A hierarchical obstacle crossing control method for complex narrow space-oriented eight-wheel robot

Abstract

The application relates to the technical field of mobile robot autonomous navigation and obstacle crossing control, and particularly discloses a hierarchical obstacle crossing control method for an eight-wheel robot in a complex narrow space, which comprises the following steps: S1, obstacle sensing and terrain feature collection; S2, adaptive obstacle passing capacity determination and grade decision; S3, corresponding step control according to the output obstacle crossing grade; S3.1, first-level direct crossing adaptive control; S3.2, second-level auxiliary crossing adaptive control; S3.3, third-level enhanced crossing adaptive control, which comprises S3.3.1, step-by-step overturning lifting timing autonomous decision, S3.3.2, multi-wheel ground contact autonomous maintenance and S3.3.3, attitude safety autonomous constraint; S3.4, detour / parking adaptive strategy; and S4, autonomous attitude recovery and parameter calibration after crossing. The method is applied to a robot with an eight-wheel chassis, can improve the obstacle crossing continuity, passing performance and platform stability under the premise of ensuring safety, and has cross-terrain adaptive capability.

Description

Technical Field

[0001] This invention relates to the field of autonomous navigation and obstacle crossing control technology for mobile robots. Background Technology

[0002] Underground utility tunnels, collapsed building interiors, narrow alleyways, and semi-enclosed passages at natural disaster sites are often characterized by limited space, significant ground undulations, and diverse obstacle types. Ground-based detection robots designed for such scenarios not only need high obstacle-crossing capabilities but also need to maintain platform stability as much as possible during movement to ensure the reliable operation of onboard sensing equipment, communication devices, and payload modules. A single fixed obstacle-crossing mode is insufficient for engineering applications; therefore, a real-time, hierarchical obstacle-crossing control method for an eight-wheeled chassis is urgently needed.

[0003] Chinese Patent Application No. 202511473589.1 discloses a real-time graded obstacle avoidance control method and device for a tunnel lining inspection vehicle. This method constructs a multi-body dynamics digital twin model based on environmental data and vehicle parameters, collects environmental and vehicle status data in real time, drives the virtual model for synchronous simulation, obtains simulation detection information, executes corresponding graded obstacle avoidance strategies according to obstacle type, and generates control commands to be sent to the actuators. This method applies the concept of graded control to obstacle avoidance in complex equipment, featuring a combination of "non-stop obstacle avoidance strategy + stopping obstacle avoidance strategy," enabling differentiated actions for different obstacles. However, it also has the following shortcomings: First, the graded obstacle avoidance scheme mainly serves the avoidance actions of the upper robotic arm or inspection mechanism, and its grading focuses on obstacle category, insufficiently considering the obstacle passability, ground clearance margin, and top and lateral envelope limitations of the ground robot chassis. Second, this method cannot fully utilize the advantages of multi-wheeled chassis in complex terrain and is not suitable for eight-wheeled independent drive chassis applications. Third, in complex and confined spaces, robots are significantly limited by width and height. If a single, brute-force obstacle-crossing strategy is still employed, it can easily lead to vehicle jamming, attitude instability, or unnecessary stopping. Fourth, while simulation-heavy, virtual model-heavy solutions are suitable for large-scale equipment rehearsals, they present problems such as high system complexity, high deployment costs, and long real-time implementation paths for engineered ground robots with limited size and computing power. Fifth, the level thresholds (such as height thresholds H1, H2, H3) and control parameters (such as suspension lift angle, torque distribution ratio, and target wheel speed) of existing graded obstacle-crossing methods are mostly fixed values ​​calibrated offline by humans. They cannot perform online autonomous judgment and adaptive adjustment based on real-time terrain characteristics (such as changes in ground friction coefficient and differences in obstacle surface material) and vehicle dynamic responses (such as pitch angle change rate and wheel speed difference fluctuations). This results in rigid obstacle-crossing strategies when robots face unstructured, unknown, and complex terrain, easily leading to energy waste from "large movements over small obstacles" or failure to pass through "small movements over large obstacles," and poor cross-terrain capability of control parameters. Summary of the Invention

[0004] The purpose of this invention is to provide a hierarchical obstacle crossing control method for eight-wheeled robots with eight-wheeled chassis, which can improve obstacle crossing continuity, passability and platform stability while ensuring safety, and has cross-terrain adaptive capability, for use in complex and narrow spaces.

[0005] To achieve the above objectives, the technical solution of the present invention is: a hierarchical obstacle-crossing control method for an eight-wheeled robot in complex and narrow spaces, characterized by the following steps:

[0006] S1 Obstacle Perception and Terrain Feature Acquisition: Real-time acquisition of obstacle point cloud and / or contour data, as well as vehicle attitude, wheel speed, suspension travel and ground contact status data, and estimation of ground friction coefficient and obstacle surface material characteristics based on data analysis to form the obstacle feature set at the current moment;

[0007] S2 Adaptive Obstacle Passage Capability Determination and Level Decision: Based on the obstacle feature set, the obstacle crossing level is autonomously determined by online correction of the dynamic threshold and rule reasoning obtained through pre-set multi-level dynamic height threshold and spatial envelope safety threshold. The autonomous obstacle crossing level reasoning process includes: S2.1 Threshold boundary region rule reasoning; S2.2 Output obstacle level, including Level 1 direct passage, Level 2 assisted obstacle crossing, and Level 3 enhanced obstacle crossing, corresponding to the three levels of dynamic height thresholds, as well as detour / stop.

[0008] S3 performs corresponding step control based on the output obstacle crossing level;

[0009] S3.1 Level 1 directly uses obstacle crossing adaptive control to maintain the suspension reference posture and cross obstacles in an adaptive low-speed straight-line manner. During the crossing process, it continuously monitors the slippage situation and determines whether a level transition is required.

[0010] S3.2 Second-level assisted obstacle crossing adaptive control calculates the front suspension lift amount to pre-lift the suspension, corrects the drive torque in real time, adjusts the center of gravity to achieve front wheel engagement, and calculates the rear suspension compensation amount to adjust the rear suspension.

[0011] S3.3 Three-level enhanced obstacle crossing adaptive control; including S3.3.1 step-by-step flip-lift timing autonomous decision-making, S3.3.2 multi-round grounding autonomous maintenance and S3.3.3 attitude safety autonomous constraint; S3.3.1 step-by-step flip-lift timing autonomous decision-making includes the first step of head-up phase, the second step of first half of climbing, and the third step of tail bottoming prevention.

[0012] S3.4 Detour / Parking Adaptive Strategy;

[0013] S4 Autonomous Attitude Recovery and Parameter Calibration after Obstacle Crossing

[0014] Furthermore, the obstacle feature set mentioned in step S1 is denoted as...

[0015] ,in, The height of the obstacle. G represents the length of the obstacle along the direction of travel, and G represents the obstacle's slope / step characteristics. For spatial envelope constraint margin, This is an estimated value for the ground friction coefficient. This is for the amount of space at the top. This is for lateral space allowance. For vehicle stability margin, The effective grounding wheel number.

[0016] Furthermore, the three-level dynamic height threshold mentioned in step S2 is based on the offline calibration of the chassis structure parameters, and then corrected online according to the real-time ground friction coefficient and vehicle pitch angle. The expression is: Where k=1,2,3, corresponding to the upper limits of obstacle clearance for Level 1, Level 2, and Level 3 obstacles, respectively. This represents the k-th level foundation threshold determined offline by chassis structure parameters. For comprehensive correction factors, Indicates the coefficient of friction of the ground. Indicates the vehicle's pitch angle. , for The coefficient can be obtained by looking up a table or interpolating μ online. for The correction factor, according to | | You can obtain the table or interpolate it online.

[0017] Furthermore, the spatial envelope safety threshold is obtained offline by the chassis shape envelope dimensions, the maximum suspension lift travel, and the minimum safety clearances at the top and sides, and is corrected online based on real-time vehicle speed, vehicle attitude, and ground friction coefficient.

[0018] Furthermore, in step S2.1, when the obstacle height is within the dynamic threshold boundary region, the rule reasoning engine is activated. It makes a multi-factor decision by comprehensively considering the relative position of the obstacle height, the ground friction coefficient level, the vehicle pitch angle, the top space margin, the lateral space margin, and the number of effective ground contact wheels. When there is a rule conflict, the conservative strategy is executed first. When the spatial envelope constraint margin is less than the preset spatial envelope safety threshold, it is determined first to be a detour / parking level or to trigger the conservative strategy. When the ground friction coefficient is small or the absolute value of the vehicle pitch angle is large, the three-level dynamic height threshold is tightened and the conservative strategy is triggered; otherwise, it is appropriately relaxed.

[0019] Furthermore, in step S3.1, maintaining the suspension reference posture is achieved by the online parameter tuning unit matching and correcting the torque distribution ratio of the front wheel group and the middle wheel group from the preset parameter library based on the current ground friction coefficient and vehicle pitch angle. Specifically, the online parameter tuning unit has a built-in preset torque distribution table that stores the torque ratios of the front wheel group, middle wheel group, and rear wheel group under different combinations of ground friction coefficients and obstacle heights, expressed as:

[0020] , Indicates the drive torque of the front wheel assembly. , Indicates the drive torque of the intermediate wheel set. Indicates the drive torque of the rear wheel assembly. This represents the distribution coefficient of the drive torque of the front wheel assembly. , This represents the distribution coefficient of the drive torque of the intermediate wheel set. This indicates the distribution coefficient of the drive torque of the rear wheel assembly. Indicates the coefficient of friction of the ground. The obstacle height is determined by looking up a table based on real-time μ and h to match the basic ratio, and then linear interpolation correction is performed based on the current wheel set's contact force feedback with the ground to ensure that each wheel neither slips nor under-drives.

[0021] Furthermore, taking the adaptive low-speed straight-ahead obstacle crossing method as the target speed for calculation, the expression is: ,in, For the target speed, Based on cruise speed, The speed coefficient, which is adjusted online based on the friction coefficient and obstacle height, is obtained by the online parameter tuning unit by looking up and interpolating from a preset speed coefficient table.

[0022] Furthermore, during the passage, the system continuously monitors the wheel speed difference, pitch angle change rate, and grounding status. If the wheel speed difference exceeds the preset wheel speed difference threshold for more than a set time, or if the vehicle pitch angle exceeds the preset vehicle pitch angle threshold, it indicates that the attitude has exceeded the limit. If continuous slippage is detected, the system automatically triggers a level transition and autonomously upgrades to Level 2 Assisted Obstacle Crossing.

[0023] Furthermore, in step S3.2, the front suspension lift is calculated online based on the obstacle height, vehicle pitch angle, and desired approach angle. The desired approach angle is determined by the obstacle height, front wheel radius, ground clearance, and top clearance. The drive torque is corrected in real time based on the front wheel contact force feedback and slippage monitoring. During the transition phase, when the front wheel group begins to engage with the obstacle, the torque of the front wheel group is appropriately increased and the torque of the rear wheel group is decreased to achieve front wheel engagement. The rear suspension compensation is interpolated online based on the real-time monitoring value of the ground clearance at the center of the chassis. The vehicle pitch angle is constrained through IMU closed-loop control, and the PID parameters are matched online to the optimal group based on the degree of vehicle oscillation.

[0024] Furthermore, in step S3.3.1, the first step of the head-up phase: the lifting angle of the front suspension group is determined by the online parameter tuning unit based on the obstacle height, the current vehicle pitch angle, and the center of gravity projection position, by looking up and interpolating from a preset head-up angle table to form the optimal head-up posture; the second step of the first half of the climb: the front wheel group traction, the middle wheel group support, and the rear wheel group push, the torque ratio of each wheel is determined by the online parameter tuning unit based on the real-time wheel-to-ground contact force and the longitudinal acceleration of the vehicle body, by looking up and interpolating from a preset torque ratio table to ensure maximum traction and no slippage; the third step of the rear bottoming-out phase: when the middle part is close to the top of the obstacle, the system autonomously determines the timing of the rear suspension group lifting based on the real-time monitoring value of the chassis rear ground clearance and the predicted trajectory. If the system detects that there is an imminent risk of the rear touching the ground, the online parameter tuning unit immediately looks up the table from the preset rear suspension compensation table and outputs the lifting angle. If the vehicle posture already meets the conditions for passing, unnecessary lifting is skipped.

[0025] Furthermore, in step S3.3.2, the system monitors the grounding status of each wheel in real time, and the online parameter tuning unit autonomously adjusts the suspension height and torque distribution to ensure that no less than four drive wheels are in an effective grounding state. When a wheel shows a tendency to lift off the ground, the online parameter tuning unit looks up the preset fine-tuning table and outputs the suspension fine-tuning amount of that wheel and the torque compensation amount of the adjacent wheel.

[0026] And / or, in step S3.3.3, the vehicle pitch angle and lift angle are limited to a preset threshold, but the threshold boundary is adaptively widened or tightened online according to the current vehicle speed and the curvature of the top of the obstacle.

[0027] Furthermore, in step S3.4, if it is determined that the obstacle exceeds the allowable range of the chassis structure, or even if the suspension is raised, it will still cause a top collision, lateral jamming, or insufficient stability margin, or if the rule reasoning outputs the detour / parking level in step S2, then obstacle crossing will not be performed, and detour will be switched autonomously. The detour path selection is determined autonomously by the system. Based on the lateral space margin, ground flatness, and path length obtained by the environmental perception unit, the online parameter tuning unit matches the optimal detour mode from the preset detour strategy library. When the detour path is unavailable, parking and waiting or manual intervention will be performed.

[0028] Furthermore, in step S4, after the obstacle is cleared, attitude recovery, drive recovery, and data recording and parameter calibration are performed. Attitude recovery includes the suspension hydraulic cylinders gradually returning to the reference position, and the return speed is matched from the preset regression curve library by the online parameter tuning unit according to the current oscillation damping of the vehicle body. Drive recovery includes the drive output returning to the cruise mode, and the recovery process adopts a smooth transition curve, with torque and speed returning to the reference value in an exponential smoothing manner. Parameter calibration includes recording the data of the entire obstacle-crossing process, and using statistical analysis methods to automatically calibrate the correction coefficients and lookup parameters.

[0029] By adopting the above technical solution, the beneficial effects of this invention are as follows: Through a complete closed loop of "obstacle perception and terrain feature acquisition - adaptive obstacle passability determination and level decision - level-based obstacle crossing control - post-obstacle posture recovery and parameter calibration", the obstacle crossing process of the eight-wheeled robot is transformed from a fixed mode to an autonomous hierarchical control based on real-time terrain, vehicle posture, spatial envelope and stability constraints. Its beneficial effects are that it can autonomously select direct passage, assisted obstacle crossing, enhanced obstacle crossing or detour / stop strategy according to factors such as obstacle geometry, ground friction coefficient, top / lateral space margin, stability margin and effective grounding status. Thus, it can balance obstacle crossing continuity, passability and platform posture stability in complex and narrow spaces, reduce energy waste of "small obstacle big action" and failure to pass "big obstacle small action", and reduce the risk of jamming, collision and posture loss of control. At the same time, it can be realized in real time on the airborne controller without relying on complex digital twin pre-simulation, making engineering deployment simpler and cross-terrain adaptive capability stronger.

[0030] The above-mentioned further improvements in Level 1 obstacle crossing utilize a pre-set torque distribution table and speed coefficient table, combined with real-time ground friction coefficient, obstacle height, and wheel-ground contact force feedback for table lookup matching and linear interpolation correction. This enables autonomous torque distribution across eight wheels and adaptive low-speed straight-line travel. Furthermore, it autonomously triggers a level transition when wheel speed difference exceeds a threshold or pitch angle exceeds a limit, thereby achieving low-energy, smooth passage and improved continuity under low obstacles. In Level 2 obstacle crossing, the timing coordination of front suspension pre-lift, front wheel engagement, and rear suspension compensation, combined with the period... By observing the approach angle, front wheel contact force feedback, chassis center ground clearance, and IMU closed-loop attitude constraints, the risk of scraping and pitch angle fluctuations during medium obstacle crossings are effectively reduced. In level three obstacle crossings, through autonomous decision-making on step-by-step flip-up and lifting sequence, autonomous maintenance of multi-wheel ground contact, and autonomous attitude safety constraints, it is ensured that no less than four drive wheels are effectively grounded during high obstacle crossings, reducing the risk of tail bottoming out, jamming, and attitude instability. Furthermore, the attitude threshold is relaxed or tightened online according to the curvature of the obstacle top, improving the success rate and smoothness of high obstacle crossings.

[0031] Furthermore, by clearly defining obstacle feature sets and parameters such as spatial envelope margin, stability margin, and effective grounding wheel count, a unified input is provided for hierarchical decision-making and online parameter tuning; offline calibration and online correction of dynamic height thresholds and spatial envelope safety thresholds enhance the adaptability to complex and unknown terrains; rule-based reasoning and a conservative priority strategy during conflicts improve the safety of level switching; detour / stop strategies prevent forced obstacle crossing when it is not safe to pass; and smooth recovery of attitude and drive after obstacle crossing and self-calibration of historical data continuously optimize obstacle crossing performance and enhance the ability of control parameters to migrate across terrains. Detailed Implementation

[0032] To further explain the technical solution of the present invention, the present invention will be described in detail below through specific embodiments.

[0033] This embodiment discloses a hierarchical obstacle-crossing control method for an eight-wheeled robot in complex and narrow spaces. Its control system includes an environmental perception unit, a state estimation unit, an adaptive obstacle hierarchical decision-making unit, an online parameter tuning unit, a suspension execution unit, and a drive execution unit.

[0034] The environmental perception unit is used to acquire information such as the height and length of obstacles ahead, slope / step features, top space margin, lateral space margin, and ground material / friction coefficient estimation.

[0035] The state estimation unit is used to obtain vehicle pitch angle, roll angle, speed of each drive wheel, stroke of each suspension hydraulic cylinder, effective ground contact status of the wheels, and vehicle stability margin, etc.

[0036] The adaptive obstacle classification decision unit autonomously calculates and dynamically adjusts the obstacle crossing level judgment threshold based on obstacle geometric parameters, terrain features, spatial envelope, and stability constraints, and selects the best obstacle level.

[0037] The online parameter tuning unit has a built-in preset parameter library. Based on the current obstacle crossing level, real-time vehicle status and terrain feedback, it matches the optimal parameter set from the parameter library and performs online interpolation correction, outputting suspension lift angle sequence, torque distribution ratio of each wheel, target wheel speed and differential fine adjustment amount, etc.

[0038] The suspension actuator controls the hydraulic cylinders of the front and rear suspension groups to extend and retract in stages according to the set parameters.

[0039] The drive actuator controls the eight drive wheels to output torque and make differential fine adjustments according to the tuned settings.

[0040] The following describes in detail the graded obstacle crossing control process that influences the graded obstacle crossing control method of the eight-wheeled robot of this invention, including the following steps: S1 Obstacle perception and terrain feature acquisition

[0041] During its movement, the robot collects obstacle point cloud / contour data, as well as vehicle posture, wheel speed, suspension travel, and ground contact status data in real time. Simultaneously, it estimates the ground friction coefficient and obstacle surface material characteristics through wheel-ground contact force variation characteristics and visual texture analysis, forming the obstacle feature set for the current moment. This set serves as the input for S2 adaptive obstacle clearance determination and level decision-making, expressed as:

[0042] ,in, The height of the obstacle. G represents the length of the obstacle along the direction of travel, and G represents the obstacle's slope / step characteristics (such as edge sharpness and slope curvature). These three parameters are used to determine the obstacle's geometric scale. It is the spatial envelope constraint margin (or defined as the maximum spatial margin, which is synonymous with the spatial envelope constraint margin). This is used to characterize the tightest of the top / lateral spatial constraints. This is an estimate of the ground friction coefficient, used for dynamic threshold correction and torque distribution. This refers to the top space allowance (the minimum distance from the top of the vehicle body to any obstacle above it). Lateral space margin (the minimum distance from both sides of the vehicle body to lateral obstacles). The vehicle stability margin (characterizing the safety margin against rollover / slippage under the current attitude) is used for stability constraints. The effective grounding rounds are used for multi-factor rule reasoning.

[0043] This step provides real-time, multi-dimensional input for subsequent adaptive grading and online parameter tuning, enabling the robot to determine whether the height and width are sufficient, whether it will collide with the top, and whether it will get stuck laterally. It provides spatial constraints for dynamic threshold correction and rule-based reasoning in S2 and detour / stopping decisions in S6, avoiding reliance solely on fixed obstacle categories or single height thresholds for obstacle-crossing decisions. S2 Adaptive Obstacle Ability Determination and Grading Decision

[0044] Three levels of dynamic height thresholds are preset. , , The spatial envelope safety threshold and the three-level dynamic height threshold can be determined by offline calibration of the chassis structure parameters and then adjusted based on the real-time ground friction coefficient. and vehicle pitch angle Online correction, the expression is: Where k=1,2,3, corresponding to the upper limits of obstacle clearance for Level 1, Level 2, and Level 3 obstacles, respectively. This represents the k-th level foundation threshold determined offline by chassis structure parameters. For comprehensive correction factors, Indicates the coefficient of friction of the ground. Indicates the vehicle's pitch angle. , for The coefficient can be obtained by looking up a table or interpolating μ online. for The correction factor, according to | | The threshold can be obtained by online table lookup or interpolation and formed into a dynamic threshold.

[0045] Based on the obstacle feature set F, the robot autonomously determines the obstacle crossing level through dynamic threshold and rule reasoning, replacing the traditional fixed threshold determination method.

[0046] S2.1 Threshold Boundary Region Rule Reasoning

[0047] When h is within the dynamic threshold boundary region, the rule inference engine is activated, taking into account the relative position of the obstacle height, μ level, |θp|, top / lateral margin, and number of effective grounding wheels. When making multi-factor decisions, prioritize conservative strategies when rules conflict.

[0048] when When the obstacle height is less than the preset spatial envelope safety threshold, it indicates that the current spatial envelope constraint margin is insufficient. Regardless of whether the obstacle height is within the passable range, it should be prioritized as a detour / stopping level, or a conservative strategy should be triggered. The spatial envelope safety threshold can be obtained offline from chassis shape envelope dimensions, maximum suspension lift travel, minimum top / lateral safety clearance, etc., and can be corrected online based on real-time vehicle speed, vehicle attitude, and ground friction coefficient.

[0049] when Smaller or | When it is large The three-level dynamic height threshold is tightened, triggering a conservative strategy; conversely, it is appropriately relaxed to improve throughput. Larger and | When it is smaller, .

[0050] S2.2 Output Barrier Level

[0051] The output obstacle crossing levels include Level 1 direct passage, Level 2 assisted obstacle crossing, and Level 3 enhanced obstacle crossing, which correspond to three levels of dynamic height thresholds, as well as detour / stop.

[0052] This step enables adaptive hierarchical decision-making, allowing the robot to autonomously select the most suitable obstacle-crossing level in unknown and complex terrain, reducing "small obstacles with large movements" and "large obstacles with small movements," and solving problems such as fixed thresholds, single obstacle-crossing modes, and insufficient consideration of spatial envelope and stability constraints.

[0053] S3 performs corresponding control steps based on the obstacle crossing level output by S2.2.

[0054] S3.1 Level 1 directly uses obstacle-crossing adaptive control.

[0055] This level is suitable for low obstacles, enabling low-energy and smooth passage. If a risk arises, the obstacle-crossing level can be automatically upgraded to improve the continuity of passage. Details are as follows:

[0056] Maintaining the suspension's reference posture, the online parameter tuning unit adjusts the parameters based on the current ground friction coefficient μ and the vehicle's pitch angle. The system matches and corrects the torque distribution ratio between the front and middle wheel sets from a preset parameter library to achieve autonomous torque distribution. Specifically, the online parameter tuning unit has a built-in preset torque distribution table that stores the torque ratios of the front, middle, and rear wheel sets under different combinations of μ and h, expressed as follows:

[0057] , Indicates the drive torque of the front wheel assembly. , Indicates the drive torque of the intermediate wheel set. Indicates the drive torque of the rear wheel assembly. This represents the distribution coefficient of the drive torque of the front wheel assembly. , This represents the distribution coefficient of the drive torque of the intermediate wheel set. This represents the distribution coefficient of the drive torque of the rear wheel group. The base ratio is matched by looking up a table based on real-time μ and h, and then linear interpolation is performed based on the current wheel group's contact force feedback with the ground to ensure that each wheel neither slips nor under-drives. It should be noted that the wheel group of the eight-wheeled robot is divided into four groups: the front wheel group facing forward, the rear wheel group facing backward, and the middle wheel group between the front and rear wheel groups.

[0058] The target speed is calculated using an adaptive low-speed straight-ahead obstacle-crossing method. ,in Based on cruise speed, The speed coefficient, which is adjusted online based on the friction coefficient and obstacle height, is obtained by the online parameter tuning unit by looking up and interpolating from a preset speed coefficient table, thus achieving autonomous speed adjustment.

[0059] Continuously monitor wheel speed difference during passage. Pitch angle change rate and grounding status ,like The set time has been exceeded. The threshold for wheel speed difference, or This is due to an overshoot in posture. If the vehicle body pitch angle threshold is used to determine that continuous slippage has been detected, the system will automatically trigger a level transition, autonomously upgrading to level two assisted obstacle crossing, and simultaneously initiating online parameter tuning of S3.2 to achieve online monitoring and level transition.

[0060] S3.2 Second-level assisted obstacle crossing adaptive control.

[0061] Suitable for medium-height obstacles, the system autonomously executes the sequence of "front suspension pre-lift—front wheel engagement—rear suspension compensation." Each action parameter is matched in real-time from a preset parameter library by the online parameter tuning unit and corrected by linear interpolation, rather than using fixed values. Details are as follows:

[0062] The front suspension lift is calculated online based on obstacle height, vehicle pitch angle, and desired approach angle to achieve pre-lift of the front suspension. The desired approach angle can be determined by obstacle height, front wheel radius, ground clearance, and top clearance. The drive torque is corrected in real time based on front wheel contact force feedback and slippage monitoring. During the transition phase, the torque distribution center of gravity is autonomously adjusted to achieve front wheel engagement. That is, when the front wheel group begins to engage the obstacle, the torque of the front wheel group is appropriately increased and the torque of the rear wheel group is reduced to prevent front wheel slippage or vehicle body tilting backward. The rear suspension compensation is obtained by online interpolation based on the real-time monitoring value of the ground clearance at the center of the chassis to adjust the rear suspension and reduce the risk of scraping. Throughout the process, the vehicle pitch angle is constrained by IMU closed loop, and the PID parameters are matched online with the optimal group according to the degree of vehicle body oscillation to achieve closed-loop attitude stability.

[0063] S3.3 Level 3 Enhanced Obstacle Crossing Adaptive Control.

[0064] Suitable for obstacles that are relatively high but still passable, the system implements a step-by-step tilting and lifting strategy for the front and rear suspensions. The triggering timing, lifting angle, and drive torque for each step are autonomously determined by the online parameter tuning unit based on real-time conditions. This achieves autonomous decision-making for step-by-step tilting and lifting, multi-wheel contact maintenance, and attitude safety constraints, improving obstacle clearance success rate and reducing the risk of getting stuck and bottoming out. Details are as follows:

[0065] S3.3.1 Step-by-step flip-up lifting timing autonomous decision

[0066] In the first stage of head-up, the front suspension assembly lift angle is determined by the online parameter tuning unit based on the obstacle height, the current vehicle pitch angle, and the center of gravity projection position. The unit looks up the table from the preset head-up angle table, matches the values, and interpolates to correct the optimal head-up posture.

[0067] The first half of the second step involves climbing, with the front wheel group pulling, the middle wheel group supporting, and the rear wheel group pushing. The torque ratio of each wheel is adjusted by the online parameter tuning unit based on the real-time wheel-ground contact force and the longitudinal acceleration of the vehicle body. The unit looks up the preset torque ratio table and interpolates to ensure maximum traction and no slippage.

[0068] The third step is rear-end bottoming protection. When the middle section approaches the apex of the obstacle, the system autonomously determines the timing for the rear suspension to lift based on the real-time monitoring value of the chassis rear ground clearance and the predicted trajectory. If the system detects that there is an imminent risk of the rear touching the ground, the online parameter tuning unit immediately looks up the lifting angle from the preset rear suspension compensation table. If the vehicle body posture already meets the conditions for passing, unnecessary lifting is skipped to reduce energy consumption.

[0069] S3.3.2 Multi-round grounding autonomous maintenance

[0070] At this level, the system monitors the ground contact status of each wheel in real time, and the online parameter tuning unit autonomously adjusts the suspension height and torque distribution to ensure that no less than four drive wheels are in an effective ground contact state. When a wheel tends to lift off the ground, the online parameter tuning unit looks up the suspension fine-tuning amount of that wheel and the torque compensation amount of the adjacent wheel from the preset fine-tuning table to maintain the overall traction.

[0071] S3.3.3 Attitude Safety Autonomous Constraints

[0072] The vehicle's pitch and lift angles are limited to preset thresholds, but the threshold boundaries are adaptively widened or tightened online based on the current vehicle speed and the curvature of the obstacle top. For example, when the obstacle top has a rounded transition, a brief overshoot of the pitch angle is allowed (within the structural safety range) to improve smoothness of passage; when the top has a sharp edge, the threshold is tightened to ensure vehicle stability. S3.4 Adaptive detour / stop strategy.

[0073] If the obstacle is determined to exceed the allowable range of the chassis structure, or if raising the suspension will still cause top collision, lateral jamming, or insufficient stability margin, or if the rule reasoning in step S2 outputs a detour / parking level, then the forced obstacle crossing will not be executed. Instead, it will automatically switch to local detour, so as to switch to detour or parking in time when it is not safe to pass, reduce the risk of jamming and collision, and use the obstacle information for global path planning.

[0074] The detour route selection is determined autonomously by the system. Based on the lateral space margin, ground flatness and path length obtained by the environmental perception unit, the online parameter tuning unit matches the optimal detour mode (such as detouring to the left, detouring to the right, or reversing and then detouring) from the preset detour strategy library.

[0075] When the detour route is unavailable, a stop-and-wait or manual takeover will be initiated, and the current obstacle characteristics will be recorded in the environmental map for subsequent global path planning.

[0076] S4 Autonomous Attitude Recovery and Parameter Calibration After Obstacle Crossing

[0077] After overcoming the obstacle, attitude recovery, drive recovery, data recording, and parameter calibration are performed.

[0078] The suspension hydraulic cylinder gradually returns to the reference position. The return speed is matched by the online parameter tuning unit from the preset return curve library according to the current oscillation damping of the vehicle body, so as to realize the attitude recovery and avoid secondary attitude fluctuations caused by "hard return".

[0079] Drive output recovery cruise mode, the recovery process adopts a smooth transition curve, and torque and speed return to the reference value in an exponential smooth manner to achieve drive recovery.

[0080] The entire obstacle-crossing process data (obstacle feature F, selected level, tuned parameters, actual posture response, and success rate) is recorded. Through statistical analysis methods (such as sliding window averaging), the threshold correction coefficients k1, k2, and k3 in S2 and the lookup table parameters in S3.2-S3.3 are autonomously calibrated to achieve data recording and parameter calibration. This enables the robot to smoothly recover after obstacle crossing and performs parameter self-calibration, continuously optimizing and improving its obstacle-crossing performance with use, and making increasingly accurate obstacle-crossing decisions in similar terrain.

[0081] The technical solutions of the above embodiments of the present invention have the following advantages:

[0082] (1) The controlled object is more in line with actual needs. Existing solutions are designed for obstacle avoidance by the upper structure of tunnel inspection vehicles, while the present invention is designed for obstacle-crossing action of the eight-wheeled freewheel mobile platform body, which is more suitable for detection robots in complex and narrow spaces.

[0083] (2) The classification criteria are more in line with the obstacle crossing characteristics of the chassis and have adaptive capabilities. The present invention takes obstacle geometry, spatial envelope, ground clearance margin, stability margin and real-time terrain characteristics (friction coefficient, material) as the core criteria. The threshold can be dynamically corrected online, rather than just determining the action according to a fixed category and a fixed threshold. It is more targeted and has better cross-terrain adaptability.

[0084] (3) Autonomous decision-making and adaptive adjustment are realized. Existing solutions rely on manually preset strategies and fixed parameters. This invention enables the robot to autonomously match and correct the suspension lifting angle, torque distribution ratio and speed adjustment coefficient from the preset parameter library according to the real-time status through an online parameter tuning unit and rule reasoning engine, thus realizing the autonomous generation and adaptive adjustment of the control strategy.

[0085] (4) The execution chain is shorter and the engineering deployment is simpler. This invention can be directly implemented on the airborne sensing and chassis controller, and can complete real-time decision-making without relying on complex digital twin pre-simulation, which is convenient for engineering implementation.

[0086] (5) The suspension and drive are more coordinated and have the ability to self-calibrate parameters. By step-by-step tilting and lifting of the front and rear suspensions and the coordinated distribution of torque to eight wheels, combined with the data recording and parameter calibration mechanism after overcoming obstacles, the success rate of overcoming obstacles can be improved and the vehicle attitude fluctuation can be reduced. Moreover, the obstacle-crossing performance is continuously optimized with use.

[0087] (6) More suitable for safe passage through narrow spaces. The present invention incorporates top and lateral envelope constraints into the decision conditions, and can switch to detour in a timely manner when it is not safe to pass, thereby reducing the risk of getting stuck and colliding.

[0088] The above embodiments are not intended to limit the product form and style of the present invention. Any appropriate changes or modifications made by those skilled in the art should be considered as not departing from the patent scope of the present invention.

Claims

1. A hierarchical obstacle-crossing control method for an eight-wheeled robot in complex and narrow spaces, characterized in that, The steps are as follows: S1 Obstacle Perception and Terrain Feature Acquisition: Real-time acquisition of obstacle point cloud and / or contour data, as well as vehicle attitude, wheel speed, suspension travel and ground contact status data, and estimation of ground friction coefficient and obstacle surface material characteristics based on data analysis to form the obstacle feature set at the current moment; S2 Adaptive Obstacle Passage Assessment and Level Decision; Based on the obstacle feature set, the obstacle crossing level is autonomously determined by online correction of the dynamic threshold and spatial envelope safety threshold obtained by pre-set multi-level dynamic height threshold and rule reasoning. The process of autonomously determining the obstacle crossing level includes: S2.1 rule reasoning of threshold boundary area; S2.2 outputting the obstacle level, including Level 1 direct passage, Level 2 assisted obstacle crossing and Level 3 enhanced obstacle crossing, corresponding to the three levels of dynamic height threshold, as well as detour / stop. S3 performs corresponding step control based on the output obstacle crossing level; S3.1 Level 1 directly uses obstacle crossing adaptive control to maintain the suspension reference posture and cross obstacles in an adaptive low-speed straight-line manner. During the crossing process, it continuously monitors the slippage situation and determines whether a level transition is required. S3.2 Second-level assisted obstacle crossing adaptive control calculates the front suspension lift amount to pre-lift the suspension, corrects the drive torque in real time, adjusts the center of gravity to achieve front wheel engagement, and calculates the rear suspension compensation amount to adjust the rear suspension. S3.3 Three-level enhanced obstacle crossing adaptive control; including S3.3.1 step-by-step flip-lift timing autonomous decision-making, S3.3.2 multi-round grounding autonomous maintenance and S3.3.3 attitude safety autonomous constraint; S3.3.1 step-by-step flip-lift timing autonomous decision-making includes the first step of head-up phase, the second step of first half of climbing, and the third step of tail bottoming prevention. S3.4 Detour / Parking Adaptive Strategy; S4 Autonomous Attitude Recovery and Parameter Calibration after Obstacle Crossing 2. The hierarchical obstacle-crossing control method for an eight-wheeled robot in complex and narrow spaces as described in claim 1, characterized in that, The obstacle feature set mentioned in step S1 is denoted as ,in, The height of the obstacle. G represents the length of the obstacle along the direction of travel, and G represents the obstacle's slope / step characteristics. For spatial envelope constraint margin, This is an estimated value for the ground friction coefficient. This is for the amount of space at the top. This is for lateral space allowance. For vehicle stability margin, The effective grounding wheel number.

3. The hierarchical obstacle-crossing control method for an eight-wheeled robot in complex and narrow spaces as described in claim 1, characterized in that, The three-level dynamic height threshold mentioned in step S2 is based on the offline calibration of chassis structure parameters, and then corrected online according to the real-time ground friction coefficient and vehicle pitch angle. The expression is: Where k=1,2,3, corresponding to the upper limits of obstacle clearance for Level 1, Level 2, and Level 3 obstacles, respectively. This represents the k-th level foundation threshold determined offline by chassis structure parameters. For comprehensive correction factors, Indicates the coefficient of friction of the ground. Indicates the vehicle's pitch angle. , for The coefficient can be obtained by looking up a table or interpolating μ online. for The correction factor, according to | | Obtained via online table lookup or interpolation; And / or, the spatial envelope safety threshold is obtained offline by the chassis shape envelope dimensions, the maximum suspension lift travel, and the minimum safety clearances at the top and sides, and is corrected online based on real-time vehicle speed, vehicle attitude, and ground friction coefficient.

4. The hierarchical obstacle-crossing control method for an eight-wheeled robot in complex and narrow spaces as described in claim 1, characterized in that, In step S2.1, when the obstacle height is within the dynamic threshold boundary region, the rule reasoning engine is activated. It makes a multi-factor decision by comprehensively considering the relative position of the obstacle height, the ground friction coefficient level, the vehicle pitch angle, the top space margin, the lateral space margin, and the number of effective ground contact wheels. When there is a rule conflict, the conservative strategy is executed first. When the spatial envelope constraint margin is less than the preset spatial envelope safety threshold, it is determined first to be a detour / parking level or to trigger a conservative strategy. When the ground friction coefficient is small or the absolute value of the vehicle pitch angle is large, the three-level dynamic height threshold is tightened and the conservative strategy is triggered; otherwise, it is appropriately relaxed.

5. A hierarchical obstacle-crossing control method for an eight-wheeled robot in complex and narrow spaces as described in any one of claims 1-4, characterized in that, In step S3.1, maintaining the suspension reference posture is achieved by the online parameter tuning unit matching and correcting the torque distribution ratio of the front and middle wheel sets from a preset parameter library based on the current ground friction coefficient and vehicle pitch angle. Specifically, the online parameter tuning unit has a built-in preset torque distribution table that stores the torque ratios of the front, middle, and rear wheel sets under different combinations of ground friction coefficients and obstacle heights, expressed as follows: ,in, Indicates the drive torque of the front wheel assembly. , Indicates the drive torque of the intermediate wheel set. Indicates the drive torque of the rear wheel assembly. This represents the distribution coefficient of the drive torque of the front wheel assembly. , This represents the distribution coefficient of the drive torque of the intermediate wheel set. This indicates the distribution coefficient of the drive torque of the rear wheel assembly. Indicates the coefficient of friction of the ground. The obstacle height is determined by looking up a table based on real-time μ and h to match the basic ratio, and then linear interpolation correction is performed based on the current wheel set's contact force feedback with the ground to ensure that each wheel neither slips nor under-drives. And / or, in step S3.1, the target speed is calculated using the adaptive low-speed straight-ahead obstacle crossing method, and the expression is: ,in, For the target speed, Based on cruise speed, The speed coefficient, which is adjusted online based on the friction coefficient and obstacle height, is obtained by the online parameter tuning unit by looking up and interpolating from a preset speed coefficient table. And / or, in step S3.1, during the continuous monitoring of the passing process, the wheel speed difference, pitch angle change rate and grounding status are monitored. If the wheel speed difference is greater than the preset wheel speed difference threshold for more than a set time or the vehicle pitch angle is greater than the preset vehicle pitch angle threshold, it is considered that the attitude has exceeded the limit. It is judged that continuous slippage has been detected, and the system automatically triggers the level transition and autonomously upgrades to the second level of assisted obstacle crossing.

6. A hierarchical obstacle-crossing control method for an eight-wheeled robot in complex and narrow spaces as described in any one of claims 1-4, characterized in that, In step S3.2, the front suspension lift is calculated online based on the obstacle height, vehicle pitch angle, and desired approach angle. The desired approach angle is determined by the obstacle height, front wheel radius, ground clearance, and top clearance. The drive torque is corrected in real time based on the front wheel contact force feedback and slippage monitoring. During the transition phase, when the front wheel set begins to engage with the obstacle, the torque of the front wheel set is appropriately increased and the torque of the rear wheel set is decreased to achieve front wheel engagement. The rear suspension compensation is interpolated online based on the real-time monitoring value of the ground clearance at the center of the chassis. The vehicle pitch angle is constrained by the IMU closed loop, and the PID parameters are matched online to the optimal set based on the degree of vehicle oscillation.

7. A hierarchical obstacle-crossing control method for an eight-wheeled robot in complex and narrow spaces as described in any one of claims 1-4, characterized in that, In step S3.3.1, the first step of the head-up phase is as follows: the front suspension assembly lift angle is determined by the online parameter tuning unit based on the obstacle height, the current vehicle pitch angle, and the center of gravity projection position, by looking up a table in the preset head-up angle table and interpolating to correct the optimal head-up posture. The second step, the first half of the climb, is achieved by the front wheel assembly pulling, the middle wheel assembly supporting, and the rear wheel assembly pushing. The torque ratio of each wheel is determined by the online parameter tuning unit based on the real-time wheel-to-ground contact force and the vehicle's longitudinal acceleration, by looking up a table in the preset torque ratio table and interpolating to correct the torque ratio, ensuring maximum traction and preventing slippage. The third step, rear bottoming-out, involves the system autonomously determining the timing of rear suspension assembly lift when the middle section approaches the apex of the obstacle based on the real-time monitoring value of the chassis rear ground clearance and the predicted trajectory. If the system detects an imminent risk of the rear touching the ground, the online parameter tuning unit immediately looks up a table in the preset rear suspension compensation table and outputs the lift angle. If the vehicle posture already meets the passability requirements, unnecessary lifts are skipped.

8. A hierarchical obstacle-crossing control method for an eight-wheeled robot in complex and narrow spaces as described in any one of claims 1-4, characterized in that, In step S3.3.2, the system monitors the grounding status of each wheel in real time. The online parameter tuning unit autonomously adjusts the suspension height and torque distribution to ensure that no less than four drive wheels are in an effective grounding state. When a wheel shows a tendency to lift off the ground, the online parameter tuning unit looks up the preset fine-tuning table and outputs the suspension fine-tuning amount of that wheel and the torque compensation amount of the adjacent wheel. And / or, in step S3.3.3, the vehicle pitch angle and lift angle are limited to a preset threshold, but the threshold boundary is adaptively widened or tightened online according to the current vehicle speed and the curvature of the top of the obstacle.

9. A hierarchical obstacle-crossing control method for an eight-wheeled robot in complex and narrow spaces as described in any one of claims 1-4, characterized in that, If, in step S3.4, it is determined that the obstacle exceeds the allowable range of the chassis structure, or that even if the suspension is raised, it will still cause a top collision, lateral jamming, or insufficient stability margin, or if the rule reasoning outputs the detour / parking level in step S2, then obstacle crossing will not be performed, and detour will be switched autonomously. The detour path selection is determined autonomously by the system. Based on the lateral space margin, ground flatness, and path length obtained by the environmental perception unit, the online parameter tuning unit matches the optimal detour mode from the preset detour strategy library. When the detour path is unavailable, parking and waiting or manual intervention will be performed.

10. A hierarchical obstacle-crossing control method for an eight-wheeled robot in complex and narrow spaces as described in any one of claims 1-4, characterized in that, In step S4, after the obstacle is cleared, attitude recovery, drive recovery, data recording, and parameter calibration are performed. Attitude recovery includes the suspension hydraulic cylinders gradually returning to the reference position. The return speed is matched by the online parameter tuning unit from the preset regression curve library according to the current oscillation damping of the vehicle body. Drive recovery includes restoring the drive output to cruise mode. The recovery process uses a smooth transition curve, and torque and speed return to the baseline value in an exponential smoothing manner. Parameter calibration includes recording the data of the entire obstacle crossing process and using statistical analysis methods to automatically calibrate correction coefficients and lookup parameters.

Images