlement depth scalar, slip characteristic parameters, and wheel end attribution relationship of the wheel that has experienced settlement exceeding the boundary and participated in the entrapment judgment; the failure and entrapment node position record is used to provide the three-dimensional position coordinates of the failure and entrapment node in the chassis coordinate system; the suspension travel parameter record is used to provide the available travel range of each suspension support strut; and the trimmed suspension action amplitude parameters are used to limit the travel change amplitude of each suspension support strut in the subsequent adjustment process. By first reading the three-dimensional physical lock-up boundary, the failed and trapped node marker record, the failed and trapped node position record, the trimmed suspension action amplitude parameters, and the suspension travel parameter record, the generation of the vehicle pitch adjustment sequence can directly inherit the three-dimensional physical lock-up boundary and the multimodal morphological reconstruction strategy, rather than generating suspension control content solely based on preset vehicle attitude parameters. The preset vehicle attitude parameters are derived from the normal vehicle inspection attitude data stored before the unmanned vehicle inspection task configuration, and are only used as a distinction in this section. The subsequent vehicle pitch adjustment sequence is still directly generated based on the three-dimensional physical lock-up boundary, the failed and trapped node position record, and the trimmed suspension action amplitude parameters.

[0095] Preferably, when analyzing the quadrant positions of the failed and trapped nodes in the chassis coordinate system, the three-dimensional position coordinates in the failed and trapped node position record are first read. Then, based on the origin of the chassis coordinate system, the front-rear and rear-rear reference points, and the left-right reference points, the three-dimensional position coordinates of each failed and trapped node are assigned to a corresponding quadrant position. The quadrant positions include the front left region, front right region, rear left region, and rear right region of the chassis. These regions are derived from the combination of front-rear and rear-rear coordinates and left-right coordinates in the chassis coordinate system. After completing the quadrant position assignment, the failed and trapped nodes, the three-dimensional position coordinates, the settlement depth scalar, the slip characteristic parameters, and the quadrant positions are written into the failed and trapped node quadrant record. This quadrant record continues to be used to determine the target suspension support struts that require lifting operations, allowing the quadrant positions to continue participating in the selection of target suspension support struts based on the spatial distribution of the failed and trapped nodes.

[0096] Preferably, after forming the quadrant record of the failed and trapped nodes, the quadrant position is not determined solely by the number of failed and trapped nodes. Instead, the continuous segments of the locked boundary in the physical locked boundary record and the outer edge envelope surface in the outer envelope geometric convex hull are read. The front-to-back and left-to-right coverage areas of the continuous segments of the locked boundary in the chassis coordinate system are correlated with the quadrant positions in the quadrant record of the failed and trapped nodes. If multiple failed and trapped nodes are located in the same quadrant position, and the continuous segments of the locked boundary corresponding to that quadrant position have continuous trapped contact areas, then that quadrant position is marked as the primary trapped quadrant position. If the failed and trapped nodes are distributed in two adjacent quadrant positions, then the primary trapped quadrant position and the secondary trapped quadrant position are determined based on the settlement depth scalar and slip characteristic parameters corresponding to the two quadrant positions. The primary trapped quadrant position and the secondary trapped quadrant position are written into the quadrant record of the failed and trapped nodes and continue to participate in the determination of the target suspension support strut, so that the vehicle pitch adjustment sequence can be formed around the actual distribution state of the three-dimensional physical locked boundary.

[0097] Preferably, when determining the target suspension support strut requiring lifting operation based on the quadrant position, the installation position record of the suspension support strut is first read. This record originates from the chassis structure data configured before the unmanned vehicle inspection task configuration. The installation position record includes the front-rear coordinates, left-right coordinates, installation height position, and suspension support strut identifier of each strut relative to the chassis coordinate system. Subsequently, the installation position record of the suspension support strut is matched with the quadrant record of the failed and trapped node, and the suspension support strut located at or adjacent to the main trapped quadrant position is identified as a candidate suspension support strut. The candidate suspension support struts undergo a stroke availability check against the suspension stroke parameter record. This check reads the available stroke range of the candidate suspension support strut and determines whether it can perform the lifting operation corresponding to the trimmed suspension action amplitude parameters. After passing the stroke availability check, the candidate suspension support struts continue to participate in the determination of the target suspension support strut.

[0098] Preferably, when determining the target suspension support strut, the candidate suspension support struts that have passed the travel availability check are further subjected to boundary decoupling relationship judgment. The boundary decoupling relationship judgment reads the three-dimensional physical lock-up boundary, the continuous segment of the lock-up boundary, and the position record of the failed trapped node, and determines whether the candidate suspension support strut, after performing a lifting operation, can move the chassis lower surface of the corresponding quadrant position away from the continuous segment of the lock-up boundary. If the position of the candidate suspension support strut is consistent with the position of the main trapped quadrant, then the candidate suspension support strut is taken as the main target suspension support strut; if another candidate suspension support strut adjacent to the position of the main trapped quadrant can cooperate to form the vehicle body pitch adjustment attitude, then the candidate suspension support strut is taken as the cooperating target suspension support strut. The main target suspension support strut and the cooperating target suspension support strut together constitute the target suspension support strut. The target suspension support strut continues to participate in the generation of the hydraulic travel expansion command set, so that the source of the target suspension support strut maintains a correspondence with the position of the main trapped quadrant, the position of the subordinate trapped quadrant, and the continuous segment of the lock-up boundary.

[0099] Preferably, the step increment rule is pre-set before configuring the unmanned vehicle inspection task and stored in the suspension travel parameter record. The step increment rule is derived from the available travel range of the suspension support strut, the allowable range of attitude changes under normal vehicle inspection posture, the chassis ground clearance safety threshold, and the trimmed suspension action amplitude parameters. When pre-setting the step increment rule, the available travel range of the suspension support strut is first divided into multiple consecutive travel increment intervals, and then each travel increment interval is associated with the corresponding vehicle attitude change amplitude to form a travel step record. The travel step record includes the suspension support strut identifier, single travel increment, cumulative travel upper limit, time interval between adjacent increment actions, and allowable attitude change range. The travel step record, as the specific data content of the step increment rule, is further used to generate a hydraulic travel expansion command set for the target suspension support strut, enabling the step increment rule to enter the generation process of the hydraulic travel expansion command set from the suspension travel parameter record.

[0100] Preferably, when generating the hydraulic stroke expansion command set for the target suspension support strut according to a preset step-increment rule, the target suspension support strut, the stroke step record, the trimmed suspension motion amplitude parameters, and the three-dimensional physical lock-up boundary are first read. Then, the single stroke increment corresponding to the target suspension support strut is selected from the stroke step record, and the trimmed suspension motion amplitude parameters are used as the cumulative stroke upper limit. Multiple hydraulic stroke expansion commands are generated in ascending stroke increment order. Each hydraulic stroke expansion command includes a target suspension support strut identifier, the current stroke target value, the single stroke increment, the cumulative stroke value, and the attitude holding time. Multiple hydraulic stroke expansion commands are written into the hydraulic stroke expansion command set in the generation order, enabling the hydraulic stroke expansion command set to drive the target suspension support strut to perform a lifting operation in a segmented incremental manner, and allowing the hydraulic stroke expansion command set to continue participating in the generation of the vehicle pitch adjustment sequence.

[0101] Preferably, when generating the hydraulic stroke expansion command set, the spatial relationship between the primary target suspension support strut and the cooperating target suspension support strut is also read, and the current stroke target value in the hydraulic stroke expansion command set is relatively coordinated according to the spatial relationship of the target suspension support struts. The spatial relationship of the target suspension support struts is derived from the suspension support strut installation position record and is used to express the relative positions of the primary target suspension support strut and the cooperating target suspension support strut in the front-rear direction and the left-right direction of the chassis. If the primary target suspension support strut is located on the front side of the chassis, the hydraulic stroke expansion command set preferentially includes the stroke expansion command of the front primary target suspension support strut, and generates the cooperating return command of the rear cooperating target suspension support strut according to the vehicle pitch adjustment attitude; if the primary target suspension support strut is located on the left or right side of the chassis, the hydraulic stroke expansion command set preferentially includes the stroke expansion command of the corresponding side primary target suspension support strut, and generates the cooperating adjustment command of the other side cooperating target suspension support strut according to the vehicle pitch adjustment attitude. The hydraulic stroke expansion command set, after the aforementioned relatively coordinated processing, continues to participate in the generation of the vehicle pitch adjustment sequence, enabling the vehicle pitch adjustment sequence to reduce disturbance to the grounding state of other wheels while raising the position of the main trapped quadrant.

[0102] Preferably, when generating the vehicle pitch adjustment sequence based on the hydraulic stroke expansion command set, multiple hydraulic stroke expansion commands are first sequentially arranged according to the target suspension support strut identifier, current stroke target value, and attitude holding time in the hydraulic stroke expansion command set to form a suspension stroke timing record. The suspension stroke timing record includes the target suspension support strut identifier, hydraulic stroke expansion command, command execution order, current stroke target value, cumulative stroke value, and attitude holding time. Subsequently, the suspension stroke timing record is correlated with the spatial range of the three-dimensional physical lock-up boundary, and the distance change relationship between the chassis lower surface at the quadrant where the target suspension support strut is located and the continuous segment of the lock-up boundary is determined after each hydraulic stroke expansion command is executed. The distance change relationship is further used to filter the hydraulic stroke expansion commands in the suspension stroke timing record, so that the vehicle pitch adjustment sequence retains the hydraulic stroke expansion commands that can move the chassis lower surface away from the continuous segment of the lock-up boundary; the retained hydraulic stroke expansion commands are further written into the vehicle pitch adjustment sequence according to the command execution order.

[0103] Preferably, when generating the vehicle pitch adjustment sequence, the timing of the hydraulic stroke expansion command set and the wheel-end action sequence of the independent wheel-end pulse output sequence is also coordinated. This timing coordination process reads the independent wheel-end pulse output sequence, the trimmed wheel-end action amplitude parameters, and the normal operating wheel margin record from the morphological reconstruction strategy record. The target suspension support strut lifting phase in the vehicle pitch adjustment sequence is set before the wheel-end pulse drive action of the independent wheel-end pulse output sequence, or within the pre-position preparation period of the wheel-end pulse drive action. The pre-position preparation period originates from the timing coordination process and is used to express the time interval during which the target suspension support strut forms the vehicle pitch adjustment posture before the start of the wheel-end pulse drive action. After completing the timing coordination process, the target suspension support strut lifting phase, the posture holding phase, and the target suspension support strut adjustment phase are sequentially written into the vehicle pitch adjustment sequence, ensuring that the vehicle pitch adjustment sequence and the independent wheel-end pulse output sequence form a timing relationship of first posture adjustment and then wheel-end drive.

[0104] Preferably, after the vehicle pitch adjustment sequence is generated, it is written into the morphological reconstruction strategy record. The vehicle pitch adjustment sequence includes the target suspension support strut, the hydraulic stroke expansion command set, the suspension stroke timing record, the target suspension support strut lifting stage, the attitude holding stage, the target suspension support strut adjustment stage, and the trimmed suspension action amplitude parameters. The morphological reconstruction strategy record continues to participate in the subsequent processing of sending the vehicle pitch adjustment sequence to the suspension controller according to a preset priority timing, so that the vehicle pitch adjustment sequence can be formed along the technical chain of "failure and entrapment node quadrant record, target suspension support strut, step size increment rule, hydraulic stroke expansion command set, suspension stroke timing record, and vehicle pitch adjustment sequence", and maintain a data correspondence with the three-dimensional physical lock-up boundary.

[0105] Preferably, when the unmanned vehicle passes the edge of the sand and the failed and trapped nodes are mainly distributed in the front left area of ​​the chassis, the three-dimensional position coordinates in the failed and trapped node position record will be divided into the front left area of ​​the chassis and written into the failed and trapped node quadrant record. Then, the suspension support strut corresponding to the front left area of ​​the chassis is determined as the main target suspension support strut, and the suspension support struts adjacent to the front left area of ​​the chassis are determined as the matching target suspension support struts. Then, according to the step increment rule in the suspension travel parameter record, a set of hydraulic travel expansion commands for the main target suspension support strut is generated, and matching adjustment commands are generated according to the matching target suspension support struts. After the hydraulic travel expansion command set and the matching adjustment commands are arranged in time sequence, a suspension travel time sequence record is formed. The suspension travel time sequence record is then written into the vehicle body pitch adjustment sequence, so that the lower surface of the chassis in the front left area of ​​the chassis gradually moves away from the three-dimensional physical lock boundary.

[0106] Preferably, in all-weather complex terrain inspection scenarios, the vehicle underbody contour scan point cloud data and current torque feedback sequence data may experience short-term changes in the position of the failed and trapped node due to local map fluctuations, local torque fluctuations, or short-term wheel bounces. To reduce the impact of these short-term changes on the vehicle pitch adjustment sequence, the failed and trapped node position record and the continuous segment of the lock boundary are read when analyzing the quadrant position, instead of reading a single map unit. When determining the target suspension support strut, the suspension support strut installation position record and suspension travel parameter record are read, instead of directly fixing a single suspension support strut. When generating the hydraulic travel expansion command set, multiple hydraulic travel expansion commands are generated according to the step size increment rule, instead of generating a large travel adjustment command all at once. The vehicle pitch adjustment sequence formed after the above processing can maintain a correspondence with the three-dimensional physical lock boundary, the target suspension support strut, the hydraulic travel expansion command set, and the trimmed suspension action amplitude parameters, and provides a basis for attitude adjustment for the subsequent unmanned vehicle to perform creeping gait movements.

[0107] Optionally, the multimodal morphology reconstruction strategy includes the following methods for generating the independent wheel-end pulse output sequence: Obtain the estimated value of the maximum ground adhesion force of the wheel corresponding to the edge of the three-dimensional physical lock boundary; Set a peak torque target that is greater than the estimated maximum ground adhesion value; A continuous torque control signal waveform with sinusoidal fluctuations is generated based on the stated torque peak target; The continuous torque control signal waveform is discretized and sampled in the time dimension to generate the independent wheel-end pulse output sequence.

[0108] Preferably, the specific implementation process of the generation method of the independent wheel-end pulse output sequence included in the multimodal morphology reconstruction strategy is as follows: After forming the morphology reconstruction strategy record, the three-dimensional physical lock-up boundary, physical lock-up boundary record, lock-up boundary continuous segment, failure and trapped node marker record, normal operating wheel margin record, trimmed wheel-end action amplitude parameter, and independent wheel-end pulse output sequence template in the morphology reconstruction strategy record are read first; the physical lock-up boundary record is used to provide the three-dimensional physical lock-up boundary, failure and trapped node, three-dimensional position coordinates, settlement depth scalar, slip characteristic parameter, and wheel-end affiliation relationship; the lock-up boundary continuous segment is used to express the continuous trapped contact area at the edge of the three-dimensional physical lock-up boundary; the failure and trapped node marker record is used to provide the wheel identifier, settlement depth scalar, slip characteristic parameter, and wheel-end affiliation relationship of the wheel that has experienced settlement cross-boundary and participated in the trapped judgment; the normal operating wheel margin record is used to provide the current torque output margin of the normal operating wheel that has not been marked as a failure and trapped node; and the trimmed wheel-end action amplitude parameter is used to limit the amplitude range of the wheel-end pulse drive action. By first reading the three-dimensional physical lock-up boundary, the physical lock-up boundary record, the lock-up boundary continuous segment, the failed and trapped node mark record, the normal operating wheel margin record, and the wheel end motion amplitude parameters after trimming, the subsequent acquisition of the maximum ground adhesion estimation value, setting the torque peak target, generating the continuous torque control signal waveform, and generating the independent wheel end pulse output sequence all have the preceding data source.

[0109] Preferably, when obtaining the estimated value of the maximum ground adhesion force of the wheel corresponding to the edge of the three-dimensional physical lock-up boundary, the candidate map units of the lock-up boundary in the continuous segment of the lock-up boundary are first read. The candidate map units of the lock-up boundary are derived from the map units in the environmental contact stress distribution map that are located near the outer edge of the outer envelope geometric convex hull and have enhanced wheel-end torque loading or continuous tangential sliding state, and participate in the formation of the continuous segment of the lock-up boundary after the previous boundary continuity processing. Then, according to the wheel-end attribution relationship in the candidate map units of the lock-up boundary, the wheel identifier corresponding to the edge of the three-dimensional physical lock-up boundary is determined. Then, the wheel identifier is matched with the failure and entrapment node marking record. If the wheel identifier has been marked as a failure and entrapment node, the wheel identifier is written into the edge failure wheel identification record. If the wheel identifier has not been marked as a failure and entrapment node but is located near the continuous segment of the lock-up boundary, the wheel identifier is written into the edge normal wheel identification record. The edge failure wheel identification record and the edge normal wheel identification record together form the edge wheel identification record. The edge wheel identification record is then used to determine the maximum ground adhesion estimation value, so that the maximum ground adhesion estimation value can be derived from the wheel end attribution relationship at the edge of the three-dimensional physical lock boundary.

[0110] Preferably, after forming the edge wheel identification record, the wheel proximity map region corresponding to the edge wheel identification record in the environmental contact stress distribution map is read. The wheel proximity map region is composed of map units with the same wheel end affiliation relationship and located within the spatial range of the corresponding wheel. The wheel proximity map region includes the chassis projection position, terrain height state, torque load attribute, and wheel end affiliation relationship. Based on the terrain height state in the wheel proximity map region, it is determined which complex terrain contact state the corresponding wheel is in: muddy sinking, loose sand, gravel ditch top, or rugged ground support. Based on the torque load attribute in the wheel proximity map region, it is determined the wheel end torque load state of the corresponding wheel in the complex terrain contact state. Based on the slip characteristic parameters, it is determined the tangential slip state of the corresponding wheel in the complex terrain contact state. The complex terrain contact state, the wheel end torque load state, and the tangential slip state are jointly written into the edge wheel contact state record, and the edge wheel contact state record continues to participate in the determination of the maximum ground adhesion estimation value.

[0111] Preferably, when determining the estimated maximum ground adhesion value, the edge wheel contact state record, wheel settlement characterization record, unmanned vehicle tire parameter record, and chassis passage parameter record are first read. The wheel settlement characterization record provides the settlement depth scalar, first pixel coordinate, second pixel coordinate, and vertical Euclidean distance of the corresponding wheel. The unmanned vehicle tire parameter record provides the standard tire radius constant and tire contact width parameter. The chassis passage parameter record provides the chassis ground clearance safety threshold and the clearance below the chassis. Subsequently, the estimated normal load value of the corresponding wheel is determined based on the settlement depth scalar, the standard tire radius constant, the tire contact width parameter, and the terrain height state. The ground adhesion state coefficient of the corresponding wheel is determined based on the complex terrain contact state and the slip characteristic parameter. The estimated normal load value and the ground adhesion state coefficient are converted into adhesion capability to form the estimated maximum ground adhesion value of the corresponding wheel. The estimated maximum ground adhesion value is then written into the edge wheel adhesion estimation record and used as the contact mechanics basis for subsequently setting the torque peak target.

[0112] Preferably, the ground adhesion state coefficient is pre-stored in a complex terrain adhesion parameter record before the unmanned vehicle inspection task is configured. This record is derived from the inspection terrain type record, terrain height identifiable by the vehicle's underbody contour scan point cloud data, wheel-end torque load identifiable by the current torque feedback sequence data, and tangential slippage identifiable by slip characteristic parameters. The complex terrain adhesion parameter record includes adhesion state coefficients for muddy ground, sandy ground, gravel ditch, and uneven ground, and also includes a state correction relationship between the slip characteristic parameters and the ground adhesion state coefficient. When actually generating the estimated maximum ground adhesion force value, the corresponding ground adhesion state coefficient is read from the complex terrain adhesion parameter record based on the complex terrain contact state in the edge wheel contact state record, and then corrected according to the slip characteristic parameters and the state correction relationship to form a state-corrected ground adhesion state coefficient. The ground adhesion state coefficient after state correction continues to participate in the determination of the maximum ground adhesion force estimation value, so that the maximum ground adhesion force estimation value can reflect the combined influence of complex terrain contact state and tangential sliding state.

[0113] Preferably, before setting a peak torque target greater than the estimated maximum ground adhesion value, the estimated maximum ground adhesion value is first converted to a different dimension. The estimated maximum ground adhesion value is ground contact force data, while the peak torque target is wheel-end drive torque data; the two cannot be directly compared. Therefore, the standard tire radius constant and wheel-end transmission parameter records from the unmanned vehicle's tire parameter records are read first. The wheel-end transmission parameter records originate from the wheel-end drive structure data configured before the unmanned vehicle's inspection task configuration. These records include wheel identification, the transmission correspondence between the hub motor and the wheel, and the torque transmission ratio. Subsequently, based on the standard tire radius constant and the wheel-end transmission parameter records, the estimated maximum ground adhesion value is converted into an adhesion torque threshold corresponding to that value. This adhesion torque threshold expresses the wheel-end torque threshold corresponding to the wheel maintaining contact adhesion at the edge of the three-dimensional physical lock boundary. This adhesion torque threshold then participates in setting the peak torque target.

[0114] Preferably, when setting the peak torque target, the adhesion torque threshold, the trimmed wheel-end motion amplitude parameter, the normal operating wheel margin record, and the hub motor output parameter record are read. The hub motor output parameter record provides the allowable torque range for the corresponding wheel, the normal operating wheel margin record provides the current torque output margin for the normal operating wheel, and the trimmed wheel-end motion amplitude parameter limits the upper limit of the wheel-end pulse drive motion amplitude. Then, without exceeding the allowable range of the hub motor output parameter record and the limited range of the trimmed wheel-end motion amplitude parameter, the peak torque target is set to a torque value higher than the adhesion torque threshold. Wherein, the peak torque target being higher than the adhesion torque threshold means that the magnitude is set within the same dimension as the wheel-end drive torque, rather than directly comparing the peak torque target with the estimated maximum ground adhesion value. The peak torque target is written into a peak torque target record, which includes a wheel identifier, adhesion torque threshold, peak torque target, trimmed wheel-end motion amplitude parameter, and current torque output margin. The peak torque target record is further used to generate a continuous torque control signal waveform.

[0115] Preferably, if the wheel identifier in the edge wheel identification record corresponds to a failed and trapped node, the torque peak target is set in a manner that exceeds the attached torque threshold but does not exceed the wheel end motion amplitude parameter after trimming, so as to drive the wheel corresponding to the failed and trapped node to generate reciprocating tangential vibration at the edge of the three-dimensional physical lock boundary; if the wheel identifier in the edge wheel identification record corresponds to a normally operating wheel, the torque peak target also needs to correspond to the current torque output margin in the normally operating wheel margin record, so that the torque peak target of the normally operating wheel does not exceed its current torque output margin. After the above processing, the torque peak target is both higher than the attached torque threshold and constrained by the wheel end motion amplitude parameter after trimming and the current torque output margin, which can prevent the torque peak target from deviating from the actual usable output range of the wheel end. The torque peak target continues to be stored in the torque peak target record and serves as the peak value basis for generating the continuous torque control signal waveform.

[0116] Preferably, when generating a sinusoidal continuous torque control signal waveform based on the torque peak target, the torque peak target record, the independent wheel-end pulse output sequence template, and the wheel-end action sequence are first read; the independent wheel-end pulse output sequence template is derived from the strategy mapping data table, and the wheel-end action sequence is derived from the morphological reconstruction strategy record. Then, the torque peak position of the continuous torque control signal waveform is determined based on the torque peak target, and the reference torque position of the continuous torque control signal waveform is determined based on the attached torque threshold; furthermore, the fluctuation period and fluctuation direction of the continuous torque control signal waveform are determined based on the slip characteristic parameters, enabling the continuous torque control signal waveform to form a periodic rise and fall between the torque peak position and the reference torque position. The continuous torque control signal waveform is written into a continuous torque control signal waveform record, and the continuous torque control signal waveform record continues to participate in the discretized sampling slice in the time dimension.

[0117] Preferably, when generating the continuous torque control signal waveform, the continuous torque control signal waveform does not adopt a single step torque increase method, but rather forms a continuous rise and fall change near the torque peak position in a sinusoidal oscillation manner. Specifically, the waveform start time, reference torque position, torque peak position, torque peak target, oscillation period, oscillation direction, and waveform duration are first written into the continuous torque control signal waveform record; then, according to the oscillation period, the continuous torque control signal waveform is organized into a continuous rising segment, a peak hold proximity segment, and a continuous falling segment. The continuous rising segment is used to gradually bring the wheel-end torque from the reference torque position closer to the torque peak position; the peak hold proximity segment is used to create a short-term high torque effect on the wheel-end torque near the torque peak position; and the continuous falling segment is used to gradually bring the wheel-end torque back from the torque peak position to the reference torque position. The continuous rising segment, the peak hold proximity segment, and the continuous falling segment together form the continuous torque control signal waveform and continue to participate in the discretization sampling slice.

[0118] Preferably, the fluctuation period is determined jointly based on the slip characteristic parameters, the current torque feedback sequence data, and the hub motor output parameter record. The slip characteristic parameters express the tangential slip state of the corresponding wheel, the current torque feedback sequence data expresses the wheel-end torque load state of the corresponding wheel under the current complex terrain contact state, and the hub motor output parameter record expresses the allowable torque variation range of the corresponding wheel. If the slip characteristic parameters indicate that the corresponding wheel has a continuous tangential slip state, the fluctuation period is set to a shorter period to make the continuous torque control signal waveform form a denser reciprocating tangential vibration; if the slip characteristic parameters indicate that the corresponding wheel has an intermittent tangential slip state, the fluctuation period is set to a longer period to make the continuous torque control signal waveform form a gentler torque rise and fall change. The fluctuation period is written into the continuous torque control signal waveform record and continues to serve as the time basis for discretized sampling slices.

[0119] Preferably, before discretizing and slicing the continuous torque control signal waveform in the time dimension, the control cycle record, hardware acquisition timestamp information, and continuous torque control signal waveform record of the hub motor controller are read first. The control cycle record is derived from the hub motor output parameter record and is used to express the time interval at which the hub motor controller can receive and execute torque control data; the hardware acquisition timestamp information is derived from the current torque feedback sequence data and is used to ensure that the time source of the independent wheel-end pulse output sequence is consistent with that of the current torque feedback sequence data; the continuous torque control signal waveform record is used to provide the continuous torque control signal waveform to be sliced. Subsequently, using the control cycle record as the sampling time interval, sampling times are sequentially set within the waveform duration of the continuous torque control signal waveform, and the corresponding torque target value is read at each sampling time to form a discrete torque sampling segment. The discrete torque sampling segment continues to participate in the generation of the independent wheel-end pulse output sequence.

[0120] Preferably, when performing discretized sampling slicing, each discrete torque sampling segment includes a wheel identifier, sampling time, target torque value, fluctuation period, fluctuation direction, and corresponding wheel-end action state. The wheel-end action state includes a wheel-end pulse drive state and a wheel-end brief stop state. The wheel-end pulse drive state originates from a discrete torque sampling segment in the continuous torque control signal waveform near the torque peak position, and the wheel-end brief stop state originates from a discrete torque sampling segment in the continuous torque control signal waveform near the reference torque position. Multiple discrete torque sampling segments are written into a discrete torque sampling segment record according to their sampling time sequence. This discrete torque sampling segment record is then used to generate an independent wheel-end pulse output sequence, enabling the continuous torque control signal waveform to be converted into discrete control data that the hub motor controller can read sequentially.

[0121] Preferably, when generating the independent wheel-end pulse output sequence based on the discrete torque sampling segment record, the sequence of wheel-end actions, the discrete torque sampling segment record, the torque peak target record, and the normal operating wheel margin record are first read. Then, according to the sequence of wheel-end actions, the discrete torque sampling segments under different wheel identifiers are time-sequentially arranged to form sub-wheel-end pulse output records. Each sub-wheel-end pulse output record includes a wheel identifier, discrete torque sampling segment, command execution order, wheel-end action status, and torque target value. Multiple sub-wheel-end pulse output records are then integrated according to a preset priority sequence to generate an independent wheel-end pulse output sequence. The independent wheel-end pulse output sequence is written into the morphological reconstruction strategy record and continues to participate in subsequent processing sent to the hub motor controller, ensuring that the independent wheel-end pulse output sequence maintains a data correspondence with the discrete torque sampling segment record and the sub-wheel-end pulse output records.

[0122] Preferably, when generating the independent wheel-end pulse output sequence, the independent wheel-end pulse output sequence is also time-correlated with the vehicle body pitch adjustment sequence. First, the vehicle body pitch adjustment sequence, the target suspension support strut lifting stage, and the attitude holding stage are read from the morphological reconstruction strategy record. Then, the wheel-end pulse drive state in the independent wheel-end pulse output sequence is arranged within the attitude holding stage, or after the target suspension support strut lifting stage. Through this time-correlation, the independent wheel-end pulse output sequence can apply periodic wheel-end pulse drive actions to the wheels corresponding to the edge of the three-dimensional physical lock-up boundary after the vehicle body pitch adjustment sequence has moved the lower surface of the chassis away from the continuous segment of the lock-up boundary. The time-correlation result is further written into the morphological reconstruction strategy record, ensuring that the vehicle body pitch adjustment sequence and the independent wheel-end pulse output sequence maintain an execution order within the same multimodal morphological reconstruction strategy.

[0123] Preferably, when the unmanned vehicle passes through a muddy, sunken area and its front wheel is located at the edge of the three-dimensional physical lock-up boundary, the front wheel is first written into the edge wheel identification record based on the continuous segment of the lock-up boundary and the wheel end attribution relationship. Then, the maximum ground adhesion estimation value is determined based on the settlement depth scalar, slip characteristic parameters, terrain height state, and torque load attributes of the front wheel. Subsequently, the maximum ground adhesion estimation value is converted into an adhesion torque threshold, and a torque peak target higher than the adhesion torque threshold is set within the allowable range of the hub motor output parameter record. After generating a continuous torque control signal waveform based on the torque peak target, the continuous torque control signal waveform is then discretized and sampled in the time dimension according to the control cycle record of the hub motor controller to form a discrete torque sampling segment record. The discrete torque sampling segment record is further integrated into an independent wheel end pulse output sequence, enabling the front wheel to generate periodic reciprocating tangential vibration at the edge of the three-dimensional physical lock-up boundary.

[0124] Preferably, in all-weather complex terrain inspection scenarios, the data collected by lidar, vision sensors, and multi-sensor obstacle avoidance components can cause local fluctuations in the environmental contact stress distribution map, and the current torque feedback sequence data can also experience local torque fluctuations due to short-term wheel bounce. Therefore, when generating independent wheel-end pulse output sequences, a single torque sample value is not directly used as the basis for wheel-end pulse output. Instead, the wheel corresponding to the edge of the three-dimensional physical lock boundary is first determined through edge wheel identification records. Then, the maximum ground adhesion force is estimated through edge wheel contact state records and edge wheel adhesion estimation records. Finally, a continuous torque control signal waveform is formed through adhesion torque threshold and torque peak target records. The independent wheel-end pulse output sequence generated after discretization sampling slices in the time dimension can maintain a correspondence with the three-dimensional physical lock boundary, torque peak target records, continuous torque control signal waveform records, discrete torque sampling segment records, wheel-end pulse output records, and vehicle pitch adjustment sequences, and provide wheel-end drive basis for the subsequent unmanned vehicle to perform creeping gait movements.

[0125] Optionally, the step of causing the unmanned vehicle to perform a crawling gait to cross the three-dimensional physical lock-up boundary includes: Execute the vehicle body pitch adjustment sequence to cause the unmanned vehicle body to generate a longitudinal tilt angle and maintain the longitudinal tilt angle; The independent wheel-end pulse output sequence is sent to drive the locally trapped wheel to generate periodic reciprocating tangential vibration, and the change in the unmanned vehicle's displacement to get out of trouble is monitored in real time. When the change in the escape displacement exceeds the preset escape distance threshold, the transmission of the vehicle pitch adjustment sequence is stopped and the output of the independent wheel-end pulse output sequence is terminated.

[0126] Preferably, the specific implementation process of enabling the unmanned vehicle to perform a crawling gait to cross the three-dimensional physical lock-up boundary is as follows: After the multimodal morphology reconstruction strategy is configured, the morphology reconstruction strategy record, the physical lock-up boundary record, the vehicle body pitch adjustment sequence, the independent wheel-end pulse output sequence, the failed and trapped node marker record, and the normal operating wheel margin record are read first; the morphology reconstruction strategy record is used to provide the temporal correspondence between the vehicle body pitch adjustment sequence and the independent wheel-end pulse output sequence, the physical lock-up boundary record is used to provide the three-dimensional physical lock-up boundary, the lock-up boundary continuous segment, the settlement depth scalar, the slip characteristic parameter, and the wheel-end attribution relationship, the failed and trapped node marker record is used to provide the wheel identifier of the wheel corresponding to the failed and trapped node and the wheel-end attribution relationship, and the normal operating wheel margin record is used to provide the current torque output margin of the normal operating wheels that are not marked as the failed and trapped node. By first reading the morphological reconstruction strategy record, the physical lock-up boundary record, the vehicle pitch adjustment sequence, and the independent wheel-end pulse output sequence, the subsequent execution of the vehicle pitch adjustment sequence, sending the independent wheel-end pulse output sequence, monitoring the change in the escape displacement, and stopping the sending of the vehicle pitch adjustment sequence and terminating the output of the independent wheel-end pulse output sequence all have the same strategy source and the same boundary source.

[0127] Preferably, before executing the vehicle pitch adjustment sequence, the target suspension support strut, the hydraulic stroke expansion command set, the suspension stroke timing record, the target suspension support strut lifting stage, the attitude holding stage, the target suspension support strut adjustment stage, and the trimmed suspension action amplitude parameters are read from the vehicle pitch adjustment sequence. The target suspension support strut originates from the quadrant position of the failed and trapped node in the chassis coordinate system. The hydraulic stroke expansion command set originates from multiple hydraulic stroke expansion commands generated according to a preset step size increment rule. The suspension stroke timing record is used to express the command execution order and attitude holding time of the multiple hydraulic stroke expansion commands. Subsequently, based on the suspension travel timing record, the set of hydraulic travel expansion commands is written into the suspension control issuance record according to the target suspension support strut identifier and the command execution order; the suspension control issuance record is used to issue segmented travel control data to the suspension controller, and the segmented travel control data continues to cause the suspension controller to execute the segmented lifting of the target suspension support strut and the coordinated adjustment of the cooperating target suspension support strut according to the vehicle body pitch adjustment sequence.

[0128] Preferably, the execution of the vehicle body pitch adjustment sequence does not involve adjusting the target suspension support strut to its maximum travel position all at once, but rather executing it segment by segment according to the current travel target value in the hydraulic travel expansion command set. After completing the travel adjustment corresponding to each hydraulic travel expansion command, the suspension travel feedback data returned by the suspension controller is read, and the suspension travel feedback data is checked against the corresponding current travel target value to form a suspension travel execution record. The suspension travel execution record includes the target suspension support strut identifier, the current travel target value, the suspension travel feedback data, the cumulative travel value, and the attitude holding time. The suspension travel execution record continues to participate in the determination of the longitudinal lift angle, enabling the unmanned vehicle body to generate a longitudinal lift angle corresponding to the three-dimensional physical lock boundary according to the vehicle body pitch adjustment sequence, rather than adjusting the attitude solely based on a fixed lift amplitude.

[0129] Preferably, when the autonomous vehicle body generates the longitudinal lift angle, the front-to-rear height difference of the chassis is determined based on the target suspension support strut identifier, the cumulative travel value, and the adjustment stage of the target suspension support strut in the suspension travel execution record, and the front-to-rear height difference of the chassis is written into the vehicle body attitude change record; the vehicle body attitude change record includes the target suspension support strut, the target suspension support strut, the cumulative travel value, the front-to-rear height difference of the chassis, the attitude holding time, and the three-dimensional physical lock-up boundary. Subsequently, a longitudinal lift angle state record is generated based on the vehicle body attitude change record, which is used to express the lift tilt state formed by the autonomous vehicle body in the front-to-rear direction of the chassis. The longitudinal lift angle state record is further used to determine whether the autonomous vehicle body attitude has entered the attitude holding stage, so that the independent wheel-end pulse output sequence can participate in wheel-end drive after the autonomous vehicle body attitude has moved away from the continuous segment of the lock-up boundary.

[0130] Preferably, while maintaining the longitudinal lift angle, the longitudinal lift angle state record, the locked boundary continuous segment, the suspension travel execution record, and the attitude maintenance range record are read. The attitude maintenance range record is pre-stored in the suspension travel parameter record, and the attitude maintenance range record is derived from the available travel range of the target suspension support strut, the chassis ground clearance safety threshold, and the spatial range of the locked boundary continuous segment. Subsequently, based on the longitudinal lift angle state record, it is determined whether the distance between the chassis lower surface boundary of the area where the target suspension support strut is located and the chassis lower surface boundary of the locked boundary continuous segment is within the attitude maintenance range corresponding to the attitude maintenance range record. If the distance between the chassis lower surface boundary is within the attitude maintenance range, the current travel target value of the target suspension support strut is maintained according to the attitude maintenance time. If the distance between the chassis lower surface boundary is lower than the attitude maintenance range, the hydraulic travel expansion command that has not yet been executed in the hydraulic travel expansion command set is read, and the next hydraulic travel expansion command is executed under the condition that it does not exceed the trimmed suspension action amplitude parameter. Therefore, the longitudinal lifting angle can be maintained in segments around the three-dimensional physical lock-up boundary, so that the subsequent wheel-end pulse drive action is under the chassis attitude conditions corresponding to the three-dimensional physical lock-up boundary.

[0131] Preferably, before sending the independent wheel-end pulse output sequence, the individual wheel-end pulse output records, discrete torque sampling segment records, torque peak target records, wheel-end action states, and wheel-end action sequence relationships in the independent wheel-end pulse output sequence are read first. The individual wheel-end pulse output records are derived from the result of chronologically arranging discrete torque sampling segments under different wheel identifiers according to the wheel-end action sequence relationship. The discrete torque sampling segment records are derived from the result of discretizing and slicing the continuous torque control signal waveform in the time dimension. The torque peak target record is used to provide the torque peak target and adhesion torque threshold corresponding to the wheel of the failed and trapped node. Subsequently, the individual wheel-end pulse output records are time-correlated with the longitudinal lifting tilt angle state records, so that the wheel-end pulse drive states in the individual wheel-end pulse output records are arranged within the attitude holding phase or after the target suspension support strut lifting phase, to form a wheel-end pulse transmission record. The wheel-end pulse transmission record is then used to send the independent wheel-end pulse output sequence to the hub motor controller.

[0132] Preferably, when sending the independent wheel-end pulse output sequence to the hub motor controller, the discrete torque sampling segments in the independent wheel-end pulse output sequence are sent piece by piece to the corresponding hub motor controller according to the wheel identifier, sampling time, torque target value, and wheel-end action state in the wheel-end pulse transmission record; the hub motor controller outputs the corresponding wheel-end drive torque according to the torque target value and wheel-end action state in the discrete torque sampling segment. For the wheel corresponding to the failed and trapped node, if the wheel-end action state is the wheel-end pulse drive state, a higher torque is output according to the torque peak target in the torque peak target record; if the wheel-end action state is the wheel-end brief stop state, a lower torque or paused torque output is output according to the reference torque position in the continuous torque control signal waveform. Multiple wheel-end pulse drive states and multiple wheel-end brief stop states alternate according to the sampling time, causing the wheel corresponding to the failed and trapped node to generate periodic reciprocating tangential vibration.

[0133] Preferably, when driving the wheel corresponding to the failed and trapped node to generate periodic reciprocating tangential vibration, the wheel identifier, the slip characteristic parameters, and the wheel end affiliation relationship of the wheel corresponding to the failed and trapped node are first read from the failed and trapped node marking record. Then, the lock-up boundary candidate map unit corresponding to the wheel corresponding to the failed and trapped node is read from the lock-up boundary continuous segment. The wheel end affiliation relationship is used to determine the hub motor controller corresponding to the wheel corresponding to the failed and trapped node. The slip characteristic parameters are used to determine the fluctuation period and fluctuation direction of the reciprocating tangential vibration. The lock-up boundary candidate map unit is used to determine the contact area of ​​the wheel corresponding to the failed and trapped node at the edge of the three-dimensional physical lock-up boundary. Subsequently, according to the fluctuation period and the fluctuation direction, the wheel end pulse drive state and the wheel end brief stop state in the independent wheel end pulse output sequence are arranged so that the wheel corresponding to the failed and trapped node forms the reciprocating tangential vibration along the tangential direction of the contact area. The reciprocating tangential vibration continues to participate in the formation of the change in the escape displacement, so that the unmanned vehicle applies a periodic wheel-end drive action to the edge of the three-dimensional physical lock boundary while maintaining the longitudinal lifting angle.

[0134] Preferably, when monitoring the change in the unmanned vehicle's displacement to overcome obstacles in real time, the initial trapped pose record of the unmanned vehicle is read before sending the independent wheel-end pulse output sequence. This initial trapped pose record is derived from the pose result formed by autonomous navigation positioning data, LiDAR positioning data, visual positioning data, and wheel odometer data at the same time reference. The autonomous navigation positioning data is derived from the positioning calculation data formed when the unmanned vehicle travels along the inspection route or autonomously planned inspection route. The LiDAR positioning data is derived from the positioning matching data obtained by scanning the surrounding terrain boundaries and obstacle boundaries using LiDAR. The visual positioning data is derived from the positioning matching data obtained by tracking the environmental texture in the inspection scene using a visual sensor. The wheel odometer data is derived from the rotation sampling results of each wheel. Subsequently, during the sending of the independent wheel-end pulse output sequence, the current vehicle pose record is continuously read according to a preset displacement monitoring cycle. This current vehicle pose record is also derived from the pose result formed by the autonomous navigation positioning data, LiDAR positioning data, visual positioning data, and wheel odometer data at the same time reference. The current vehicle pose record and the initial trapped pose record are processed by displacement difference analysis to form the escape displacement change. The escape displacement change is written into the escape displacement monitoring record and is used to determine whether the unmanned vehicle has crossed the three-dimensional physical lock boundary.

[0135] Preferably, the monitoring of the change in the escape displacement does not solely rely on the wheel odometer data, because when the wheel corresponding to the failed trapped node experiences tangential slippage, reading the wheel rotation alone would amplify the unmanned vehicle's displacement judgment result. Therefore, when forming the current vehicle pose record, the terrain boundary matching result in the lidar positioning data, the environmental texture tracking result in the visual positioning data, and the wheel end displacement estimation result in the wheel odometer data are subjected to consistency screening. The consistency screening reads the displacement consistency tolerance range pre-stored in the displacement monitoring parameter record. The displacement consistency tolerance range is derived from the displacement deviation range of the lidar positioning data, the visual positioning data, and the wheel odometer data under normal inspection conditions. If the displacement change corresponding to the wheel odometer data deviates from the displacement change corresponding to the lidar positioning data and the visual positioning data beyond the displacement consistency tolerance range, the participation of the wheel odometer data in the current vehicle pose record is reduced; if the lidar positioning data, the visual positioning data, and the wheel odometer data are within the displacement consistency tolerance range, they collectively form the current vehicle pose record. The resulting change in displacement can reduce the impact of wheel spin on the judgment of getting out of trouble.

[0136] Preferably, the departure distance threshold is pre-stored in the escape judgment parameter record before the unmanned vehicle inspection task is configured. The escape judgment parameter record is derived from the chassis passage parameter record, the boundary thickness of the three-dimensional physical lock boundary, the spatial range of the continuous segment of the lock boundary, and the map scale mapping record of the environmental contact stress distribution map. When pre-configuring the departure distance threshold, the minimum displacement requirement for the unmanned vehicle to leave the continuous segment of the lock boundary is first determined based on the coverage range of the three-dimensional physical lock boundary in the front-rear direction and the left-right direction of the chassis. Then, combined with the chassis ground clearance safety threshold and the map scale mapping record, the minimum displacement requirement is converted into a scale expression consistent with the change in escape displacement to form the departure distance threshold. The departure distance threshold continues to participate in the comparison and judgment of the change in escape displacement, ensuring that there is no scale inconsistency in the comparison between the change in escape displacement and the departure distance threshold.

[0137] Preferably, when the change in the escape displacement is greater than the preset escape distance threshold, the change in the escape displacement, the escape distance threshold, the current vehicle pose record, the longitudinal lift tilt angle record, and the current execution segment position of the independent wheel-end pulse output sequence are first written into the boundary crossing determination record. The boundary crossing determination record is used to indicate that the unmanned vehicle has undergone a displacement change relative to the initial trapped pose record sufficient to leave the locked boundary continuous segment. Subsequently, the three-dimensional physical locked boundary and the locked boundary continuous segment in the physical locked boundary record are read, and the current vehicle pose record is mapped to the chassis coordinate system or inspection map coordinate system where the three-dimensional physical locked boundary is located to determine whether the current vehicle pose record has crossed the outer edge range of the locked boundary continuous segment. If the current vehicle pose record has crossed the outer edge range of the locked boundary continuous segment, the boundary crossing determination record is written into the escape completion determination record. The escape completion determination record is further used to trigger the process of stopping the transmission of the vehicle pitch adjustment sequence and terminating the output of the independent wheel-end pulse output sequence.

[0138] Preferably, when stopping the transmission of the vehicle pitch adjustment sequence, the system first reads the traction completion determination record and the suspension travel timing record, and identifies the hydraulic travel expansion commands and coordination adjustment commands that have not yet been transmitted in the vehicle pitch adjustment sequence. For the hydraulic travel expansion commands and coordination adjustment commands that have not yet been transmitted, the system stops writing to the suspension control issuance record; for the hydraulic travel expansion commands that are already being executed, the system reads the suspension travel feedback data, and stops increasing the current travel feedback value in the suspension travel feedback data after it reaches the corresponding current travel target value. Subsequently, the vehicle pitch adjustment sequence stop status, the target suspension support strut identifier, the current travel feedback value, and the traction completion determination record are written to the suspension stop record, which continues to be used for subsequent attitude recovery or inspection driving status recovery. Through this stopping transmission method, the vehicle pitch adjustment sequence will not continue to increase chassis attitude changes after the unmanned vehicle has crossed the three-dimensional physical lock boundary.

[0139] Preferably, when terminating the output of the independent wheel-end pulse output sequence, the escape completion determination record, the wheel-end pulse transmission record, and the individual wheel-end pulse output record are first read, and the discrete torque sampling segments that have not yet been output under each wheel identifier are identified. For the discrete torque sampling segments that have not yet been output, transmission to the corresponding wheel hub motor controller is stopped; for the discrete torque sampling segments that have been transmitted and are being executed, a wheel-end output fall-back command is generated according to the reference torque position in the continuous torque control signal waveform record, causing the corresponding wheel hub motor controller to fall back from the wheel-end pulse drive state to the reference torque position. Subsequently, the wheel identifier, the terminated discrete torque sampling segment, the reference torque position, the wheel-end output fall-back command, and the escape completion determination record are written into the wheel-end pulse termination record. The wheel-end pulse termination record is used to indicate that the independent wheel-end pulse output sequence has stopped applying the periodic reciprocating tangential vibration to the wheel corresponding to the failed trapped node.

[0140] Preferably, after stopping the transmission of the vehicle pitch adjustment sequence and terminating the output of the independent wheel-end pulse output sequence, the system continues to read the suspension stop record, the wheel-end pulse termination record, the traction completion determination record, and the environmental contact stress distribution map, and performs a status check on the wheel-adjacent map region corresponding to the wheel of the failed traction node. The status check includes reading the terrain height status, torque load attributes, and wheel-end attribution relationship in the wheel-adjacent map region, and reading the current torque sampling value in the current torque feedback sequence data. If the torque load attributes and the current torque sampling value have fallen back from the continuously enhanced load state to the normal inspection range, and the current vehicle pose record has crossed the outer edge of the continuous segment of the lockout boundary, then the traction completion determination record is retained as a valid traction completion record. The valid traction completion record continues to be used to restore the unmanned vehicle's travel along the inspection route or the autonomously planned inspection route, ensuring a data connection between the creeping gait action and subsequent inspection travel.

[0141] Preferably, when the unmanned vehicle passes the edge of the sand and the front wheel forms the wheel corresponding to the failed and trapped node, the vehicle body pitch adjustment sequence is first executed, causing the target suspension support strut to rise segment by segment according to the hydraulic stroke expansion command set, and causing the unmanned vehicle body to form the longitudinal lifting tilt angle; then, during the attitude holding phase, the independent wheel-end pulse output sequence is sent, causing the front wheel to generate periodic reciprocating tangential vibration according to the discrete torque sampling segment record. During the execution of the reciprocating tangential vibration, the autonomous navigation positioning data, the lidar positioning data, the visual positioning data, and the wheel odometer data are continuously read to form the current vehicle body pose record, and the escape displacement change is formed according to the displacement difference between the current vehicle body pose record and the initial trapped pose record. When the escape displacement change is greater than the escape distance threshold and the current vehicle body pose record crosses the outer edge range of the continuous segment of the locked boundary, the transmission of the vehicle body pitch adjustment sequence is stopped, and the output of the independent wheel-end pulse output sequence is terminated.

[0142] Preferably, in all-weather complex terrain inspection scenarios, the data collected by lidar, visual sensors, and multi-sensor obstacle avoidance components are affected by changes in illumination, differences in ground reflection, partial occlusion, and short-term vehicle vibration. The current torque feedback sequence data also fluctuates due to short-term wheel-end bounce. Therefore, when the unmanned vehicle performs the creeping gait action, it does not directly use a single wheel speed or a single torque sampling value as the basis for escaping the obstacle. Instead, it forms the longitudinal lifting angle according to the vehicle pitch adjustment sequence, and drives the wheel corresponding to the failed and trapped node to generate periodic reciprocating tangential vibration according to the independent wheel-end pulse output sequence. The change in escaping displacement is formed by the autonomous navigation positioning data, lidar positioning data, visual positioning data, and wheel odometer data. The change in escaping displacement, the escape distance threshold, the three-dimensional physical lock boundary, and the continuous segment of the lock boundary jointly participate in the determination of escaping completion, so that stopping the transmission of the vehicle pitch adjustment sequence and terminating the output of the independent wheel-end pulse output sequence corresponds to the spatial result of the unmanned vehicle crossing the three-dimensional physical lock boundary.

[0143] like Figure 3 As shown, an unmanned vehicle multimodal motion planning and escape device for complex terrain is provided, comprising: The data acquisition module is used to acquire the point cloud data of the vehicle's undercarriage contour scan and to collect the current torque feedback sequence data of each wheel of the unmanned vehicle. The map generation module is used to map the vehicle underside contour scan point cloud data onto the current torque feedback sequence data to generate an environmental contact stress distribution map. The feature extraction module is used to extract the settlement depth scalar and the slip characteristic parameters of each wheel from the environmental contact stress distribution map. The strategy configuration module is used to determine the three-dimensional physical lock-up boundary of the chassis based on the settlement depth scalar and the slip characteristic parameters, and to configure a multi-modal morphological reconstruction strategy corresponding to the three-dimensional physical lock-up boundary. The multi-modal morphological reconstruction strategy includes a vehicle body pitch adjustment sequence and an independent wheel end pulse output sequence. The action execution module is used to send the vehicle body pitch adjustment sequence to the suspension controller according to a preset priority sequence, and to send the independent wheel end pulse output sequence to the wheel hub motor controller, so that the unmanned vehicle performs a creeping gait action to cross the three-dimensional physical lock boundary.

[0144] like Figure 4As shown, an electronic device includes: a memory, a processor, and an unmanned vehicle multimodal motion planning and escape program for complex terrain stored in the memory and executable on the processor, the unmanned vehicle multimodal motion planning and escape program for complex terrain configured to implement the steps of the unmanned vehicle multimodal motion planning and escape method for complex terrain as described in any one of the present applications.

[0145] like Figure 5 As shown, a computer-readable storage medium stores a multimodal motion planning and escaping program for unmanned vehicles in complex terrain. When executed by a processor, the multimodal motion planning and escaping program for unmanned vehicles in complex terrain implements the steps of the multimodal motion planning and escaping method for unmanned vehicles in complex terrain as described in any one of this application.

[0146] Figures 2-5 For an exemplary description, please refer to the above. Figure 1 This will not be elaborated upon here.

Claims

1. A multimodal motion planning and obstacle avoidance method for unmanned vehicles in complex terrain, characterized in that, The multimodal motion planning and obstacle avoidance method for unmanned vehicles facing complex terrain includes the following steps: Acquire the point cloud data of the undercarriage contour of the unmanned vehicle, and collect the current torque feedback sequence data of each wheel of the unmanned vehicle; The vehicle underside contour scan point cloud data is mapped onto the current torque feedback sequence data to generate an environmental contact stress distribution map. The settlement depth scalar and slip characteristic parameters of each wheel are extracted from the environmental contact stress distribution map. The three-dimensional physical lock-up boundary of the chassis is determined based on the settlement depth scalar and the slip characteristic parameters, and a multi-modal morphological reconstruction strategy corresponding to the three-dimensional physical lock-up boundary is configured. The multi-modal morphological reconstruction strategy includes a vehicle body pitch adjustment sequence and an independent wheel end pulse output sequence. According to the preset priority sequence, the vehicle pitch adjustment sequence is sent to the suspension controller, and the independent wheel-end pulse output sequence is sent to the wheel hub motor controller, so that the unmanned vehicle performs a creeping gait action to cross the three-dimensional physical lock-up boundary.

2. The multimodal motion planning and obstacle avoidance method for unmanned vehicles in complex terrain as described in claim 1, characterized in that, The step of mapping the vehicle underside contour scan point cloud data onto the current torque feedback sequence data to generate an environmental contact stress distribution map includes: Identify the three-dimensional coordinate nodes in the scanned point cloud data of the vehicle underside contour; Obtain the hardware acquisition timestamp information from the current torque feedback sequence data; Based on the hardware acquisition timestamp information, torque load attributes are configured for the three-dimensional coordinate nodes to obtain multiple enhanced three-dimensional coordinate nodes; Multiple enhanced three-dimensional coordinate nodes are spatially topologically stitched together to construct the environmental contact stress distribution map.

3. The multimodal motion planning and obstacle avoidance method for unmanned vehicles in complex terrain as described in claim 2, characterized in that, The step of configuring torque load attributes for the three-dimensional coordinate nodes based on the hardware-acquired timestamp information to obtain enhanced three-dimensional coordinate nodes includes: Retrieve the laser echo reception time of the three-dimensional coordinate nodes; Compare the laser echo reception time with the hardware acquisition timestamp information; When the difference between the laser echo reception time and the hardware acquisition timestamp information is within a preset tolerance range, the corresponding instantaneous torque value is extracted. The instantaneous torque value is written as a node attribute into the corresponding three-dimensional coordinate node to obtain the enhanced three-dimensional coordinate node.

4. The multimodal motion planning and obstacle avoidance method for unmanned vehicles in complex terrain as described in claim 1, characterized in that, The step of extracting the settlement depth scalar of each wheel from the environmental contact stress distribution map includes: Locate the first pixel coordinate representing the axle rotation center in the environmental contact stress distribution map; Locate the second pixel coordinate representing the actual terrain contact profile in the environmental contact stress distribution map. Calculate the vertical Euclidean distance from the first pixel coordinate to the second pixel coordinate; The settlement depth scalar is determined by subtracting the pre-stored standard tire radius constant from the vertical Euclidean distance.

5. The multimodal motion planning and obstacle avoidance method for unmanned vehicles in complex terrain as described in claim 1, characterized in that, The step of determining the three-dimensional physical lock-up boundary of the chassis based on the settlement depth scalar and the slip characteristic parameters includes: Determine whether the settlement depth scalar exceeds the preset chassis ground clearance safety threshold; If the settlement depth scalar exceeds the chassis ground clearance safety threshold, the corresponding wheel is marked as a failed and trapped node. Obtain the three-dimensional position coordinates of all the failed and trapped nodes; Construct the outer envelope geometric convex hull based on all the aforementioned three-dimensional position coordinates; The three-dimensional physical lock-up boundary of the chassis is determined based on the outer envelope geometric convex hull.

6. The multimodal motion planning and obstacle avoidance method for unmanned vehicles in complex terrain as described in claim 5, characterized in that, The steps for configuring the multimodal morphological reconstruction strategy corresponding to the three-dimensional physically locked boundary include: Calculate the projected area of ​​the outer envelope geometric convex hull on the horizontal plane; Calculate the entrapment ratio coefficient of the projected area relative to the total projected area of ​​the unmanned vehicle chassis; Based on the trapped ratio coefficient, a matching address is performed in a preset strategy mapping data table to determine the initial set of escape actions; Collect the current torque output margin of normally operating wheels that are not marked as the failed and trapped node; The action amplitude parameters in the initial set of escape actions are pruned and restricted based on the current torque output margin in order to configure the multimodal morphological reconstruction strategy.

7. The multimodal motion planning and obstacle avoidance method for unmanned vehicles in complex terrain as described in claim 6, characterized in that, The multimodal morphology reconstruction strategy includes the following methods for generating the vehicle pitch adjustment sequence: Analyze the quadrant positions of the failed and trapped nodes in the chassis coordinate system; The target suspension support strut that needs to be lifted is determined based on the quadrant position. A set of hydraulic stroke expansion commands for the target suspension support strut is generated according to a preset step-increment rule; The vehicle pitch adjustment sequence is generated based on the hydraulic stroke expansion command set.

8. A multimodal motion planning and obstacle avoidance device for unmanned vehicles tackling complex terrain, characterized in that: The unmanned vehicle multimodal motion planning and obstacle avoidance device for complex terrain includes: The data acquisition module is used to acquire the point cloud data of the vehicle's undercarriage contour scan and to collect the current torque feedback sequence data of each wheel of the unmanned vehicle. The map generation module is used to map the vehicle underside contour scan point cloud data onto the current torque feedback sequence data to generate an environmental contact stress distribution map. The feature extraction module is used to extract the settlement depth scalar and the slip characteristic parameters of each wheel from the environmental contact stress distribution map. The strategy configuration module is used to determine the three-dimensional physical lock-up boundary of the chassis based on the settlement depth scalar and the slip characteristic parameters, and to configure a multi-modal morphological reconstruction strategy corresponding to the three-dimensional physical lock-up boundary. The multi-modal morphological reconstruction strategy includes a vehicle body pitch adjustment sequence and an independent wheel end pulse output sequence. The action execution module is used to send the vehicle body pitch adjustment sequence to the suspension controller according to a preset priority sequence, and to send the independent wheel end pulse output sequence to the wheel hub motor controller, so that the unmanned vehicle performs a creeping gait action to cross the three-dimensional physical lock boundary.

9. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and an unmanned vehicle multimodal motion planning and escape program for complex terrain stored in the memory and executable on the processor, wherein the unmanned vehicle multimodal motion planning and escape program for complex terrain is configured to implement the steps of the unmanned vehicle multimodal motion planning and escape method for complex terrain as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a multimodal motion planning and escaping program for unmanned vehicles in complex terrain. When the multimodal motion planning and escaping program for unmanned vehicles in complex terrain is executed by a processor, it implements the steps of the multimodal motion planning and escaping method for unmanned vehicles in complex terrain as described in any one of claims 1 to 7.

Images