A Ground-Based Coupling Adaptive Control Method and System Driven by Ground State Cognition

Abstract

This invention belongs to the field of intelligent perception and control technology for tracked mobile robots. It provides a ground-coupled adaptive control method and system driven by ground state cognition. The method involves collecting multi-source sensor information from the tracked mobile robot during its operation under complex ground conditions and preprocessing it to obtain signal sequences. Based on the signal sequences, ground state features are extracted, and multiple identification evidence sources are constructed from different sensor channels or different feature groups. The credibility of each identification evidence source is determined, and the credibility of multiple identification evidence sources is weighted and fused to obtain the ground category distribution result. A ground state vector is constructed based on the ground category distribution result, and the identification confidence level is calculated. Based on the continuous ground state vector and the identification confidence level, the control target is reconstructed online. A two-layer adaptive control is executed based on the reconstructed control target and the robot's current operating state to achieve stable operation of the robot under complex ground conditions.

Description

Technical Field

[0001] This invention belongs to the field of intelligent perception and control technology for tracked mobile robots, and particularly relates to a ground-coupled adaptive control method and system driven by ground state cognition. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Tracked mobile robots have been widely used in complex environments such as tunnels and mines. Under continuously changing ground conditions such as hard, gravel, slippery, and soft surfaces, the contact state between the robot and the ground will change significantly, directly affecting its traction performance, posture stability, and passage safety.

[0004] Existing technologies mainly fall into three categories: first, ground classification and mode switching schemes based on visual or empirical rules; second, state estimation schemes based on single signals such as vibration and current; and third, fixed-parameter or ordinary adaptive control schemes. However, the above schemes still have the following shortcomings: First, existing identification methods mostly output discrete categories such as "hard ground" and "soft ground," which fail to reflect continuously changing characteristics such as ground adhesion and slip risk, and are also inadequate for describing transitional working conditions, easily leading to abrupt changes in control commands. Second, existing solutions often rely on a single information source, failing to fully utilize the machine-ground coupling relationship between signals such as drive torque, speed decay, and attitude response. This results in insufficient robustness under interference from water accumulation, dust, etc., and a lack of effective mechanisms for handling conflicts from multiple information sources. Finally, existing control strategies often design identification and control separately, using identification results only for mode switching. They cannot reconstruct parameters such as target slip rate and torque distribution online based on continuous changes in ground conditions, nor can they dynamically adjust control intensity based on the reliability of identification results. This leads to problems such as response lag, insufficient anti-slip capability, and attitude instability when the robot suddenly transitions from a hard surface to a slippery surface. Summary of the Invention

[0005] To address at least one of the technical problems mentioned above, this invention provides a ground-state co-coupled adaptive control method and system driven by ground state cognition. This method can stably identify and continuously characterize the ground state under complex ground conditions, and adjust the target slip ratio, target speed, torque distribution, and control parameters online based on the identification results and their reliability, thereby improving the anti-slip capability, attitude stability, and passage safety of tracked mobile robots in complex environments.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of the present invention provides a ground-coupled adaptive control method driven by ground state cognition, comprising the following steps: Multi-source sensor information is collected during the operation of a tracked mobile robot in complex ground conditions, and the signal sequence is obtained through preprocessing. Ground state features are extracted based on signal sequences, including time-domain features, frequency-domain or time-frequency features, and ground-machine coupling response features. Based on different sensing channels and ground state characteristics, cross-decoupling and aggregation are performed to construct multiple identification evidence sources. The credibility of each identification evidence source is determined based on its volatility, stability and inconsistency. The credibility weighting and conflict suppression fusion of multiple identification evidence sources are then performed to obtain the ground category distribution results. Based on the distribution results of ground categories, a ground state vector is constructed, and the recognition confidence is calculated; Based on continuous ground state vectors and identification confidence levels, the control target is reconstructed online. Based on the reconstructed control objective and the robot's current operating state, a two-layer adaptive control is executed. The upper-layer control adjusts the control objective according to the continuous changes in the ground condition, while the lower-layer control adjusts the control gain online based on the control error, identification confidence, and disturbance estimation results, and outputs the left and right track drive control quantities to achieve stable operation of the robot under complex ground conditions.

[0007] Furthermore, the multi-source sensing information includes at least inertial information, drive information, speed information, and attitude response information; wherein, the inertial information includes one or more of longitudinal acceleration, vertical acceleration, yaw rate, and pitch rate; the drive information includes one or more of motor current, motor speed, and output torque; the speed information includes track linear velocity and vehicle body actual speed; and the attitude response information includes one or more of pitch angle, roll angle, and suspension displacement.

[0008] Furthermore, in the ground state characteristics, the time-domain characteristics include one or more of the following: root mean square value, peak factor, kurtosis, and fluctuation intensity; The frequency domain or time-frequency features include one or more of the following: spectral entropy, main band energy ratio, and wavelet subband energy distribution. The ground-machine coupling response characteristics include one or more of the following: torque-slip coupling coefficient, vibration-drive synchronous change characteristics, velocity decay rate, attitude recovery time constant, and ground excitation response delay.

[0009] By performing confidence-weighted and conflict-suppressed fusion on multiple sources of identification evidence, the ground category distribution results are obtained, including: Furthermore, for each source of evidence, an initial basic confidence assignment function is generated based on the Mahalanobis distance between its feature subvector and the feature center values ​​of each ground prototype in the preset ground prototype library; A credibility update mechanism is introduced to calculate the credibility coefficient of each evidence source; The initial basic confidence assignment is weighted and corrected using the confidence coefficient, and the remaining confidence quality is assigned to the uncertain set of the whole set. Based on the corrected confidence assignment, the fusion conflict degree is calculated. When the fusion conflict degree exceeds the preset threshold, the evidence source with the highest inconsistency is subject to secondary penalty and weight reduction. The basic confidence assignment after suppression and adjustment is fused using orthogonal and DS synthesis rules to obtain the ground category distribution results and uncertain category distribution.

[0010] Furthermore, the formula for calculating the credibility of each source of evidence is as follows: , , in, For the first The credibility coefficient of each source of evidence For the first The volatility intensity index of the first evidence source is the first... The initial basic confidence assignment of each evidence source output within the time-domain standard deviation of the current sliding time window. For the first Inconsistency between one source of evidence and other sources of evidence and For adjustment coefficients, The total number of evidence sources, and The first The and the first The initial probability distribution corresponding to each source of evidence. This represents the Jensen-Shannon divergence.

[0011] Furthermore, the formula for calculating the degree of fusion conflict is as follows: , in, To identify the total number of evidence sources, Indicates the first The ground identification proposition of selecting evidence sources The propositions indicating that the intersection of all chosen sources of evidence is an empty set are represented. Indicates the first Each source of evidence corresponds to a proposition. The quality of the basic confidence assignment after reconstruction.

[0012] Furthermore, the control target includes one or more of the following: target slip ratio, target speed, and differential torque limiting parameters for left and right tracks; online reconstruction of the control target specifically includes: determining the target slip ratio based on adhesion capability index and slip risk index; determining the target speed correction amount based on slip risk index and softening degree index; coordinating and adjusting the differential torque limiting parameters for left and right tracks based on vehicle attitude response information; and determining the differential torque limiting parameters for left and right tracks based on adhesion capability index and recognition confidence level.

[0013] A second aspect of the present invention provides a ground-state co-located adaptive control system driven by ground state awareness, comprising: The data acquisition module is used to collect multi-source sensor information during the operation of the tracked mobile robot in complex ground conditions, and preprocess it to obtain signal sequences; The feature extraction module is used to extract ground state features based on signal sequences. The ground state features include time-domain features, frequency-domain or time-frequency features, and ground-machine coupling response features. The state recognition module is used to cross-decouple and aggregate based on different sensing channels and ground state characteristics to construct multiple recognition evidence sources. The credibility of each recognition evidence source is determined based on its volatility, stability and inconsistency. The credibility weighting and conflict suppression fusion of multiple recognition evidence sources are performed to obtain the ground category distribution results. The state characterization module is used to construct a ground state vector based on the ground category distribution results and calculate the recognition confidence. The target reconstruction module is used to reconstruct the control target online based on the continuous ground state vector and the identification confidence level; The adaptive control module is used to perform two-layer adaptive control based on the reconstructed control objective and the robot's current operating state. The upper layer control adjusts the control objective according to the continuous changes in the ground condition, while the lower layer control adjusts the control gain online based on the control error, identification confidence, and disturbance estimation results, and outputs the left and right track drive control quantities to achieve stable operation of the robot under complex ground conditions.

[0014] A third aspect of the present invention provides a computer-readable storage medium.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a ground-coupled adaptive control method driven by ground state cognition as described above.

[0016] A fourth aspect of the present invention provides a computer device.

[0017] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a ground-coupled adaptive control method driven by ground state cognition as described above.

[0018] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention breaks through the traditional fragmented architecture of "recognition → mode switching" and constructs an integrated adaptive closed loop that couples the machine and the ground, from ground state cognition to motion control. It expands discrete ground categories into multi-dimensional continuous state vectors (adhesion capability, roughness, softening degree, slip risk), and calculates the recognition confidence based on the information entropy of the fusion result. Then, using the continuous state vector and confidence, it reconstructs the target slip ratio, target velocity, and differential torque limit coefficient online in real time. Finally, it outputs the left and right track drive torque through a two-layer adaptive control (the upper layer dynamically shapes the target, and the lower layer adjusts the gain online based on error, confidence, and disturbance estimation). This scheme completely eliminates the control command jump problem caused by discrete classification, enabling smooth, stable, and highly passable movement of the robot when transitioning from hard ground to slippery, gravelly, or soft ground conditions.

[0019] 2. This invention constructs vehicle-ground coupling response characteristics, including torque-slip coupling coefficient, velocity decay rate, attitude recovery time constant, and ground excitation response delay, deeply exploring the physical essence of vehicle-ground dynamics interaction, and performing multi-domain fusion by combining time-domain, frequency-domain, and wavelet packet time-frequency characteristics. Compared with single physical quantity detection, this scheme effectively suppresses false alarms in terrain estimation caused by water accumulation, dust, and local impact noise, significantly improving the identification and anti-interference capabilities and robustness under harsh operating conditions.

[0020] 3. This invention proposes an adaptive calculation mechanism for the credibility of evidence sources based on wave intensity and Jensen-Shannon divergence, along with a conflict secondary penalty suppression mechanism. When severe information conflicts arise from sudden environmental changes in multi-source sensors, the system can proactively reduce the weight of unreliable evidence sources and safely transfer the conflict probability to the uncertain set of the entire set, significantly enhancing the filtering fault tolerance and smooth transition performance in unstructured scenarios.

[0021] 4. This invention further introduces conservative control allocation and differential torque dynamic limiting. It integrates adaptive and conservative control through confidence-weighted fusion, while simultaneously coordinating with a disturbance observer for feedforward compensation. This architecture can sensitively fine-tune the gain of each control loop online according to changes in road conditions, effectively preventing track slippage and instability on one side, and balancing trajectory tracking accuracy, attitude maintenance capability, and safe passage performance.

[0022] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0023] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0024] Figure 1 This is a flowchart of a ground-state co-located adaptive control method driven by ground state cognition provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of ground state identification and continuous ground state characterization provided in an embodiment of the present invention; Figure 3 This is a block diagram of the two-layer adaptive control structure provided in the embodiments of the present invention; Figure 4 This is a block diagram of a ground-coupled adaptive control system driven by ground state cognition provided in an embodiment of the present invention. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0026] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0027] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0028] Example 1 like Figure 1 As shown, this embodiment provides a ground-state co-located adaptive control method driven by ground state cognition, including the following steps: Step 1: Collect multi-source sensor information during the operation of the mobile robot under complex ground conditions; It should be noted that the mobile robot in this embodiment is a tracked mobile robot. The robot can be used in scenarios such as tunnel inspection, underground utility tunnel operations, mine roadway inspection, and transportation in complex terrain.

[0029] The complex ground conditions mentioned here specifically refer to various conditions such as hard ground, gravel ground, wet and slippery ground, and soft ground. The ground condition will change continuously with the operating environment.

[0030] In this embodiment, the acquired multi-source sensing information includes at least inertial information, driving information, velocity information, and attitude response information; Specifically, inertial information includes longitudinal acceleration. Vertical acceleration yaw rate and pitch angular velocity One or more of the following; drive information includes motor current. Motor speed and output torque One or more of the following; speed information includes track linear speed. and the actual speed of the vehicle body Attitude response information includes pitch angle. Roll angle and suspension displacement One or more of them.

[0031] The multi-source sensing information comes from the IMU, encoder, current sensor, motor driver feedback unit, attitude measurement unit and suspension displacement sensor installed on the robot body; among them, the track linear velocity is measured by the encoder, and the actual speed of the vehicle body is obtained from the odometer, visual odometer, inertial navigation fusion result or external positioning result.

[0032] It should be noted that the sampling frequency of each sensor can be set according to the platform performance. Preferably, the IMU sampling frequency is 100-500Hz, the encoder and drive feedback frequency is 50-200Hz, and the attitude and suspension response frequency is 50-100Hz.

[0033] Step 2: Preprocess the acquired multi-source sensor information, including time synchronization, filtering and noise reduction, and normalization. In this embodiment, during the operation of the robot, the controller periodically collects raw data from each sensor and performs time synchronization processing on the data from different channels. The synchronized data is then filtered, denoised, and normalized.

[0034] The process employs a sliding time window with a length of 1 second and a step size of 0.2 seconds for windowing. The original sequence within each time window... Perform normalization operation: , In the formula, This is the average value within the current time window. For vertical vibration signals, the standard deviation is [value]. and drive current signal , A preset small positive number (to prevent zero denominator factor) is used to prevent the standard deviation from tending to zero when the robot is stationary or the signal is extremely stable. This can lead to situations where the denominator of the formula is zero and the calculation diverges.

[0035] Step 3: Extract multi-domain feature sets from the preprocessed multi-source sensor information to characterize complex ground conditions, including time-domain features, frequency-domain or time-frequency features, and machine-ground coupled response features; Among them, time-domain features include one or more of the following: root mean square value, peak factor, kurtosis, and fluctuation intensity; Among them, signal root mean square value The overall energy level of the signal is represented as: , Peak factor Used to evaluate the degree of impulse or extremum in a signal, denoted as: , in, This represents the maximum absolute value of the signal.

[0036] cliff The sharpness and impact characteristics reflecting the probability density distribution of signal vibration are expressed as: , Fluctuation intensity It is used to measure the dynamic deviation and dispersion of a signal around its mean, and is expressed as: , Frequency domain or time-frequency characteristics include one or more of the following: spectral entropy, main band energy ratio, and wavelet subband energy distribution. Among them, spectral entropy This extends the concept of information entropy to a wider domain, used to quantify the uniformity of the signal power spectrum distribution across the entire frequency domain, expressed as: , in, For the first The proportion of energy in each frequency band to the total energy.

[0037] Main frequency band energy ratio This refers to the proportion of energy within a specific target frequency band to the total energy across the entire frequency band, expressed as: , in, Main frequency band energy This represents the total energy.

[0038] Wavelet packet subband energy distribution refers to the distribution of signal energy corresponding to each frequency subband in the overall energy after wavelet packet decomposition of the original vibration signal. It is used to characterize the energy characteristics of the vibration signal in different frequency ranges.

[0039] Specifically, the vibration signal is first analyzed. Layer wavelet packet decomposition, the frequency domain will be equally divided into The signal is divided into sub-bands; then, the energy of the reconstructed signal corresponding to each sub-band is calculated, and the energy of each sub-band is normalized to obtain the energy proportion of each frequency sub-band; finally, the normalized energy proportions of each sub-band are combined to form the wavelet packet sub-band energy distribution feature vector. , is represented as: , in, For the first The normalized energy percentage corresponding to each sub-band.

[0040] Under different ground conditions, the energy distribution of vibration signals in each frequency sub-band varies significantly. Therefore, the energy distribution characteristics of the wavelet packet sub-band can be used to identify complex ground conditions.

[0041] The ground-machine coupling response characteristics are one of the key features of this embodiment. Unlike schemes that utilize only a single vibration signal or a single current signal, this embodiment characterizes the ground state by examining the relationship between the drive input, slip response, attitude disturbance, and vibration response. Specifically, it includes the following features: Torque-slip coupling coefficient Used to characterize drive torque input The degree of correlation between the slip response and the surface adhesion; this characteristic changes significantly when the surface adhesion decreases or the slipperiness increases, and its expression is: , in, Indicates the first The driving torque at each sampling point Indicates the driving torque within the current time window The arithmetic mean, Indicates track slip ratio, , For the first Track linear velocity at each sampling point For the first The actual speed of the vehicle at each sampling point To prevent tiny positive numbers with a denominator of zero; This represents the arithmetic mean of the slip rates within the current time window. Indicates the number of sampling points; Vibration-current synchronous change index It is used to reflect the consistency between vehicle body vibration and changes in drive load, and is defined as vertical acceleration. With drive current Pearson correlation coefficient within the sliding time window; Velocity decay rate This reflects the efficiency of converting the driving input into actual forward speed, and is expressed as: , Attitude recovery time constant , used to describe the recovery speed of a robot from pitch disturbance after being excited by the ground, is defined as the time required for the pitch angle to return to a stable value from its peak value in the response curve after being subjected to a ground impact; Ground excitation response delay It is used to describe the time delay relationship between the drive input and the vibration response, and is defined as the drive torque. With vertical acceleration The time delay corresponding to when the correlation function between them reaches its maximum value.

[0042] Combine features from multiple domains to form a total feature vector: , in, Represents the time-domain eigenvector. Represents the frequency domain or time-frequency eigenvector. This represents the characteristic vector of the machine-ground coupled response.

[0043] Step 4: Construct multiple sources of identification evidence, determine the corresponding credibility based on the volatility, stability and inconsistency of each source of identification evidence, and perform credibility weighting and conflict suppression fusion on multiple sources of identification evidence to obtain the ground category distribution results; like Figure 2 As shown, the specific steps include the following: Step 401: Based on multi-source sensor information and multi-domain feature groups, perform cross-decoupling and aggregation to construct three physically complementary independent sources of identification evidence; To achieve highly robust ground condition recognition, this invention constructs a multi-source information fusion recognition algorithm based on the Dempster-Shafer Theory (DST) framework. See also... Figure 2 The specific progressive derivation and logical integration process is as follows: To avoid misjudgment due to transient interference from a single feature or sensor, this embodiment cross-decouples and aggregates the different sensing channels acquired in step 1 and the multi-domain feature groups extracted in step 3, constructing three physically complementary independent sources of identification evidence. ; Source of evidence —Vibration characteristic sources in the time and frequency domains: based on vertical acceleration Temporal features (root mean square value) extracted from sensor channels , cliff ) and frequency domain characteristics (spectral entropy) Main frequency band energy ratio Construct the first feature vector The source of the evidence. Used to characterize the macroscopic roughness of the ground and the dynamic impact response of the chassis.

[0044] Source of evidence —Time-varying characteristic source of dynamics in the time-frequency domain: based on longitudinal acceleration yaw rate and suspension displacement Wavelet packet energy distribution vector extracted from the sensing channel Constructing the second feature vector The source of the evidence. Used to capture transient time-frequency domain features of mobile robots as they traverse unstructured obstacles or abrupt terrain changes.

[0045] Source of evidence —The dynamic characteristic source of machine-ground coupling: it is based on motor current. Output torque Track linear velocity and the actual speed of the vehicle body Machine-to-ground coupling characteristics extracted from sensor channels (torque-slip coupling coefficient) Velocity decay rate Ground excitation response delay Constructing the third feature vector The source of the evidence. Based on the essence of vehicle dynamics, it is used to characterize the current adhesion limit and slip instability risk of the ground.

[0046] Step 402: Generate an initial basic confidence assignment for each source of identified evidence; The identification framework for the ground prototype is constructed from offline test samples under typical ground conditions through feature extraction, normalization, and statistical calibration. It can also be initialized and corrected by combining empirical settings or clustering results.

[0047] Let the identification framework of the system's ground prototype be: , in, It is a hard surface. The ground is made of gravel. The ground is slippery. It is a soft ground.

[0048] For each source of identified evidence (in Its eigenvectors are represented as Within the current sliding time window, the Mahalanobis distance between the feature vector and the center values ​​of the feature features of each ground prototype in the preset ground prototype library is calculated. This is transformed into the local likelihood of each surface prototype, and then used to generate its corresponding initial basic confidence assignment function. It satisfies the following constraints: , In the formula, Indicates the first One source of evidence was assigned to the ground prototype. The initial basic probability, This represents the complete set of identification frameworks composed of various landform prototypes.

[0049] Step 403: Introduce a credibility update mechanism to calculate the credibility coefficient of each evidence source; To improve the robustness of identification under conditions of interference such as water accumulation, dust, and impact, this embodiment introduces a credibility update mechanism for each evidence source. The credibility of an evidence source is updated based on its volatility within the current time window and its consistency with other evidence sources. The calculation formula is as follows: , in, For the first The credibility coefficient of each source of evidence For the first The volatility intensity index of the first evidence source is the first... The initial basic confidence assignment of each evidence source output within the time-domain standard deviation of the current sliding time window. For the first Inconsistency between one source of evidence and other sources of evidence and This is the adjustment coefficient.

[0050] In this embodiment, inconsistency Adopting the first The Jensen-Shannon divergence mean of one source of evidence with all other sources of evidence is measured and expressed as: , in, The total number of evidence sources, and The first The and the first The initial probability distribution corresponding to each evidence source is represented by the Jensen-Shannon divergence. Using the Jensen-Shannon divergence helps improve the symmetry and stability of the inconsistency measure.

[0051] Preferably, in this embodiment If the noise level is high, the volume can be increased appropriately. .

[0052] Step 404: Use the calculated confidence coefficients to perform weighted correction on the initial basic confidence assignment to obtain the corrected confidence assignment; Using the calculated credibility coefficient The initial basic confidence assignment is weighted and corrected to reconstruct the basic confidence assignment function. It satisfies the following constraints: , in, For the revised version One source of evidence was assigned to the ground prototype. Confidence level; To assign to the total set of uncertainties representing the total uncertainty of the system The confidence quality, the This represents an unknown or ambiguous state of system cognition (i.e., the current ground belongs to one of the known terrain types, but due to strong noise interference, it is impossible to determine which one it is).

[0053] In this way, when the credibility of a channel decreases, the probability of its specific ground location will be reduced proportionally, and the remaining untrusted part (i.e., Then it will automatically and safely transfer to the indeterminate set of the whole set. This design prevents high-noise, high-conflict false recognition probabilities from being forcibly input into the underlying adaptive control system by adjusting the confidence quality of the conflict and disturbance components. The process of smooth absorption and isolation buffering ensures the stability of subsequent multi-source fusion decisions and provides clear confidence quality input for the conservative transition of the subsequent controller.

[0054] Step 405: Based on the corrected confidence assignment, calculate the fusion conflict degree, determine whether the fusion conflict degree exceeds the set threshold, and perform confidence-weighted fusion according to different results to obtain the ground category distribution results; This invention introduces a fusion conflict degree. This is used to quantitatively characterize the severity of the contradictions among the three revised sources of evidence. The degree of fusion conflict... Defined as the sum of the probabilities of completely mutually exclusive ground types pointed to by each source of evidence (i.e., the intersection is an empty set), its general calculation formula is: , In the formula, To identify the total number of evidence sources, Indicates the first The ground identification proposition selected from the evidence sources (which can be specific ground) or an uncertain set ), The propositions indicating that the intersection of all chosen sources of evidence is an empty set are represented. Indicates the first Each source of evidence corresponds to a proposition. The reconstructed basic confidence assignment quality. Due to the uncertainty of the entire set... The introduction of, due to It naturally plays an excellent role in conflict transition and buffering during computation, avoiding distortion of the probability space.

[0055] When the system is on a complex transitional surface or experiences a strong physical impact, the sensors are prone to producing conflicting conclusions. The system then determines whether the fusion conflict level exceeds a preset conflict threshold. (In the embodiments of the present invention, it is preferably set) ): like This indicates that the conflict is within a safe range, and conventional synthesis fusion is performed directly (i.e., multi-source fusion is performed using orthogonal and DS synthesis rules). like This indicates a serious conflict between the channels. At this point, a secondary conflict suppression strategy is implemented: the current inconsistency is identified. The biggest source of conflict is the introduction of a penalty and weighting factor. (Preferred setting) Apply a second penalty to the credibility coefficient of the conflict source: This effectively suppresses its weight in the final fusion until the conflict degree is recalculated. Below the threshold.

[0056] Suppression and Adjustment of Basic Confidence Assignment Function Multi-source fusion was performed using orthogonal geometry and Dempster's Rule of Combination to calculate the ground category distribution results under the final identification framework. and uncertain category distribution : , , In the formula, To be allocated to the fused first The final probabilities of the ground-like prototypes constitute the final ground category distribution result output by the embodiment and serve as direct input for constructing the continuous ground state vector in the next step. Uncertain category distribution. This quantitatively characterizes the probability that a multi-source fusion perception system cannot definitively classify the current terrain into any single ground prototype (e.g., hard, gravel, slippery, soft). When the robot enters unknown complex terrain or is disturbed by external physical impacts, the system assigns mass to uncertain propositions across the entire set. Automatically increases (tends to 1) to prevent the system from being misclassified, building a security barrier at the perception front end.

[0057] In this way, the embodiments of the present invention can effectively suppress abnormal conflicts between sensors and achieve stable, low-noise, and high-real-time confidence output for four prototypes: hard, gravel, wet and slippery, and soft.

[0058] Step 5: Construct a ground state vector based on the ground category distribution results and calculate the recognition confidence score; In this embodiment, the ground category distribution results correspond to multiple preset ground prototypes, let the first one be... The prototype state vector corresponding to the ground-like surface is Construct a continuous ground state vector: , in, Indicates adhesion ability index, Represents the roughness index, Indicators representing the degree of softening This indicates a slippage risk indicator.

[0059] In this embodiment, the prototype state vectors of the four typical ground types are preferably set as follows: Hard surfaces: , gravel ground: , Slippery surfaces: , Soft ground: , The merged ground category distribution is as follows The continuous ground state vector is then expressed as: , Calculate the recognition confidence level based on the information entropy of the fusion result: , in, Indicates the fusion result Information entropy ,because ,therefore .

[0060] Step 6: Based on continuous ground conditions and identification confidence, reconstruct the control target online; In this embodiment, the control target includes one or more of the following: target slip ratio, target speed, and differential torque limiting parameters of the left and right tracks; enabling the controller to dynamically adjust the control target according to changes in ground conditions and the reliability of identification, thereby improving the system's adaptability to complex ground conditions.

[0061] In each control cycle, based on the current ground state vector and identification confidence Online adjustment of control objectives. This specifically includes online reconfiguration of control objectives, including: The target slip ratio is determined based on adhesion capacity indicators and slip risk indicators; In this embodiment, the target slip ratio Reconstruct according to the following formula: , in, Indicates the baseline target slip ratio. Indicates the adhesion adjustment coefficient. This represents the slippage risk adjustment coefficient. This represents the current estimated value of the ground adhesion coefficient. This indicates the current slip risk indicator.

[0062] In this embodiment, .

[0063] The target speed correction amount is determined based on slip risk indicators and softening degree indicators; Target speed Reconstruct using the following formula: , in, Indicates the reference target speed. This represents the slippage risk adjustment coefficient. Indicates the adjustment coefficient for the degree of ground softening. In this embodiment, an index representing the degree of ground softening is used. .

[0064] To improve the mobile robot's anti-slip and stable steering capabilities on low-adhesion or asymmetrical transitional terrain, the system adaptively adjusts the torque distribution weights of the two tracks. It is particularly important to note that when identifying confidence levels... Approaching When this occurs, it indicates that there is a serious conflict or strong interference between the multi-source sensor information. At this time, the continuous ground state vector calculated in step 5 will be lost. (especially the adhesion index) It is no longer credible. If it is still... Directly using it to calculate the differential torque limit coefficient may result in an incorrectly excessive differential torque output on low-adhesion surfaces, which could exacerbate the risk of slippage or roll, creating a vicious feedback loop of "unreliable perception → dangerous control output → further deterioration of the condition".

[0065] To address this issue, this invention designs a confidence-weighted modulation adhesion index, which is calculated as follows: , In the formula, This is a preset low-adhesion safety constant. When When (perception is completely trustworthy), Control relies entirely on real-time identification results; when When (perception is completely unreliable), The system forcibly disconnects from unreliable systems. The dependency is converted to a preset safe and conservative value.

[0066] In this invention, it is preferred to set .

[0067] Based on this, a modulated adhesion index is introduced. Reconstructed differential torque limiting coefficient .

[0068] In this embodiment of the invention, the differential torque limiting coefficient is used as a torque distribution parameter. Reconstruct according to the following formula: , After reconfiguration, the system uses the reconfigured differential torque limiting coefficient when distributing torque between the left and right tracks. The absolute value of the differential torque between the left and right tracks Dynamic amplitude limiting constraints are expressed as follows: , in, This represents the differential torque limiting coefficient after reconstruction, used to dynamically constrain the maximum permissible driving torque deviation between the two tracks, preventing slippage and instability on one side; This represents a preset reference differential torque limiting coefficient. In this embodiment of the invention, preferably, it is set to... ; This represents the adhesion index in the continuous ground state vector calculated at the current moment; This represents a preset low-confidence safety protection coefficient, used to conservatively tighten the drive difference when the terrain identification height is uncertain. In this embodiment of the invention, preferably, it is set to... ; This represents the total driving traction torque of the robot at the current moment.

[0069] Step 7: Execute two-layer adaptive control based on the reconstructed control objective and the robot's current operating state; wherein, the upper layer control adjusts the control objective according to the continuous changes in the ground state, and the lower layer control adjusts the control gain online based on the control error, identification confidence and disturbance estimation results, and outputs the left and right track drive control quantities to achieve stable operation of the robot under complex ground conditions.

[0070] like Figure 3 As shown, in this embodiment, the upper-layer control receives the control target obtained in step 6. Dynamic trajectory shaping and adaptive coordination are implemented using measured vehicle states and confidence levels, specifically including: Target speed and target slip ratio Introducing recognition confidence Smooth weighted inertial filtering is used to generate a smoothly changing dynamic desired vehicle speed. With dynamic expected slip ratio : , , In the formula, This is the current sampling control time; The previous sampling control time and The preset fixed filter coefficients are preferably set to [value] in the embodiments. , When identifying confidence level At that time, the actual filter gain and The automatic reduction in speed corresponds to a corresponding increase in filter inertia, significantly lengthening the transition time between the target slip ratio and the target vehicle speed. By introducing confidence level as an adjustment slider for the filter weight, adaptive damping and gradual release of control command jumps on the time scale are achieved, preventing severe impacts on the tracked chassis caused by sudden changes in control quantity.

[0071] Based on the actual measured pitch angle velocity of the vehicle body and roll rate The differential torque limiting coefficient generated in step 6 Perform attitude stability coordination adjustments and generate dynamic differential torque limiting coefficients. : , In the formula, To coordinate the allocation of adjustment factors; in this embodiment of the invention, it is preferable to set When the vehicle body is impacted by the ground, causing an increase in attitude angular velocity, the upper-level control adaptively reduces the differential torque distribution range of the left and right tracks, thereby reducing the imbalance of the left and right drive outputs, suppressing violent fluctuations in the vehicle body attitude, and improving the robot's motion stability and anti-rollover capability in complex ground environments.

[0072] In this embodiment, the lower-level control receives the reconstructed and shaped dynamic target from the upper-level control, calculates the multi-loop error, performs online dynamic self-learning of the control gain, and combines state observation feedforward to output the final physical execution torque.

[0073] The error quantity is defined as follows: , , , in, Indicates speed tracking error. Indicates slip ratio tracking error. Indicates the attitude angle tracking error. This represents the final target velocity after confidence modulation and conservative safety correction. This represents the final target slip rate after confidence modulation and conservative safety correction. Indicates the current pitch angle. Indicates the target pitch angle.

[0074] The control gain update format is as follows: , in, This represents the online update value of the control feedback gain calculated at the next sampling control moment. , For channel The corresponding adaptive feedback control loop has a preset self-updating regulation rate. For channel The corresponding ground condition adjustment function, It is a saturation function. This represents the discrete sampling and gain update period of the control system. This embodiment sets up a disturbance observer to estimate unknown disturbances caused by complex ground conditions. The disturbance estimate is updated in the following form: , in, This represents the estimated disturbance value at the current moment. This represents the estimated disturbance value at the previous moment. This represents the perturbation filter coefficients, in this embodiment .

[0075] The final control output adopts a confidence-weighted form: , in, Indicates the final control output; This indicates the adaptive adjustment control output, calculated as follows: , This indicates a conservative control output, which is a preset safety control value that does not depend on online parameter updates. The total longitudinal traction torque expected to be output by the chassis Defined as the final control output of the system Using terrain confidence Perform weighted linear sliding smooth fusion: , The control quantity can be further allocated to the left and right track drive torque. and : , , in, Total drive torque, The differential torque between the left and right tracks. Dynamic differential torque limiting coefficient of upper control output Constraints: This is to prevent excessive slippage of one side of the track.

[0076] Example 2 like Figure 4 As shown, this embodiment provides a ground-state co-located adaptive control system driven by ground state cognition, including: The data acquisition module is used to collect multi-source sensor information during the operation of the tracked mobile robot in complex ground conditions, and preprocess it to obtain signal sequences; The feature extraction module is used to extract ground state features based on signal sequences. The ground state features include time-domain features, frequency-domain or time-frequency features, and ground-machine coupling response features. The state recognition module is used to construct multiple recognition evidence sources from different sensor channels or different feature groups, determine the corresponding credibility based on the volatility, stability and inconsistency of each recognition evidence source, and perform credibility weighting and conflict suppression fusion on multiple recognition evidence sources to obtain the ground category distribution results. The state characterization module is used to construct a ground state vector based on the ground category distribution results and calculate the recognition confidence. The target reconstruction module is used to reconstruct the control target online based on the continuous ground state vector and the identification confidence level; The adaptive control module is used to perform two-layer adaptive control based on the reconstructed control objective and the robot's current operating state. The upper layer control adjusts the control objective according to the continuous changes in the ground condition, while the lower layer control adjusts the control gain online based on the control error, identification confidence, and disturbance estimation results, and outputs the left and right track drive control quantities to achieve stable operation of the robot under complex ground conditions.

[0077] It should be noted that the specific implementation of the ground state cognition-driven machine-ground coupled adaptive control system of this invention is similar to the specific implementation of the ground state cognition-driven machine-ground coupled adaptive control method of this invention. Please refer to the description in the method section for details. In order to reduce redundancy, it will not be repeated here.

[0078] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in a ground-state co-coupled adaptive control driven by ground state cognition as described above.

[0079] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in a ground-state co-coupled adaptive control driven by ground state cognition as described above.

[0080] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0081] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0082] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0083] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0084] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0085] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A ground-state co-located adaptive control method driven by ground state cognition, characterized in that, Includes the following steps: Multi-source sensor information is collected during the operation of a tracked mobile robot in complex ground conditions, and the signal sequence is obtained through preprocessing. Ground state features are extracted based on signal sequences, including time-domain features, frequency-domain or time-frequency features, and ground-machine coupling response features. The ground-machine coupling response characteristics include one or more of the following: torque-slip coupling coefficient, vibration-drive synchronous change characteristics, velocity decay rate, attitude recovery time constant, and ground excitation response delay. Based on different sensor channels and ground state characteristics, cross-decoupling and aggregation are performed to construct multiple identification evidence sources. The credibility of each identification evidence source is determined based on its volatility, stability, and inconsistency. The multiple identification evidence sources are then weighted by credibility and fused with conflict suppression to obtain the ground category distribution results. The formula for calculating the credibility of each source of evidence is as follows: ， ， in, For the first The credibility coefficient of each source of evidence For the first The volatility intensity index of the first evidence source is the first... The initial basic confidence assignment of each evidence source output within the time-domain standard deviation of the current sliding time window. For the first Inconsistency between one source of evidence and other sources of evidence and For adjustment coefficients, The total number of evidence sources, and The first The and the first The initial probability distribution corresponding to each source of evidence. Indicates the Jensen-Shannon divergence; Based on the distribution results of ground categories, a ground state vector is constructed, and the recognition confidence is calculated; Based on continuous ground state vectors and identification confidence levels, the control target is reconstructed online. Based on the reconstructed control objective and the robot's current operating state, a two-layer adaptive control is executed. The upper-layer control adjusts the control objective according to the continuous changes in the ground condition, while the lower-layer control adjusts the control gain online based on the control error, identification confidence, and disturbance estimation results, and outputs the left and right track drive control quantities to achieve stable operation of the robot under complex ground conditions.

2. The ground-state coherent adaptive control method driven by ground state coherence as described in claim 1, characterized in that, The multi-source sensing information includes at least inertial information, drive information, speed information, and attitude response information; wherein, the inertial information includes one or more of longitudinal acceleration, vertical acceleration, yaw rate, and pitch rate; the drive information includes one or more of motor current, motor speed, and output torque; the speed information includes track linear velocity and vehicle body actual speed; and the attitude response information includes one or more of pitch angle, roll angle, and suspension displacement.

3. The ground-state co-located adaptive control method driven by ground state cognition as described in claim 1, characterized in that, In the ground state characteristics, the time-domain characteristics include one or more of the following: root mean square value, peak factor, kurtosis, and fluctuation intensity; The frequency domain or time-frequency features include one or more of the following: spectral entropy, main band energy ratio, and wavelet packet energy distribution.

4. The ground-state coherent adaptive control method driven by ground state coherence as described in claim 1, characterized in that, By performing confidence-weighted and conflict-suppressed fusion on multiple sources of identification evidence, the ground category distribution results are obtained, including: For each source of evidence, an initial basic confidence assignment function is generated based on the Mahalanobis distance between its feature subvector and the feature center values ​​of each ground prototype in the preset ground prototype library; a confidence update mechanism is introduced to calculate the confidence coefficient of each source of evidence. The initial basic confidence assignment is weighted and corrected using the confidence coefficient, and the remaining confidence quality is assigned to the uncertain set of the whole set. Based on the corrected confidence assignment, the fusion conflict degree is calculated. When the fusion conflict degree exceeds the preset threshold, the evidence source with the highest inconsistency is subject to secondary penalty and weight reduction. The basic confidence assignment after suppression and adjustment is fused using orthogonal and DS synthesis rules to obtain the ground category distribution results and uncertain category distribution.

5. The ground-coupled adaptive control method driven by ground state cognition as described in claim 4, characterized in that, The formula for calculating the degree of fusion conflict is: ， in, To identify the total number of evidence sources, Indicates the first The ground identification proposition of selecting evidence sources The propositions indicating that the intersection of all chosen sources of evidence is an empty set are represented. Indicates the first Each source of evidence corresponds to a proposition. The quality of the basic confidence assignment after reconstruction.

6. The ground-state coherent adaptive control method driven by ground state coherence as described in claim 1, characterized in that, The control objectives include one or more of the following: target slip ratio, target speed, and differential torque limiting parameters for the left and right tracks; Online reconfiguration of control objectives specifically includes: The target slip ratio is determined based on adhesion capacity indicators and slip risk indicators; The target speed correction amount is determined based on slip risk indicators and softening degree indicators; The differential torque limiting parameters for the left and right tracks are determined based on the adhesion capability index and the recognition confidence level.

7. A ground-coupled adaptive control system driven by ground state cognition, characterized in that, A ground-coupled adaptive control method for ground state awareness drive as described in any one of claims 1-6 includes: The data acquisition module is used to collect multi-source sensor information during the operation of the tracked mobile robot in complex ground conditions, and preprocess it to obtain signal sequences; The feature extraction module is used to extract ground state features based on signal sequences. The ground state features include time-domain features, frequency-domain or time-frequency features, and ground-machine coupling response features. The state recognition module is used to cross-decouple and aggregate based on different sensing channels and ground state characteristics to construct multiple recognition evidence sources. The credibility of each recognition evidence source is determined based on its volatility, stability and inconsistency. The credibility weighting and conflict suppression fusion of multiple recognition evidence sources are performed to obtain the ground category distribution results. The state characterization module is used to construct a ground state vector based on the ground category distribution results and calculate the recognition confidence. The target reconstruction module is used to reconstruct the control target online based on the continuous ground state vector and the identification confidence level; The adaptive control module is used to perform two-layer adaptive control based on the reconstructed control objective and the robot's current operating state. The upper layer control adjusts the control objective according to the continuous changes in the ground condition, while the lower layer control adjusts the control gain online based on the control error, identification confidence, and disturbance estimation results, and outputs the left and right track drive control quantities to achieve stable operation of the robot under complex ground conditions.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the ground state coherence driven machine-ground coupled adaptive control method as described in any one of claims 1-6.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the ground-coupled adaptive control method driven by ground state cognition as described in any one of claims 1-6.

Images