Self-adaptive vehicle speed control method for special robot based on road roughness index

Abstract

The application discloses a special robot self-adaptive vehicle speed control method based on road bumping index, and relates to the technical field of special robots.The application extracts four types of complementary features to calculate a dynamic bumping index through multi-source fusion of a laser radar, an inertial measurement unit, a track tension sensor and a visual camera;an actual mass center is dynamically estimated based on a preset storage state mass center position in combination with a real-time posture of a mechanical arm to adaptively adjust a bumping grade boundary;maximum safe vehicle speeds considering postures, accelerations and impact energy constraints are generated, and after multi-target correction, model predictive control is adopted to realize collaborative optimization of vehicle speed, a mechanical arm end posture and an optoelectronic load pan-tilt;simultaneously, sensor health monitoring and degradation protection, a forward-looking path re-planning based on a bumping thermal diagram, federal learning cloud end bumping prediction and system parameter self-calibration functions are integrated.

Description

Technical Field

[0001] This invention relates to the field of special robot technology, specifically to an adaptive speed control method for special robots based on road surface bump index. Background Technology

[0002] Tracked special-purpose robots are widely used in complex operational scenarios such as reconnaissance and surveying, urban counter-terrorism, fire rescue, and emergency bomb disposal. These scenarios often involve diverse road conditions, including rubble, potholes, steep slopes, and muddy terrain. When robots travel on such surfaces, improper speed control can lead to severe shaking, damaging onboard precision detection instruments and photoelectric payloads. In severe cases, it can also cause rollovers, skidding, and other safety accidents, directly impacting the progress of reconnaissance and rescue missions and personnel safety.

[0003] Currently, the speed control of tracked special robots generally has significant shortcomings. Most solutions adopt a fixed speed or a simple graded speed adjustment method of faster speed on flat roads and slower speed on bumpy roads. This cannot accurately match the speed according to the actual bumpiness of the road surface, and is always caught in the dilemma of either sacrificing work efficiency for safety or creating safety hazards for efficiency. At the same time, the road condition perception capabilities of existing technologies are very limited. Many rely solely on the inertial measurement unit to detect vehicle vibration to judge road conditions, and cannot identify special road surfaces such as sand and mud that appear flat but are bumpy when driven on, resulting in frequent misjudgments. Moreover, the influence of the robot's own state is completely ignored. The tolerance to bumps varies greatly when fully loaded and unloaded, and when the center of gravity is normal and deviated, yet the same speed standard is still used. Ultimately, this leads to safety accidents when the robot is carrying heavy detection equipment, while it cannot fully utilize its work efficiency when unloaded.

[0004] In addition, the existing control method suffers from the problem of neglecting overall coordination, adjusting only the vehicle speed without integrating it with the robotic arm's end effector posture and the photoelectric payload gimbal. Even when the speed is reduced, the detection equipment still experiences severe shaking, leading to blurred images and distorted detection data. Furthermore, the system lacks the ability to anticipate problems, only reacting passively to slow down after the tracks hit bumpy surfaces, resulting in inconsistent speeds that affect stability and pose safety risks. Moreover, its overall reliability is poor, lacking corresponding sensor health monitoring and failure protection mechanisms. If even one sensor malfunctions, the entire speed control system will fail, leaving the robot stranded and awaiting rescue. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an adaptive speed control method for special robots based on road surface bump index, thus solving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a special robot adaptive speed control method based on road surface bump index, comprising the following steps: S1. Multi-source sensor data acquisition and preprocessing: Real-time acquisition of road surface and vehicle body status data through lidar, inertial measurement unit, track tension sensor and vision camera, and time synchronization, outlier removal and spatiotemporal registration; S2. Multi-feature fusion dynamic bump index calculation: Extract road elevation change features, vehicle vibration energy features, track tension fluctuation features, and road texture roughness features, and obtain the dynamic bump index through adaptive weighted fusion. ; S3. Load and attitude adaptive bump level classification: Based on the robot's preset storage and transportation state center of mass position, combined with the real-time attitude dynamic estimation of the actual center of mass position of the robotic arm, the bump level boundary value is adaptively adjusted to divide the road surface into four bump levels. S4. Generation of maximum safe vehicle speed with multiple constraints: Based on the bump level and real-time vehicle attitude, generate the maximum safe vehicle speed threshold that takes into account acceleration constraints and impact energy limits. S5. Multi-target dynamic speed correction: The maximum safe speed threshold is dynamically corrected by taking into account the urgency of the task, the stability of the robotic arm and the center of gravity offset. S6. Vehicle speed smoothing and load attitude coordinated control: Model predictive control is used to generate a smooth vehicle speed sequence, and at the same time, feedforward commands for end-effector attitude compensation and feedforward commands for photoelectric load gimbal stabilization are output to achieve coordinated optimization of vehicle speed and load attitude. S7. Sensor health monitoring and failure protection: Real-time assessment of the health of each sensor, and automatic switching to degraded operation mode when multi-source fusion is abnormal; S8. Proactive route replanning: Construct an environmental turbulence heatmap based on historical turbulence data, and replan a low-turbulence route before entering a high-turbulence area; S9. Cloud-based collaborative turbulence prediction: By training a regional turbulence prediction model through federated learning, it enables proactive speed adjustment in untouched areas. S10. System parameter self-calibration: Utilizes historical driving data and online feedback to automatically calibrate control model parameters, improving long-term operational accuracy.

[0007] Preferably, the dynamic turbulence index in step S2 The calculation formula is as follows: ; This is a composite bump index that integrates multiple features, comprehensively quantifying the impact of road surface unevenness, vehicle vibration intensity, track tension fluctuations, and road texture characteristics on driving; among which: These are characteristic values ​​of road surface elevation changes. The characteristic value of vehicle body vibration energy. This represents the characteristic value of track tension fluctuation. This represents the surface texture roughness characteristic value; The dynamic weighting coefficients satisfy the following conditions: ; Step S2 further includes the following specific steps: S2-0. Perform statistical filtering on the original point cloud of the lidar to remove distance outliers and outliers, and retain the effective point cloud area. S2-1. Perform ground segmentation and elevation interpolation on the denoised lidar point cloud to generate a continuous road surface elevation sequence. Calculate the mean of the absolute values ​​of the second derivative of the elevation using a sliding window to obtain the characteristic values ​​of road surface elevation changes. ; S2-2. The IMU triaxial acceleration signal is detrended to eliminate the influence of gravitational acceleration and uniform motion. Then, a fast Fourier transform is performed to extract the energy spectral density integral of the load device's sensitive frequency band from 0.5Hz to 20Hz, thus obtaining the vehicle body vibration energy characteristic value. ; S2-3. Perform high-pass filtering on the track tension sensor signal to eliminate static load interference, and calculate the mean of the root mean square values ​​of the left and right track tension fluctuations to obtain the characteristic values ​​of track tension fluctuations. ; S2-4. Perform grayscale conversion, Gaussian filtering, and edge detection on the road surface images acquired by the visual camera, and calculate the texture entropy value of the gray-level co-occurrence matrix to obtain the road surface texture roughness feature value. The formula for calculating texture entropy is: ; Texture entropy reflects the uniformity and complexity of the gray-level distribution in a road surface image; a higher entropy value indicates a coarser road surface texture. For gray levels, Elements in the gray-level co-occurrence matrix The probability of occurrence can effectively distinguish special road surfaces such as sand and mud that are difficult for traditional sensors to identify; S2-5. Calculate the weight coefficients dynamically based on the signal-to-noise ratio corresponding to each feature value. The higher the signal-to-noise ratio, the more reliable the feature data, and the greater the corresponding weight. The sum of all weight coefficients is 1. S2-6. Apply second-order exponential smoothing to the weight coefficient sequence. By setting different smoothing factors, balance the response speed and smoothing effect, suppress weight abrupt changes, and avoid vehicle speed jitter.

[0008] Preferably, the adaptive calculation formula for the bump level boundary value in step S3 is as follows: ; The system dynamically adjusts the bump level boundary value based on load and center of gravity status. The greater the load and the more severe the center of gravity shift, the higher the driving risk corresponding to the same bump index, and the lower the boundary value is accordingly. Specifically: =1,2,3, corresponding to the three dividing points of the fourth level of turbulence; Define boundary values ​​for unloaded flat surfaces; The current payload mass; ρ is the robot's basic mass; ρ is the load sensitivity coefficient. This is the centroid offset. The reference centroid position for the pre-defined storage and transportation state; τ is the centroid offset sensitivity coefficient; Step S3 further includes the following specific steps: S3-0. Based on the preset reference center of mass position for storage and transportation, and combined with the joint angles, joint mass distribution, and end-effector load mass of the robotic arm, the actual center of mass position of the robot is calculated in real time using a rigid body dynamics model. The center of mass offset is then obtained by comparing it with the reference center of mass position. ; S3-1. During the initial operation of the equipment, the no-load boundary values ​​are obtained through standard bumpy road test calibration. ; S3-2. Before starting a single task, complete the no-load boundary values ​​based on IMU vibration data under no-load conditions. Rapid self-calibration; S3-3. During equipment operation, load mass and robotic arm posture data are collected in real time, and the center of gravity position and bump level boundary values ​​are dynamically updated. S3-4. When the center of gravity offset exceeds the preset safety threshold, all bump level boundary values ​​will be uniformly reduced by 15% to achieve conservative speed control. S3-5. Apply upper and lower limit constraints to the boundary values ​​of the bump level to ensure... Furthermore, the boundary value update frequency is limited to no more than 1Hz to avoid frequent fluctuations that could lead to unstable vehicle speed.

[0009] Preferably, the maximum safe vehicle speed threshold in step S4 It is generated using a piecewise nonlinear function, and the calculation formula is as follows: ; An exponential decay model is used to achieve a smooth transition of vehicle speed with the bump rate, avoiding sudden changes in vehicle speed caused by piecewise linear functions; where: The robot's nominal maximum speed; The bump attenuation coefficient controls the rate at which vehicle speed decreases with increasing bump index. It is the safety margin factor and 0 < <1, leaving a safety margin; This represents the minimum controllable speed for the robot. Step S4 further includes the following specific steps: S4-1. Real-time acquisition of vehicle pitch angle θ and roll angle φ, and attitude correction of bump attenuation coefficient κ, using the following correction formula: ; When the vehicle body pitches or rolls, its driving stability decreases, requiring an increased bump damping coefficient and a further reduction in safe speed; among which... Based on the attenuation value, To adjust the gain, These are the robot's maximum allowable pitch angle and maximum roll angle, respectively. S4-2. Apply vehicle speed acceleration constraints, limiting the rate of change of vehicle speed to no more than ± To avoid vehicle instability and load tipping caused by sudden acceleration or deceleration; S4-3. Calculate the road impact energy corresponding to the current vehicle speed in real time. If the impact energy exceeds the limit, the maximum safe vehicle speed will be lowered again. The impact energy calculation formula is: ; The road impact energy is proportional to the robot's total mass, the square of its speed, and the bump index, reflecting the impact intensity of road bumps on the robot's structure and the cargo it carries; among which... For the total mass of the robot, Current vehicle speed; S4-4. Apply upper and lower limit constraints to the maximum safe speed to ensure .

[0010] Preferably, the multi-target dynamic vehicle speed correction in step S5 includes, in sequence, task urgency compensation, robotic arm stability compensation, and center of gravity offset compensation. S5-1. Construct a formula for assessing task urgency: ; The urgency of a task is assessed by combining the remaining time and the progress already completed; the less time remaining and the slower the progress, the higher the urgency. Specifically: Remaining task time, The deadline for the task. This represents the distance already traveled. The total distance to the target. As a weighting coefficient; apply upper and lower limits to the urgency of the task to ensure The urgency of the mission compensates for the vehicle speed as follows: , This is an adjustable urgency gain coefficient; S5-2. Construct the formula for calculating the stability factor of the robotic arm: ; The stability of the robotic arm is evaluated by combining its rotational inertia and external overturning moment. The smaller the stability factor, the more prone the robotic arm is to instability. Among these factors: For load quality, The extension length of the robotic arm. For the maximum permissible moment of inertia, The joint angular velocity, For external forces, The height of the force, For the total mass of the robot, Track ground contact length; stability factor When the value is less than 0, it is assigned the value of 0. After compensation by the robotic arm, the vehicle speed is... When the stability factor When the value is less than 0.3, the angular velocity of the robotic arm joints is also limited to no more than 0.3. ; S5-3. Constructing a graded centroid offset compensation formula: ; A three-level linear compensation mechanism is adopted; the larger the center of gravity offset, the smaller the compensation coefficient, and the lower the vehicle speed. Maximum permissible centroid offset; centroid compensation coefficient When the value is less than 0, it is assigned the value of 0, and the final corrected vehicle speed is... .

[0011] Preferably, the vehicle speed smoothing and load attitude coordinated control in step S6 includes the following steps: S6-0. Construct a discrete prediction model that includes vehicle speed and bump index to predict future changes in system state in the time domain; S6-1. Constructing the model to predict the control objective function: ; Taking into account three optimization objectives: vehicle speed tracking accuracy, vehicle speed change smoothness, and driving impact intensity; among which: To predict the time domain, These are the weighting coefficients. For reference vehicle speed, The rate of change of vehicle speed; S6-2. Solve the quadratic programming problem online, output the optimal vehicle speed control sequence, and extract the first control variable as the real-time vehicle speed output; S6-3. Based on the predicted turbulence index over the next three sampling periods, generate the robotic arm end effector attitude feedforward compensation angle. ; ; The PD control algorithm is used to generate the pitch and roll compensation angles of the robotic arm end effector, thus preemptively offsetting the load vibration caused by road bumps; where: These are the pitch compensation angle and tilt compensation angle at the end effector of the robotic arm. These are PID control parameters; S6-4. Based on the predicted turbulence index of the next 3 sampling periods, generate feedforward commands for stabilization of the photoelectric load gimbal. The gimbal stabilization gain is adjusted in real time according to the turbulence level: L1 level uses conventional gain, L2 level increases by 30%, L3 level increases by 60%, L4 level increases by 100%, and the vibration amplitude of the photoelectric load is reduced in coordination with vehicle speed control. S6-5. Apply upper and lower limit constraints to the end effector compensation angle of the robotic arm and the gimbal stabilization gain to ensure... , This is to prevent overcompensation from damaging the robotic arm or gimbal.

[0012] Preferably, the sensor health monitoring and failure protection in step S7 includes the following steps: S7-1. Real-time calculation of sensor data signal-to-noise ratio, variance, and multi-source data consistency indicators to quantitatively assess sensor health. The formula for calculating health is: ; The health status of sensors is assessed by comprehensively evaluating three dimensions: signal quality, data fluctuation, and multi-source consistency; among which, For reference signal-to-noise ratio and reference variance, The consistency coefficient for multi-source data. These are the weighting coefficients, and ; S7-2. When the health status of a single type of sensor is lower than a preset threshold, its weight in the turbulence index calculation is reduced proportionally. For every 10% decrease in health status, the weight is reduced by 20%. S7-3. If the deviation between multi-source fused data and single-source data exceeds 20% within 5 consecutive sampling periods, the multi-source sensor fusion is determined to be abnormal. S7-4. Automatically switches to a simplified dual-source bump model combining an IMU and a track tension sensor: ; In downgraded mode, the bump index is calculated using data from the two most reliable sensors to ensure basic driving safety; among them The characteristic value of vehicle body vibration energy. This represents the characteristic value of track tension fluctuation. S7-5. Speed ​​limit in downgraded mode: In the formula To preset a lower safe speed; S7-6. Trigger equipment degradation alarm and store abnormal data for fault tracing and diagnosis; S7-7. After the sensor health status recovers to the normal threshold, a gradual recovery mechanism is adopted, increasing the sensor weight by 20% every 10 sampling cycles until the normal weight is restored, avoiding vehicle speed fluctuations caused by sudden switching.

[0013] Preferably, the forward-looking path replanning in step S8 includes the following steps: S8-0. Set the grid resolution of the environmental grid bump heatmap to 1m×1m and the time decay coefficient to 0.95, which means that the weight of historical data decays by 5% every hour; S8-1. During normal operation of the equipment, GPS coordinates and turbulence index λ are collected synchronously to construct an environmental grid turbulence heat map. S8-2. Use a time decay algorithm to update the heatmap and reduce the weight of historical old data; S8-3. Detected a presence within 50 meters ahead. When there are continuous high-bump areas, or when the total cumulative bump value of the current path exceeds the threshold, the path replanning procedure is initiated. S8-4. The improved A* algorithm is used to plan the global path, and the node cost function is as follows: ; Add a bump cost factor to prioritize routes with lower bump levels; where: The cost from the starting point to the current node. The heuristic cost from the current node to the destination. Weighting for turbulence costs; S8-5. Constrain the total time of the path after programming to no more than 1.2 times that of the original path, and output a low-bump alternative path; S8-6. When you are 10 meters away from a bumpy area, reduce your speed to the corresponding safe speed level in advance. S8-7. Real-time detection of sudden bumps during driving; when a sudden bump is detected within 10 meters ahead... In the event of a sudden bump, local path replanning is initiated to generate a temporary obstacle avoidance path.

[0014] Preferably, the cloud-based collaborative turbulence prediction in step S9 includes the following steps: S9-0. Federated learning is performed using the federated averaging algorithm. The specific process is as follows: each robot trains the model locally → uploads the model parameters to the cloud platform → the cloud platform aggregates the model parameters → distributes the aggregated model parameters → each robot updates its local model. S9-1. The robot uploads data such as bump index, GPS coordinates, vehicle pose, and road surface image to the cloud platform after desensitizing them. The S9-2 cloud platform uses a federated learning framework to train regional turbulence prediction models while protecting the privacy of the original data. S9-3. The model input parameters include satellite topographic maps, meteorological data, historical turbulence database, and road surface image features; S9-4. Model output: Predicted turbulence index for each environmental grid. And the confidence level, which is calculated based on historical prediction errors; the smaller the prediction error, the higher the confidence level. S9-5. The cloud platform periodically iterates and updates the model, and distributes it to each terminal robot; S9-6. Before the robot enters an unknown area, a safe speed threshold is preloaded based on the predicted bump index. S9-7. When the confidence level of the prediction model is higher than 90%, the pre-deceleration operation is completed 30 meters in advance; when the confidence level is lower than 50%, the prediction result is only used as a reference, and the local sensor data is mainly relied upon. S9-8. The robot feeds back the actual bump data to the cloud platform to complete the model iteration and optimization; S9-9. The cloud platform performs offline verification of the updated model, and only after the verification is passed is it distributed to the robot to ensure that the model performance does not degrade.

[0015] Preferably, the system parameter self-calibration in step S10 includes the following steps: S10-0. Set parameter calibration trigger conditions: after a single task is completed, the continuous driving mileage exceeds 20km, the ambient temperature change exceeds 20℃, and the system detects a change in mechanical performance. S10-1. Establish a parameter calibration database to store driving condition data, control commands, and vehicle response data; S10-2. When the calibration triggering condition is met, offline calibration of control model parameters is completed based on historical data; S10-3. Calibration parameters include bump attenuation coefficient. Load sensitivity coefficient ρ, safety margin factor PID control parameters for end effector compensation of robotic arm and PID control parameters for gimbal stabilization of photoelectric load; S10-4. Taking into account the three objectives of vehicle speed tracking accuracy, bump index prediction accuracy, and driving stability, a parameter calibration optimization objective is constructed. S10-5. Minimize the optimization objective and optimize the model parameters using the gradient descent method; S10-6. When the equipment is running online, the parameters are fine-tuned in real time by recursive least squares method; S10-7. When the parameter change exceeds the preset threshold, the system self-check is triggered to troubleshoot mechanical faults; S10-8. Apply upper and lower limit constraints to the calibrated parameters to ensure that the parameters are within a reasonable range; S10-9. After calibration, the parameters are uploaded to the cloud platform to achieve parameter sharing and collaborative optimization among multiple robots.

[0016] This invention provides an adaptive speed control method for special robots based on road surface bump index, which has the following beneficial effects: 1. To address the driving characteristics of tracked robots, a dynamic bump index is constructed using multi-source sensor fusion technology. This index can accurately identify complex road surfaces such as sand and mud, which are difficult to distinguish using traditional single sensors. It also improves the robustness of perception in harsh environments such as dust, rain, and strong light. Based on a preset storage and transportation state baseline center of mass combined with the real-time posture of the robotic arm, the center of mass of the entire vehicle is dynamically estimated, and the bump level boundary is adaptively adjusted. This solves the problem that traditional fixed thresholds cannot balance driving safety and operational efficiency.

[0017] 2. Model predictive control is used to generate a smooth vehicle speed sequence. At the same time, based on the predicted bump index, the robot arm end attitude feedforward compensation and photoelectric load gimbal graded stabilization commands are output to actively offset the impact of road bumps from the load end, effectively ensuring the accuracy of reconnaissance and detection data.

[0018] 3. Construct an environmental bump heat map based on historical driving data and plan a low-bump path. Through a federated learning framework, achieve multi-robot collaborative perception and proactive pre-deceleration while protecting data privacy. Establish a multi-dimensional sensor health monitoring and degradation protection mechanism, coupled with system parameter self-calibration function, to improve long-term operational reliability and cluster equipment operation and maintenance efficiency in complex environments. Attached Figure Description

[0019] Figure 1 A flowchart illustrating an adaptive speed control method for special robots based on road bump index. Figure 2 The relationship curve between bump index and maximum safe speed; Figure 3 Bumper heat map and forward-looking route planning diagram; Figure 4 System auxiliary function architecture diagram; Figure 5 Relationship between stabilization gain and turbulence level of optoelectronic payload gimbal; Figure 6 Relationship between bump level classification and maximum safe speed. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to preferred embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This embodiment provides an adaptive speed control method for special robots based on the road surface bump index. It is applicable to tracked special robots used in reconnaissance, urban counter-terrorism, fire rescue, and emergency bomb disposal on unpaved and complex road surfaces. It is particularly suitable for robots equipped with multi-degree-of-freedom robotic arms, precision photoelectric detection payloads, and toxic and hazardous gas detection equipment. This method achieves accurate perception of road surface bump conditions through multi-source sensor fusion, dynamically generates the optimal safe speed by combining the robot's own load, attitude, and task status, and achieves coordinated control of vehicle speed, robotic arm end-effector attitude, and photoelectric payload gimbal. It also features forward-looking path planning, cloud-based collaborative prediction, and fault self-protection capabilities, maximizing operational efficiency while ensuring driving safety and equipment integrity.

[0022] The complete implementation process of this method is as follows: Figure 1 As shown, the adaptive vehicle speed control method of the present invention adopts a four-layer closed-loop architecture of perception-decision-control-assistance. The main process includes six core execution steps in sequence: multi-source sensor data acquisition and preprocessing (S1), multi-feature fusion dynamic bump index calculation (S2), load and attitude adaptive bump level classification (S3), multi-constraint maximum safe vehicle speed generation (S4), multi-objective dynamic vehicle speed correction (S5), vehicle speed smoothing and load and attitude coordinated control (S6). At the same time, it integrates four auxiliary assurance functions: sensor health monitoring and failure protection (S7), forward-looking path replanning (S8), cloud-based collaborative bump prediction (S9), and system parameter self-calibration (S10), to realize full-link adaptive control from road perception to execution output.

[0023] S1. Multi-source sensor data acquisition and preprocessing: In this embodiment, the tracked special robot is equipped with a 16-line mechanical lidar, a nine-axis inertial measurement unit (IMU), two distributed track tension sensors, and a binocular high-definition vision camera. The sampling frequencies of each sensor are set as follows: lidar 10Hz, IMU 100Hz, track tension sensor 50Hz, and vision camera 30Hz.

[0024] First, hardware-level time synchronization is achieved by the main controller sending hardware synchronization pulse signals at regular intervals, unifying the timestamps of data from all sensors and controlling the time synchronization error to within 1ms. Then, the 3σ criterion is used to remove outliers from the data of each sensor. Discrete points exceeding ±3 times the standard deviation of the corresponding data are directly removed to avoid interference from outlier data in subsequent calculations. Finally, by using the pre-calibrated extrinsic parameter matrices of each sensor (obtained through hand-eye calibration), the lidar point cloud, IMU acceleration / angular velocity data, track tension data, and visual images are uniformly converted to the robot's body coordinate system to complete spatiotemporal registration, ensuring that all data are fused under the same spatial and temporal reference.

[0025] S2. Calculation of Dynamic Tumble Index by Multi-Feature Fusion: Four complementary features—road elevation change, vehicle vibration energy, track tension fluctuation, and road texture roughness—are extracted and adaptively weighted fusion is used to obtain the dynamic bump index λ, which comprehensively quantifies the degree of road bumpiness. The calculation formula is as follows: ; This formula constructs a multi-dimensional composite bump evaluation system adapted for tracked robots. It employs a weighted summation method to fuse multi-source features, with weights and constraints ensuring the bump index remains normalized within the 0-1 range for easier subsequent processing. The four features describe road bump characteristics from different physical dimensions: It reflects the early sensing characteristics of road surface geometry and can predict bumps before the tracks contact the road surface; It reflects the actual vibration response characteristics of the vehicle body and directly demonstrates the impact of bumps on the robot; Reflecting the interaction characteristics of the contact force between the track and the ground is a unique sensing dimension of tracked robots, which can more accurately reflect the slippage, jamming and grounding status of the track than the wheel pressure of wheeled robots. It reflects the texture characteristics of road surface materials and can identify special road surfaces that are difficult to distinguish with traditional sensors. The four types of complementary features can completely solve the problem of blind spots in the perception of a single sensor.

[0026] Dynamic weighting coefficients The signal-to-noise ratio (SNR) of each feature is dynamically adjusted based on its real-time SNR. Features with higher SNR are more reliable and thus have higher weights. For example, when the SNR of a lidar point cloud decreases in rainy weather, its weight... The IMU weights will automatically decrease. The corresponding increase enhances the robustness of the system environment.

[0027] The specific calculation steps for each eigenvalue and weight coefficient are as follows: S2-0. Perform statistical filtering to denoise the raw point cloud of the lidar: calculate the average distance from each point to its 50 nearest neighbors, remove outliers and abnormal points whose distance is greater than the mean plus 2 standard deviations, and retain only the effective point cloud; S2-1. The RANSAC random sampling consensus algorithm is used to segment the denoised point cloud for ground. Bilinear interpolation is then performed on the segmented ground point cloud to generate a continuous road surface elevation sequence with a resolution of 0.1m × 0.1m. The elevation sequence is traversed through a 1m sliding window, and the mean of the absolute values ​​of the second derivatives of the elevations within the window is calculated to obtain the characteristic values ​​of road surface elevation changes. The larger the value, the more severe the road surface unevenness; S2-2. Perform first-order differential detrending processing on the IMU triaxial acceleration signal to eliminate the DC component generated by gravitational acceleration and uniform motion; then perform Fast Fourier Transform (FFT) on the processed signal to extract the energy spectral density integral in the 0.5Hz-20Hz frequency band to obtain the vehicle vibration energy characteristic value. This frequency band is a sensitive band for precision electronic equipment and optoelectronic loads, and can accurately reflect the impact of bumps on detection equipment; S2-3. Perform a high-pass filter with a cutoff frequency of 0.5Hz on the raw signals from the two track tension sensors to eliminate DC interference generated by the robot's static load; calculate the root mean square (RMS) value of the tension signal fluctuation of the left and right tracks respectively, and take the average of the two tracks to obtain the track tension fluctuation characteristic value. This value reflects the degree of change in the contact force between the track and the ground; the greater the tension fluctuation, the more uneven the road surface. S2-4. Perform grayscale conversion, 5×5 Gaussian filtering, and Canny edge detection on the road surface images acquired by the vision camera in sequence, and calculate the gray-level co-occurrence matrix (number of gray levels). The texture entropy value (with a step size of 1 and a direction of 0°) is used to obtain the surface texture roughness feature value. The calculation formula is: ; This formula quantifies the complexity of road surface texture using the entropy value of the gray-level co-occurrence matrix. The gray-level co-occurrence matrix describes the gray-level spatial correlation between two pixels that are a certain distance apart in an image. Indicates grayscale The pixels and the pixels separated by one step, with a gray level of The probability of pixels appearing simultaneously. Entropy is a measure of the disorder of a system; a higher entropy value indicates a more uneven grayscale distribution and coarser texture. While sandy or muddy surfaces have small elevation changes, their strong surface graininess and complex texture result in significantly higher texture entropy values ​​than hardened surfaces, effectively distinguishing these seemingly flat but bumpy surfaces. Grayscale levels The image retains its complete grayscale information, and the settings of step size 1 and direction 0° balance computational efficiency and texture recognition accuracy.

[0028] S2-5. Calculate the signal-to-noise ratio (SNR) (the ratio of the signal mean to the noise standard deviation) for each feature value in real time, and dynamically allocate weight coefficients based on the SNR: the higher the SNR, the more reliable the feature data, and the greater the corresponding weight, ensuring that the fusion result always leans towards the most reliable sensor data; S2-6. Apply second-order exponential smoothing to the weight coefficient sequence and set the smoothing factor. , It balances system response speed and smoothing effect, and suppresses vehicle speed jitter caused by sudden weight changes.

[0029] S3. Load and Attitude Adaptive Bump Level Classification: The bump level boundary values ​​are dynamically adjusted based on the robot's current load mass and center of gravity offset, classifying the road surface into four bump levels: L1 (flat road surface, Level 2 (mild bumps) Level L3 (moderate bumps) Level L4 (severe turbulence), ).

[0030] The adaptive calculation formula for the boundary value of the bump level is as follows: ; This formula dynamically adjusts the bump level boundary value according to the robot's own state. The core idea is that the greater the load and the more severe the center of gravity shift, the higher the driving risk corresponding to the same bump index. Therefore, it is necessary to reduce the boundary value and adopt a more conservative speed control strategy. A product form is used instead of an additive form because the impact of load and center of gravity shift on driving risk is amplified: the risk of being fully loaded with center of gravity shift is far greater than the sum of the risks of being fully loaded or having center of gravity shift alone. The three dividing points corresponding to the fourth level of turbulence; Determine the boundary values ​​for the unloaded flat surface; these serve as the benchmark for all dynamic adjustments. For load sensitivity coefficient, The two coefficients are the centroid offset sensitivity coefficients, and both were calibrated through extensive real-vehicle tests. In this embodiment... This indicates that when the load mass equals the robot's base mass, the bump level threshold is reduced by 30%. This means that when the centroid offset is equal to the reference centroid position, the boundary value is reduced by 40%.

[0031] The specific implementation steps are as follows: S3-0. Based on the preset reference center of mass position for storage and transportation, and combined with the joint angles, joint mass distribution, and end-effector load mass of the robotic arm, the actual center of mass position of the robot is calculated in real time using a rigid body dynamics model. This is then compared with the pre-calibrated reference center of mass position to obtain the center of mass offset. ; S3-1. Before the equipment's first operation, complete the no-load boundary values ​​at the standard bump test track. The calibration ensures the accuracy of the basic parameters; S3-2. Before starting a single task, the robot remains stationary for 10 seconds under no-load conditions to collect IMU vibration data and complete the no-load boundary value. Rapid self-calibration with calibration error controlled within 5%, eliminating the influence of ambient temperature and mechanical wear on parameters; S3-3. During equipment operation, load mass and robotic arm posture data are collected in real time, and the center of gravity position and bump level boundary values ​​are dynamically updated. S3-4. When the center of gravity offset exceeds 50% of the maximum permissible center of gravity offset, all bump level boundary values ​​will be uniformly reduced by 15%, and conservative speed control will be implemented to avoid the risk of rollover. S3-5. Apply upper and lower limit constraints to the boundary values ​​of the bump level to ensure... At the same time, the boundary value update frequency is limited to no more than 1Hz to avoid frequent fluctuations in boundary values ​​that could lead to unstable vehicle speed.

[0032] S4. Generation of Maximum Safe Vehicle Speed ​​with Multiple Constraints: Based on the current bump level and real-time vehicle attitude, a maximum safe vehicle speed threshold is generated that considers acceleration constraints and impact energy limits. It is generated using a piecewise nonlinear exponential decay model to avoid sudden speed changes caused by piecewise linear functions. The calculation formula is as follows: ; like Figure 2 As shown, this invention employs an exponential decay model to achieve a smooth transition of vehicle speed with the bump index. Compared to the abrupt speed changes at the level boundary points of the traditional piecewise linear model, the vehicle speed curve of this invention is continuously differentiable throughout, effectively avoiding vehicle body impact, load sway, and sudden stress changes in the robotic arm caused by sudden speed changes. The curve is divided into three functional sections: L1 level flat road surface. Maintain the nominal maximum speed V_base to maximize operational efficiency; L2-L3 level light to moderate bumpy road surface. Vehicle speed decreases exponentially with the bump index, precisely matching the mechanical properties of road impact energy being proportional to the square of vehicle speed; L4 level severely bumpy road surface. Locked to This ensures safe passage at low speeds, preventing the engine from stalling due to excessively low speeds.

[0033] This formula employs a three-stage nonlinear speed generation strategy, balancing safety and operational efficiency under varying levels of bumpiness. It abandons the traditional piecewise linear abrupt change structure, ensuring a smooth transition throughout and meeting the stability requirements of tracked special robots. The first stage, on a flat surface, maintains the nominal maximum speed. To maximize operational efficiency; the second moderately bumpy section employs an exponential decay model, aligning with the mechanical property that road impact energy is proportional to the square of vehicle speed—the higher the bump intensity, the faster the vehicle speed decreases, precisely suppressing the impact on the vehicle body caused by bumps; the third severely bumpy section is locked at... This ensures safe passage at low speeds, preventing engine stalling due to excessively low speeds. The bump attenuation coefficient controls the rate at which vehicle speed decreases with increasing bump index. The safety margin factor is set to 0.8 to reserve safety redundancy. This is the minimum controllable speed for the robot.

[0034] The specific implementation steps are as follows: S4-1. Real-time acquisition of vehicle pitch angle yaw angle For the turbulence attenuation coefficient The attitude correction is performed using the following formula: ; When the vehicle body pitches or tilts, the robot's center of gravity shifts, chassis stability deteriorates, and the risk of overturning under the same bumpy conditions increases significantly. Therefore, the bump attenuation coefficient is amplified by amplifying the attitude data to forcibly reduce the safe speed. A normalized angle calculation method is used to eliminate differences in attitude angle dimensions, ensuring that the correction ratio of pitch and tilt to the attenuation coefficient is balanced and the adjustment is smooth without abrupt changes. Base attenuation value; To adjust the gain; , These represent the robot's maximum allowable pitch angle and maximum roll angle, respectively, which are the critical attitude angles for safe driving of the tracked robot.

[0035] S4-2. Apply vehicle speed acceleration constraints, limiting the rate of change of vehicle speed to no more than [a certain value]. In this embodiment Strictly avoid problems such as vehicle body swaying, load tilting, and sudden stress changes in the robotic arm structure caused by rapid acceleration and deceleration; S4-3. Calculate the road impact energy corresponding to the current vehicle speed in real time. If the impact energy exceeds the limit, the maximum safe vehicle speed will be lowered again. The impact energy calculation formula is: ; Road impact energy is a core physical indicator for measuring the damage caused by bumps to a robot's structure and onboard detection equipment. The higher the energy value, the greater the probability of structural deformation and damage to precision instruments. The formula shows that impact energy is related to the robot's total mass. Vehicle speed square Bump index There is a positive correlation, with vehicle speed having the most significant impact on impact energy (square relationship), therefore vehicle speed is the priority variable for control. In this embodiment, the safe threshold for impact energy is set at 400J. Once the energy exceeds the limit in real time, the maximum safe vehicle speed is automatically reduced by 10% to enhance protection capabilities.

[0036] S4-4. Apply upper and lower limit constraints to the maximum safe speed to ensure To prevent the vehicle speed from exceeding the controllable range of the robot's walking mechanism, thus preventing low-speed instability and high-speed loss of control malfunctions.

[0037] S5. Multi-objective dynamic vehicle speed correction: The maximum safe speed threshold is dynamically corrected through three dimensions: task urgency compensation, robotic arm stability compensation, and center of gravity offset compensation, to obtain the final execution speed, thus balancing safety, efficiency, and operational requirements.

[0038] S5-1. Mission Urgency Compensation: Construct a formula for assessing task urgency, quantifying the urgency of a task by combining remaining time and progress already completed: ; This formula quantifies the urgency of a task from both time and schedule dimensions, avoiding the unreasonable acceleration issues caused by a single evaluation dimension. (Time item) The urgency of a task's remaining time is indicated by the number of seconds remaining; the less time left, the higher the value. (Schedule item) This represents the degree of lag in the current completion rate; the slower the progress, the larger the value. Weighting parameters. Prioritizing strict deadlines aligns with the priority logic of emergency rescue and counter-terrorism reconnaissance missions. An upper limit of 0.4 is imposed on the urgency level, rigidly limiting the maximum acceleration to prevent sacrificing driving safety for excessive pursuit of mission timeliness.

[0039] After mission urgency compensation, the vehicle speed is: ; A multiplicative correction method is used to ensure that the lower the basic safe speed, the smaller the absolute value of acceleration. Under severe bumps, safety is prioritized, while efficiency is less weighted. (Emergency gain coefficient) Combined with the upper limit constraint of urgency, the maximum speed-up rate of the robot is strictly controlled within 1.2 times, so as to achieve small, controllable and safe efficiency compensation.

[0040] S5-2. Robotic arm stability compensation: A formula for calculating the stability factor of a robotic arm is constructed, and the operational stability of the robotic arm is evaluated by comprehensively considering the rotational inertia and external overturning moment. Specific speed constraints are implemented for robotic arms. ; Stable factor The normalization value ranges from 0 to 1. The closer the value is to 1, the smaller the inertial disturbance and overturning moment of the robotic arm, and the better the operational stability. The formula contains two disturbance terms: the first term... Characterizing the inertial hazards of robotic arm motion, the longer the arm span, the greater the end-effector load, and the faster the joint rotation, the stronger the inertial disturbance and the lower the stability; the second item This characterizes the risk of overturning caused by external forces; the greater the external force and the higher the altitude at which it acts, the greater the risk of overturning. The track ground contact length, compared to the wheelbase of wheeled robots, more accurately reflects the anti-tipping capability of tracked robots. The fallback constraint is used when calculating the stability factor. If the value is forcibly assigned to 0, it is judged as an extremely unstable state, and normal driving is prohibited.

[0041] The vehicle speed after compensation by the robotic arm is: ; Vehicle speed and the robotic arm's stability factor are linearly positively correlated; stability degradation synchronously reduces vehicle speed. A safety judgment logic is added when... If the operation is deemed to be in a high-risk state, the maximum angular velocity of the robotic arm is simultaneously locked to half of the rated value, and a dual protection strategy of vehicle speed and joint angular velocity is adopted.

[0042] S5-3. Centroid offset compensation: A three-level linear centroid offset compensation formula is constructed to control vehicle speed according to the degree of centroid offset, thereby suppressing the rollover risk caused by off-center loading and biased loads. ; A three-level piecewise linear partitioning method is adopted, dividing the centroid offset into a safe zone, a warning zone, a danger zone, and a prohibited zone. This hierarchical control logic is clear and easy to implement in engineering, avoiding the complex debugging issues of nonlinear algorithms. Within the safe zone, the centroid offset is minimal, having no significant impact on driving stability; the compensation coefficient remains at 1, without reducing operational efficiency. The warning zone features linear speed reduction for a smooth transition. The danger zone rapidly reduces vehicle speed to enhance protection. The prohibited zone directly locks the vehicle speed to prevent rollover accidents. A fallback constraint compensation coefficient is provided. When the value is forcibly set to 0, the power output is cut off.

[0043] The final corrected speed is: ; The final execution speed is obtained by superimposing three correction coefficients, which takes into account task efficiency, robotic arm operation safety, and vehicle body off-center load stability, forming a comprehensive speed optimization result with multiple constraints coupled together. Upper and lower limit constraints are applied to ensure that the final speed is always within the robot's controllable range.

[0044] S6. Coordinated control of vehicle speed smoothing and load attitude: Model predictive control (MPC) is used to generate a smooth vehicle speed sequence without abrupt changes. At the same time, feedforward commands for attitude compensation of the robotic arm end effector and feedforward commands for stabilization of the photoelectric payload gimbal are output to achieve coordinated optimization of vehicle speed and payload attitude. This reduces the vibration of the detection equipment from the root and solves the problems of vehicle speed jitter and image blurring on bumpy roads.

[0045] The specific implementation steps are as follows: S6-0. Construct a discrete prediction model that includes vehicle speed and bump index, and set the prediction time domain. Control Time Domain It is used to predict system state changes within the next 15 sampling periods, balancing prediction accuracy and computing power; S6-1. Construct a model predictive control objective function to comprehensively optimize three objectives: vehicle speed tracking accuracy, vehicle speed change smoothness, and driving impact intensity. ; The objective function uses a weighted sum of squares form, transforming the multi-objective optimization problem into a single-cost solution problem, which is suitable for fast computation by a quadratic programming solver. The first term is the tracking error term, ensuring the actual vehicle speed accurately matches the corrected reference speed; the second term is the smoothness constraint term, suppressing sudden speed changes and rapid acceleration / deceleration; the third term is the impact suppression term, limiting high-speed driving behavior under high-bump conditions. Weighting coefficients... The priorities decrease sequentially, which aligns with the control requirements of tracked special robots for precise speed, smooth driving, and low impact.

[0046] S6-2. Solve the quadratic programming (QP) problem online, output the optimal vehicle speed control sequence, extract only the first control variable in the sequence as the real-time vehicle speed output at the current moment, realize rolling optimization, and continuously correct the control deviation; S6-3. Based on the predicted bump index over the next three sampling periods, the PD control algorithm is used to generate the robotic arm end-effector attitude feedforward compensation angle to preemptively offset the load vibration caused by road bumps: ; ; Employing a feedforward PD control algorithm, compensation commands are generated in advance based on predicted turbulence indices, significantly reducing lag compared to traditional feedback control. (Proportional term) Using the predicted turbulence index as input, turbulence intensity-matched compensation is achieved; the greater the turbulence, the larger the compensation angle. (Differential term) Using the gradient of bump changes as input, the system suppresses impact vibrations caused by sudden bumps. Control parameters are calibrated. , , , The pitch compensation force is greater than the tilt compensation force, matching the working condition characteristics of longitudinal bumps on the ruined road surface. Upper and lower limits are imposed on the compensation angle to prevent over-compensation from damaging the robotic arm structure.

[0047] S6-4. Based on the predicted turbulence index of the next 3 sampling periods, generate feedforward commands for stabilization of the photoelectric load gimbal. The gimbal stabilization gain is adjusted in real time according to the turbulence level: L1 level uses conventional gain, L2 level increases by 30%, L3 level increases by 60%, L4 level increases by 100%, and the vibration amplitude of the photoelectric load is reduced in coordination with vehicle speed control. S6-5. Apply upper and lower limit constraints to the end effector compensation angle of the robotic arm and the gimbal stabilization gain to ensure... , This is to prevent overcompensation from damaging the robotic arm or gimbal.

[0048] S7. Sensor health monitoring and failure protection: Real-time assessment of the health status of each sensor, establishment of a graded failure protection mechanism, automatic switching to degraded operation mode when sensors are abnormal, interfered with, or fail, ensuring continuous and reliable operation of the system and improving fault tolerance in complex field and urban ruin environments.

[0049] The specific implementation steps are as follows: S7-1. Quantitatively assess the health of each sensor from three dimensions: signal quality, data fluctuation, and multi-source consistency. The calculation formula is: ; This system integrates multiple dimensions to assess sensor health, avoiding the pitfalls of misjudging faults based on a single indicator, and is adaptable to complex operating conditions such as dust from rubble, rain, and strong light interference. (Signal-to-noise ratio item) Evaluate signal purity; variance term To assess data stability, a larger variance indicates lower health; consistency items... The health score is determined by the degree of matching between multi-sensor fusion data and data from other sensors; the greater the deviation from the data, the lower the health score. (Weight) Signal quality is given priority. The health score is normalized to 0-1; the higher the score, the better the sensor's performance.

[0050] S7-2. When the health of a single type of sensor is below 0.7, its weight in the turbulence index calculation is reduced proportionally: for every 10% decrease in health, the corresponding weight is reduced by 20%, thus weakening the interference of abnormal data on the fusion results; S7-3. If the deviation between the multi-source fused data and a single-source data exceeds 20% within 5 consecutive sampling periods, it is determined to be an anomaly in multi-source sensor fusion and the backup protection logic is triggered. S7-4. Automatically switches to a simplified dual-source bump model combining an IMU and a track tension sensor. These two types of sensors have the strongest resistance to environmental interference, ensuring basic sensing capabilities under harsh working conditions. ; In degraded mode, LiDAR and vision cameras are removed, retaining only the most reliable IMU and track tension sensors, reducing computing power consumption while improving stability. IMU vibration characteristics directly characterize the vehicle's bump response and have higher priority than track tension, thus being assigned a higher weight of 0.6.

[0051] S7-5. Maximum driving speed in limited downgrade mode: ; In the degraded perception mode, a speed limit is enforced, ensuring that the maximum speed does not exceed 4 km / h, regardless of the theoretically safe speed, so that the robot can be controlled at low speeds and withdraw safely.

[0052] S7-6. Trigger the device's audible and visual degradation alarm, and simultaneously store sensor data and control commands for the abnormal time period for subsequent fault tracing, diagnosis, and maintenance analysis; S7-7. Once the sensor health level recovers to above 0.9, a gradual recovery mechanism is adopted: the weight of the sensor is increased by 20% every 10 sampling cycles until the normal weight is restored, so as to avoid the vehicle speed oscillation caused by sudden weight changes.

[0053] S8. Forward-looking path replanning: Based on historical driving data, an environmental bump heat map is constructed. Before entering a high-bump area, the system can predict, slow down, and detour, achieving integrated proactive control of perception, prediction, planning, and speed adjustment, thereby reducing the impact of bumps at the path level.

[0054] The specific implementation steps are as follows: S8-0. Initialize the environmental grid bump heat map, set the grid resolution to 1m×1m, the time decay coefficient to 0.95 (i.e., the weight of historical data decays by 5% per hour), and remove outdated road surface data to ensure the timeliness of the heat map; S8-1. During normal operation of the equipment, GPS geographic location coordinates and real-time turbulence index are collected synchronously. The data is mapped to the corresponding grid, and the heatmap is updated iteratively. S8-2. The heat map is updated in real time using a time decay algorithm to reduce the weight of old historical data and adapt to the dynamic changes in road surface caused by rainwater erosion and building collapse. S8-3. When a presence is detected within 50 meters ahead. When there is a continuous high-bump area (more than 10m in length), or when the total cumulative bump value of the current planned path exceeds the preset threshold, the path replanning program will be automatically started. S8-4. Adopting improved A The algorithm plans a global path, in traditional A Add a bump cost term to the algorithm's cost function to achieve smoothness-first path selection: ; The traditional A* algorithm only considers path length and heuristic cost, failing to distinguish the degree of road bumpiness. This formula adds a bumpiness accumulation term. Quantify the overall bumpiness of the path and prioritize paths with lower bumpiness. Bumpiness cost weighting. Prioritize detouring through bumpy sections of road, while avoiding excessive detours that could reduce operational efficiency, thus meeting the timeliness requirements of emergency rescue missions. For example... Figure 3As shown, the environmental grid bump heatmap constructed based on historical driving data uses different colors to distinguish four levels of bump areas, with darker colors indicating higher road bump levels. The improved A* algorithm, when planning the global path, adds a cumulative bump cost to the traditional path length cost, automatically avoiding consecutive high-bump areas and generating optimized low-bump alternative paths. Simultaneously, the system sets an early deceleration point 10 meters away from high-bump areas, smoothly reducing the vehicle speed to the corresponding safety level in advance, avoiding vehicle impact caused by sudden deceleration near bumpy areas.

[0055] S8-5. The total time of the planned route shall not exceed 1.2 times that of the original route, strictly control detour costs, and take into account both driving stability and operational timeliness. S8-6. When you are 10 meters away from a high bumpy area, reduce your speed to the safe speed corresponding to the bump level in advance to avoid the impact on the vehicle body caused by sudden deceleration near the bumpy area. S8-7. Real-time detection of sudden bumps during driving; when a sudden bump is detected within 10 meters ahead... In the event of a sudden severe turbulence, local path replanning should be initiated immediately to generate a temporary obstacle avoidance path.

[0056] S9. Cloud-based collaborative turbulence prediction: By training a regional bump prediction model using a federated learning framework, data sharing among multiple robots can be achieved without disclosing the original data of individual robots and protecting operational privacy. This enables the prediction of road conditions in unknown areas and the proactive adjustment of vehicle speed.

[0057] The specific implementation steps are as follows: S9-0. Federated learning is performed using the FedAvg algorithm. The process is as follows: each robot trains its model locally → uploads the model parameters to the cloud platform → the cloud platform aggregates the model parameters of all robots → distributes the aggregated global model parameters → each robot updates its local model. No original sensor data is uploaded throughout the process. S9-1. The robot performs desensitization processing on data such as bump index, GPS coordinates, vehicle pose, and road surface image, removing confidential information and geographically sensitive information, and only uploading the feature parameters required for model training. S9-2. The cloud platform builds a federated learning framework, which aggregates the local models of multiple robots and iteratively trains a general regional turbulence prediction model without obtaining the original data. S9-3. The model input parameters include: satellite topographic maps, meteorological data (rainfall, snowfall, etc.), historical turbulence database, road surface image features, and multi-source environmental data to improve prediction accuracy; S9-4. Model output: Predicted turbulence index for each environmental grid. and confidence level The confidence level is calculated based on historical prediction error statistics. The complete calculation formula is as follows: ; The closer the confidence score is to 1, the smaller the model's historical prediction error and the higher the reliability of the prediction results. For the total number of historical prediction samples of the model, Let i be the historical predicted turbulence index. This represents the actual turbulence index collected at the corresponding time. Normalization interval constraints ensure that the confidence level output remains stable within the valid range of 0 to 1.

[0058] The S9-5 cloud platform iterates and updates the global model every 12 hours and distributes it to all terminal robots to complete batch upgrades and adapt to the characteristics of rapid changes in road surfaces at urban disaster sites. S9-6. Before the robot enters an unknown area, it obtains the predicted bump index of the area from the cloud platform and preloads the corresponding level of safe speed threshold. S9-7. When the confidence level of the prediction model is higher than 90%, the pre-deceleration operation is completed 30 meters in advance; when the confidence level is lower than 50%, the prediction result is only used as a reference, and the control is mainly based on the real-time data of local sensors to prevent prediction distortion. S9-8. During the robot's movement, the actual collected bump data is fed back to the cloud platform as training samples for model iteration, continuously optimizing the prediction accuracy; S9-9. The cloud platform performs offline verification of the updated model. Only when the verification accuracy is not less than 95% will the model be sent to the robot. This strict control of model quality prevents the release of degraded models that could lead to prediction errors.

[0059] S10. System parameter self-calibration: By utilizing historical driving data and online feedback, the control model parameters are automatically calibrated, and system deviations caused by mechanical wear, changes in ambient temperature, and sensor drift are adaptively compensated, thereby improving the long-term operational stability and control accuracy of the robot.

[0060] The specific implementation steps are as follows: S10-0. Set parameter calibration trigger conditions. Calibration will be started if any one of the following conditions is met: ① After a single task is completed; ② Continuous driving distance exceeds 20km; ③ Ambient temperature change exceeds 20℃; ④ The system detects abnormal changes in mechanical performance. S10-1. Build a local parameter calibration database to store driving condition data, control commands and vehicle response data for the most recent 20km, and limit the data capacity to avoid storage redundancy; S10-2. When the calibration triggering conditions are met, offline calibration of the control model parameters is completed based on historical data in the database; S10-3. Calibration parameters include: bump attenuation coefficient. Load sensitivity coefficient Safety margin factor The system includes PID control parameters for end effector compensation of the robotic arm and PID control parameters for gimbal stabilization of the photoelectric load, covering core control variables. S10-4. Taking into account the three objectives of vehicle speed tracking accuracy, bump index prediction accuracy, and driving stability, a parameter calibration optimization objective function is constructed: ; The objective function integrates three calibration objectives, obtaining the optimal parameter combination by minimizing the cost function. The first term is the vehicle speed tracking error, characterizing the degree of alignment between the actual and theoretical vehicle speeds; the second term is the bump prediction error, characterizing the road perception accuracy; and the third term is the vehicle speed fluctuation error, characterizing the ride smoothness performance. Weighting configuration... Prioritize the accuracy of vehicle speed control while also considering perception accuracy and driving smoothness.

[0061] S10-5. Use gradient descent to minimize the objective function, iteratively optimize the model parameters, and set the learning rate to 0.01 to ensure smooth convergence and prevent divergence. S10-6. When the equipment is running online, the parameters are fine-tuned in real time by recursive least squares method. The combination of offline coarse calibration and online fine-tuning adapts to long-term operating condition drift. S10-7. When the parameter change in a single calibration exceeds the preset threshold (20% in this embodiment), the system self-test program is triggered to check for potential faults such as track wear, sensor drift, and structural loosening. S10-8. Apply upper and lower limit constraints to all calibrated parameters to ensure that the parameters are always within a reasonable calibration range and to avoid system malfunction due to calibration abnormalities. S10-9. Upload the calibrated optimal parameters to the cloud platform to achieve parameter sharing and collaborative optimization among multiple robots, thereby reducing the maintenance and calibration costs of cluster equipment in batches.

[0062] Example: A tracked special-purpose robot equipped with precision detection instruments performs disaster relief missions. Taking a tracked rescue robot performing environmental reconnaissance in a disaster area as an example, the complete implementation process of this method is as follows: The robot's basic mass Nominal maximum speed Minimum controllable speed Maximum permissible pitch angle yaw angle Maximum allowable centroid offset Preset storage and transportation status reference centroid position ; The no-load boundary values ​​were calibrated in advance through a standard bump test track. , , Load sensitivity coefficient Centroid shift sensitivity coefficient Basic bump attenuation coefficient Attitude adjustment gain Safety margin factor Urgency gain Road impact energy safety threshold .

[0063] This mission involves entering the disaster area from point A, equipped with toxic gas detectors, infrared thermal imagers, and high-definition visible light cameras, to reconnoiter the concentration of toxic gases and the location of trapped personnel within a designated area. The mission deadline is [date missing]. Total target driving distance By the 12th minute of the mission, the distance already traveled was... ,time left At this moment, the robot's real-time status is: the load mass at the end of the robotic arm. (Including detection equipment), extension length Joint angular velocity No external force The centroid offset was calculated using a rigid body dynamics model combined with the robot arm's posture. IMU collects vehicle pitch angle yaw angle .

[0064] Multi-source sensor data preprocessing: LiDAR (10Hz), IMU (100Hz), track tension sensor (50Hz), and vision camera (30Hz) are synchronized at the hardware level by the main controller sending hardware synchronization pulse signals at regular intervals, with the synchronization error controlled within 1ms; the 3σ criterion is used to remove abnormal values ​​from each sensor, and all data are uniformly converted to the vehicle coordinate system through a pre-calibrated external parameter matrix.

[0065] Dynamic turbulence index calculation: After statistical filtering for noise reduction, RANSAC ground segmentation, and bilinear interpolation, the characteristic values ​​of road surface elevation variation were calculated using a 1m sliding window from the lidar point cloud. ; After the IMU triaxial acceleration signal is processed by first-order differential detrending term and FFT transformation, the energy spectral density integral in the 0.5Hz-20Hz frequency band is extracted to obtain the vehicle body vibration energy characteristic value. ; After the track tension signal is filtered by a 0.5Hz high-pass filter, the root mean square value of the tension fluctuation of the left and right tracks is calculated to obtain the characteristic value of the track tension fluctuation. ; After the visual road surface image is converted to grayscale, filtered by a 5×5 Gaussian filter, and subjected to Canny edge detection, the texture entropy value of the gray-level co-occurrence matrix is ​​calculated to obtain the road surface texture roughness feature value. ; Calculate the signal-to-noise ratio of each feature: , , , After normalization, the dynamic weights are obtained. , , , (satisfy ); After second-order exponential smoothing, the dynamic turbulence index is calculated. .

[0066] Load and attitude adaptive turbulence level classification: Substitute into the formula to calculate the dynamic boundary value: , , ; current Between and Between these points, it was determined to be Level L3 moderate turbulence.

[0067] Generation of maximum safe vehicle speed with multiple constraints: Attitude correction turbulence attenuation coefficient: ; Substitute into the piecewise nonlinear function for calculation: In The interval is used to obtain the initial value. ;like Figure 6 As shown, in this embodiment, the boundary values ​​of the bump level after adaptive adjustment based on load and attitude are λ1=0.23, λ2=0.46, and λ3=0.69, respectively, dividing the road surface into four safety level intervals. The current dynamic bump index λ=0.47 is slightly higher than the λ2 boundary value, falling within the L3 level moderate bump interval, corresponding to an initial maximum safe speed of 9.2 km / h. After a second 10% reduction due to excessive road impact energy, the final safe speed is 8.3 km / h. The figure clearly shows the correspondence between the adaptively adjusted bump level division boundaries, the initial calculated speed, and the speed after impact energy correction, intuitively demonstrating the safety redundancy design of the multi-constraint speed generation mechanism.

[0068] Calculate the impact energy of the road surface: Total mass of the robot Current vehicle speed , After exceeding the limit, the adjustment was lowered by 10% a second time, and then corrected. ; Apply acceleration constraints And upper and lower limit constraints, ultimately .

[0069] Multi-objective dynamic speed correction: Task urgency calculation: , , Take the upper limit Speed ​​after task compensation ; Robotic arm stability factor calculation: , The vehicle speed after compensation by the robotic arm ; Centroid offset compensation: In interval, Finally, the vehicle speed was corrected. .

[0070] Vehicle speed smoothing and load attitude coordinated control: Constructing the prediction time domain Control Time Domain The discrete prediction model is used to solve a quadratic programming problem by substituting the objective function into the output, and the optimal vehicle speed sequence is output to extract the first control variable. As real-time vehicle speed; Based on a federated learning time-series prediction model trained in the cloud, and combining historical turbulence time-series data and environmental characteristics, the turbulence index for future sampling periods is predicted. The complete calculation formula is as follows: Under this operating condition, the model outputs a future single-step predicted turbulence index. ; Substitute the values ​​into the attitude compensation formula to calculate the compensation angle at the end of the robotic arm: , All are within the constraints; L3 level turbulence corresponds to a 60% increase in the stabilization gain of the optoelectronic payload gimbal.

[0071] System auxiliary functions are executed as follows: Real-time monitoring shows that the health status of all sensors is greater than 0.9, indicating normal system operation; the local bump heatmap shows no continuous high bump area ahead (50 meters), so the system continues along the original path; the cloud-based federated learning model predicts the bump index 150 meters ahead. (Confidence level 93%), reduce speed to 6.5 km / h 30 meters in advance.

[0072] Results: The robot traveled smoothly at 7.8 km / h, with a peak body vibration acceleration of less than 0.4g. The infrared thermal imager and visible light camera produced clear images without ghosting. The toxic gas detector data was stable and without fluctuations, and no equipment damage was observed. The reconnaissance mission was completed 3 minutes ahead of schedule, successfully locating the two trapped individuals and transmitting real-time images back. This verified the effectiveness of this method in balancing driving safety and reconnaissance efficiency in complex road conditions such as urban ruins and fire scenes.

[0073] This invention constructs a high-precision bump index through spatiotemporal fusion of multi-source sensors, and achieves graded vehicle speed control by combining load, attitude, and task status. It is further enhanced by MPC speed smoothing control and collaborative vibration reduction via the robotic arm end effector and photoelectric payload gimbal. Simultaneously, it integrates fault degradation protection, intelligent path planning, cloud-based federated prediction, and parameter self-calibration functions. Compared to traditional control methods with fixed speed and single-sensor detection, this invention exhibits strong environmental adaptability and excellent anti-interference capabilities. It can effectively adapt to complex unpaved road surfaces in urban counter-terrorism, fire rescue, and emergency bomb disposal, maximizing operational efficiency while ensuring the safe operation of tracked special robots and protecting precision detection payloads. It possesses both engineering application and promotional value.

[0074] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A special robot adaptive speed control method based on road surface bump index, characterized in that, Includes the following steps: S1. Multi-source sensor data acquisition and preprocessing: Real-time acquisition of road surface and vehicle body status data through lidar, inertial measurement unit, track tension sensor and vision camera, and time synchronization, outlier removal and spatiotemporal registration; S2. Multi-feature fusion dynamic bump index calculation: Extract road elevation change features, vehicle vibration energy features, track tension fluctuation features, and road texture roughness features, and obtain the dynamic bump index through adaptive weighted fusion. ; S3. Load and attitude adaptive bump level classification: Based on the robot's preset storage and transportation state center of mass position, combined with the real-time attitude dynamic estimation of the actual center of mass position of the robotic arm, the bump level boundary value is adaptively adjusted to divide the road surface into four bump levels. S4. Generation of maximum safe vehicle speed with multiple constraints: Based on the bump level and real-time vehicle attitude, generate the maximum safe vehicle speed threshold that takes into account acceleration constraints and impact energy limits. S5. Multi-target dynamic speed correction: The maximum safe speed threshold is dynamically corrected by taking into account the urgency of the task, the stability of the robotic arm and the center of gravity offset. S6. Vehicle speed smoothing and load attitude coordinated control: Model predictive control is used to generate a smooth vehicle speed sequence, and at the same time, feedforward commands for end-effector attitude compensation and feedforward commands for photoelectric load gimbal stabilization are output to achieve coordinated optimization of vehicle speed and load attitude. S7. Sensor health monitoring and failure protection: Real-time assessment of the health of each sensor, and automatic switching to degraded operation mode when multi-source fusion is abnormal; S8. Proactive route replanning: Construct an environmental turbulence heatmap based on historical turbulence data, and replan a low-turbulence route before entering a high-turbulence area; S9. Cloud-based collaborative turbulence prediction: By training a regional turbulence prediction model through federated learning, it enables proactive speed adjustment in untouched areas. S10. System parameter self-calibration: Utilizes historical driving data and online feedback to automatically calibrate control model parameters, improving long-term operational accuracy.

2. The adaptive speed control method for special robots based on road surface bump index according to claim 1, characterized in that, Dynamic turbulence index in step S2 The calculation formula is as follows: ； This is a composite bump index that integrates multiple features, comprehensively quantifying the impact of road surface unevenness, vehicle vibration intensity, track tension fluctuations, and road texture characteristics on driving; among which: These are characteristic values ​​of road surface elevation changes. The characteristic value of vehicle body vibration energy. This represents the characteristic value of track tension fluctuation. This represents the surface texture roughness characteristic value; The dynamic weighting coefficients satisfy the following conditions: ; Step S2 further includes the following specific steps: S2-0. Perform statistical filtering on the original point cloud of the lidar to remove distance outliers and outliers, and retain the effective point cloud area. S2-1. Perform ground segmentation and elevation interpolation on the denoised lidar point cloud to generate a continuous road surface elevation sequence. Calculate the mean of the absolute values ​​of the second derivative of the elevation using a sliding window to obtain the characteristic values ​​of road surface elevation changes. ; S2-2. The IMU triaxial acceleration signal is detrended to eliminate the influence of gravitational acceleration and uniform motion. Then, a fast Fourier transform is performed to extract the energy spectral density integral of the load device's sensitive frequency band from 0.5Hz to 20Hz, thus obtaining the vehicle body vibration energy characteristic value. ; S2-3. Perform high-pass filtering on the track tension sensor signal to eliminate static load interference, and calculate the mean of the root mean square values ​​of the left and right track tension fluctuations to obtain the characteristic values ​​of track tension fluctuations. ; S2-4. Perform grayscale conversion, Gaussian filtering, and edge detection on the road surface images acquired by the visual camera, and calculate the texture entropy value of the gray-level co-occurrence matrix to obtain the road surface texture roughness feature value. The formula for calculating texture entropy is: ； Texture entropy reflects the uniformity and complexity of the gray-level distribution in a road surface image; a higher entropy value indicates a coarser road surface texture. For gray levels, Elements in the gray-level co-occurrence matrix The probability of occurrence can effectively distinguish special road surfaces such as sand and mud that are difficult for traditional sensors to identify; S2-5. Calculate the weight coefficients dynamically based on the signal-to-noise ratio corresponding to each feature value. The higher the signal-to-noise ratio, the more reliable the feature data, and the greater the corresponding weight. The sum of all weight coefficients is 1. S2-6. Apply second-order exponential smoothing to the weight coefficient sequence. By setting different smoothing factors, balance the response speed and smoothing effect, suppress weight abrupt changes, and avoid vehicle speed jitter.

3. The adaptive speed control method for special robots based on road surface bump index according to claim 2, characterized in that, The adaptive calculation formula for the bump level boundary value in step S3 is as follows: ； The system dynamically adjusts the bump level boundary value based on load and center of gravity status. The greater the load and the more severe the center of gravity shift, the higher the driving risk corresponding to the same bump index, and the lower the boundary value is accordingly. Specifically: =1,2,3, corresponding to the three dividing points of the fourth level of turbulence; Define boundary values ​​for unloaded flat surfaces; The current payload mass; ρ is the robot's basic mass; ρ is the load sensitivity coefficient. This is the centroid offset; The reference centroid position for the pre-defined storage and transportation state; τ is the centroid offset sensitivity coefficient; Step S3 further includes the following specific steps: S3-0. Based on the preset reference center of mass position for storage and transportation, and combined with the joint angles, joint mass distribution, and end-effector load mass of the robotic arm, the actual center of mass position of the robot is calculated in real time using a rigid body dynamics model. The center of mass offset is then obtained by comparing it with the reference center of mass position. ; S3-1. During the initial operation of the equipment, the no-load boundary values ​​are obtained through standard bumpy road test calibration. ; S3-2. Before starting a single task, complete the no-load boundary values ​​based on IMU vibration data under no-load conditions. Rapid self-calibration; S3-3. During equipment operation, load mass and robotic arm posture data are collected in real time, and the center of gravity position and bump level boundary values ​​are dynamically updated. S3-4. When the center of gravity offset exceeds the preset safety threshold, all bump level boundary values ​​will be uniformly reduced by 15% to achieve conservative speed control. S3-5. Apply upper and lower limit constraints to the boundary values ​​of the bump level to ensure... Furthermore, the boundary value update frequency is limited to no more than 1Hz to avoid frequent fluctuations that could lead to unstable vehicle speed.

4. The adaptive speed control method for special robots based on road surface bump index according to claim 3, characterized in that, Maximum safe vehicle speed threshold in step S4 It is generated using a piecewise nonlinear function, and the calculation formula is as follows: ； An exponential decay model is used to achieve a smooth transition of vehicle speed with the bump rate, avoiding sudden changes in vehicle speed caused by piecewise linear functions; where: The robot's nominal maximum speed; The bump attenuation coefficient controls the rate at which vehicle speed decreases with increasing bump index. It is the safety margin factor and 0 < <1, leaving a safety margin; This represents the minimum controllable speed for the robot. Step S4 further includes the following specific steps: S4-1. Real-time acquisition of vehicle pitch angle θ and roll angle φ, and attitude correction of bump attenuation coefficient κ, using the following correction formula: ； When the vehicle body pitches or rolls, its driving stability decreases, requiring an increased bump damping coefficient and a further reduction in safe speed; among which... Based on the attenuation value, To adjust the gain, These are the robot's maximum allowable pitch angle and maximum roll angle, respectively. S4-2. Apply vehicle speed acceleration constraints, limiting the rate of change of vehicle speed to no more than ± To avoid vehicle instability and load tipping caused by sudden acceleration or deceleration; S4-3. Calculate the road impact energy corresponding to the current vehicle speed in real time. If the impact energy exceeds the limit, the maximum safe vehicle speed will be lowered again. The impact energy calculation formula is: ； The road impact energy is proportional to the robot's total mass, the square of its speed, and the bump index, reflecting the impact intensity of road bumps on the robot's structure and the cargo it carries; among which... For the total mass of the robot, Current vehicle speed; S4-4. Apply upper and lower limit constraints to the maximum safe speed to ensure .

5. The adaptive speed control method for special robots based on road surface bump index according to claim 4, characterized in that, The multi-target dynamic vehicle speed correction in step S5 includes, in sequence, task urgency compensation, robotic arm stability compensation, and center of gravity offset compensation. S5-1. Construct a formula for assessing task urgency: ； The urgency of a task is assessed by combining the remaining time and the progress already completed; the less time remaining and the slower the progress, the higher the urgency. Specifically: Remaining task time, The deadline for the task. This represents the distance already traveled. The total distance to the target. As a weighting coefficient; apply upper and lower limits to the urgency of the task to ensure The urgency of the mission compensates for the vehicle speed as follows: , This is an adjustable urgency gain coefficient; S5-2. Construct the formula for calculating the stability factor of the robotic arm: ； The stability of the robotic arm is evaluated by combining its rotational inertia and external overturning moment. The smaller the stability factor, the more prone the robotic arm is to instability. Among these factors: For load quality, The extension length of the robotic arm. For the maximum permissible moment of inertia, The joint angular velocity, For external forces, The height of the force, For the total mass of the robot, Track ground contact length; stability factor When the value is less than 0, it is assigned the value of 0. After compensation by the robotic arm, the vehicle speed is... When the stability factor When the value is less than 0.3, the angular velocity of the robotic arm joints is also limited to no more than 0.

3. ; S5-3. Constructing a graded centroid offset compensation formula: ； A three-level linear compensation mechanism is adopted; the larger the center of gravity offset, the smaller the compensation coefficient, and the lower the vehicle speed. Maximum permissible centroid offset; centroid compensation coefficient When the value is less than 0, it is assigned the value of 0, and the final corrected vehicle speed is... .

6. The adaptive speed control method for special robots based on road surface bump index according to claim 5, characterized in that, Step S6, the coordinated control of vehicle speed smoothing and load attitude, includes the following steps: S6-0. Construct a discrete prediction model that includes vehicle speed and bump index to predict future changes in system state in the time domain; S6-1. Constructing the model to predict the control objective function: ； Taking into account three optimization objectives: vehicle speed tracking accuracy, vehicle speed change smoothness, and driving impact intensity; among which: To predict the time domain, These are the weighting coefficients. For reference speed, The rate of change of vehicle speed; S6-2. Solve the quadratic programming problem online, output the optimal vehicle speed control sequence, and extract the first control variable as the real-time vehicle speed output; S6-3. Based on the predicted turbulence index over the next three sampling periods, generate the robotic arm end effector attitude feedforward compensation angle. ； ； The PD control algorithm is used to generate the pitch and roll compensation angles of the robotic arm end effector, thus preemptively offsetting the load vibration caused by road bumps; where: These are the pitch compensation angle and tilt compensation angle at the end effector of the robotic arm. These are PID control parameters; S6-4. Based on the predicted turbulence index of the next 3 sampling periods, generate feedforward commands for stabilization of the photoelectric load gimbal. The gimbal stabilization gain is adjusted in real time according to the turbulence level: L1 level uses conventional gain, L2 level increases by 30%, L3 level increases by 60%, L4 level increases by 100%, and the vibration amplitude of the photoelectric load is reduced in coordination with vehicle speed control. As shown in Figure 5, the stabilization gain of the photoelectric payload gimbal exhibits a positively correlated, graded adjustment relationship with the bump level. Differential gain settings balance the gimbal's response speed and stabilization effect: Level L1 (flat road surface) uses 100% conventional gain to ensure rapid target tracking; Level L2 (light bumps) increases gain to 130% to suppress minor vibrations; Level L3 (moderate bumps) increases gain to 160% to offset moderate-intensity bumps; and Level L4 (severe bumps) increases gain to 200% (maximum gain) to strongly suppress severe vibrations. This graded stabilization mechanism, combined with vehicle speed control, can reduce the vibration amplitude of the photoelectric payload by more than 40%, significantly improving the imaging clarity and detection data stability of equipment such as infrared thermal imagers and visible light cameras. S6-5. Apply upper and lower limit constraints to the end effector compensation angle of the robotic arm and the gimbal stabilization gain to ensure... , This is to prevent overcompensation from damaging the robotic arm or gimbal.

7. The adaptive speed control method for special robots based on road surface bump index according to claim 6, characterized in that, The sensor health monitoring and failure protection in step S7 includes the following steps: S7-1. Real-time calculation of sensor data signal-to-noise ratio, variance, and multi-source data consistency indicators to quantitatively assess sensor health. The formula for calculating health is: ； The health status of sensors is assessed by comprehensively evaluating three dimensions: signal quality, data fluctuation, and multi-source consistency; among which, For reference signal-to-noise ratio and reference variance, The consistency coefficient for multi-source data. These are the weighting coefficients, and ; S7-2. When the health status of a single type of sensor is lower than a preset threshold, its weight in the turbulence index calculation is reduced proportionally. For every 10% decrease in health status, the weight is reduced by 20%. S7-3. If the deviation between multi-source fused data and single-source data exceeds 20% within 5 consecutive sampling periods, the multi-source sensor fusion is determined to be abnormal. S7-4. Automatically switches to a simplified dual-source bump model combining an IMU and a track tension sensor: ； In downgraded mode, the bump index is calculated using data from the two most reliable sensors to ensure basic driving safety; among them The characteristic value of vehicle body vibration energy. This represents the characteristic value of track tension fluctuation. S7-5. Speed ​​limit in downgraded mode: In the formula To preset a lower safe speed; S7-6. Trigger equipment degradation alarm and store abnormal data for fault tracing and diagnosis; S7-7. After the sensor health status recovers to the normal threshold, a gradual recovery mechanism is adopted, increasing the sensor weight by 20% every 10 sampling cycles until the normal weight is restored, avoiding vehicle speed fluctuations caused by sudden switching.

8. The adaptive speed control method for special robots based on road surface bump index according to claim 7, characterized in that, The forward-looking path replanning in step S8 includes the following steps: S8-0. Set the grid resolution of the environmental grid bump heatmap to 1m×1m and the time decay coefficient to 0.95, which means that the weight of historical data decays by 5% every hour; S8-1. During normal operation of the equipment, GPS coordinates and turbulence index λ are collected synchronously to construct an environmental grid turbulence heat map. S8-2. Use a time decay algorithm to update the heatmap and reduce the weight of historical old data; S8-3. Detected a presence within 50 meters ahead. When there are continuous high-bump areas, or when the total cumulative bump value of the current path exceeds the threshold, the path replanning procedure is initiated. S8-4. The improved A* algorithm is used to plan the global path, and the node cost function is as follows: ； Add a bump cost factor to prioritize routes with lower bump levels; where: The cost from the starting point to the current node. The heuristic cost from the current node to the destination. Weighting for turbulence costs; S8-5. Constrain the total time of the path after programming to no more than 1.2 times that of the original path, and output a low-bump alternative path; S8-6. When you are 10 meters away from a bumpy area, reduce your speed to the corresponding safe speed level in advance. S8-7. Real-time detection of sudden bumps during driving; when a sudden bump is detected within 10 meters ahead... In the event of a sudden bump, local path replanning is initiated to generate a temporary obstacle avoidance path.

9. The adaptive speed control method for special robots based on road surface bump index according to claim 8, characterized in that, The cloud-based collaborative turbulence prediction in step S9 includes the following steps: S9-0. Federated learning is performed using the federated averaging algorithm. The specific process is as follows: each robot trains the model locally → uploads the model parameters to the cloud platform → the cloud platform aggregates the model parameters → distributes the aggregated model parameters → each robot updates its local model. S9-1. The robot uploads data such as bump index, GPS coordinates, vehicle pose, and road surface image to the cloud platform after desensitizing them. The S9-2 cloud platform uses a federated learning framework to train regional turbulence prediction models while protecting the privacy of the original data. S9-3. The model input parameters include satellite topographic maps, meteorological data, historical turbulence database, and road surface image features; S9-4. Model output: Predicted turbulence index for each environmental grid. And the confidence level, which is calculated based on historical prediction errors; the smaller the prediction error, the higher the confidence level. S9-5. The cloud platform periodically iterates and updates the model, and distributes it to each terminal robot; S9-6. Before the robot enters an unknown area, a safe speed threshold is preloaded based on the predicted bump index. S9-7. When the confidence level of the prediction model is higher than 90%, the pre-deceleration operation is completed 30 meters in advance; when the confidence level is lower than 50%, the prediction result is only used as a reference, and the local sensor data is mainly relied upon. S9-8. The robot feeds back the actual bump data to the cloud platform to complete the model iteration and optimization; S9-9. The cloud platform performs offline verification of the updated model, and only after the verification is passed is it distributed to the robot to ensure that the model performance does not degrade.

10. The adaptive speed control method for special robots based on road surface bump index according to claim 9, characterized in that, The system parameter self-calibration in step S10 includes the following steps: S10-0. Set parameter calibration trigger conditions: after a single task is completed, the continuous driving mileage exceeds 20km, the ambient temperature change exceeds 20℃, and the system detects a change in mechanical performance. S10-1. Establish a parameter calibration database to store driving condition data, control commands, and vehicle response data; S10-2. When the calibration triggering condition is met, offline calibration of control model parameters is completed based on historical data; S10-3. Calibration parameters include bump attenuation coefficient. Load sensitivity coefficient ρ, safety margin factor PID control parameters for end effector compensation of robotic arm and PID control parameters for gimbal stabilization of photoelectric load; S10-4. Taking into account the three objectives of vehicle speed tracking accuracy, bump index prediction accuracy, and driving stability, a parameter calibration optimization objective is constructed. S10-5. Minimize the optimization objective and optimize the model parameters using the gradient descent method; S10-6. When the equipment is running online, the parameters are fine-tuned in real time by recursive least squares method; S10-7. When the parameter change exceeds the preset threshold, the system self-check is triggered to troubleshoot mechanical faults; S10-8. Apply upper and lower limit constraints to the calibrated parameters to ensure that the parameters are within a reasonable range; S10-9. After calibration, the parameters are uploaded to the cloud platform to achieve parameter sharing and collaborative optimization among multiple robots.

Images