Self-adaptive gait control method, system and equipment for quadruped robot in limited space and medium

Abstract

The invention discloses a self-adaptive gait control method, system and device for a quadruped robot in a limited space and a medium, and belongs to the technical field of gait control. Identifying terrain categories through a classifier and acquiring initial gait parameters from a gait library; adjusting the gait parameters according to the terrain gradient and the roughness to obtain optimized gait parameters; calculating stability indexes and fusing scores; when the score is lower than a threshold value, an adjustment strategy is determined according to an instability mode to correct gait parameters; a foot end track is generated and control is executed; and the executed information is fed back to terrain perception to form a closed loop. According to the method, the terrain recognition accuracy is improved through multi-modal information fusion, the adaptive mapping from the terrain to the gait is established to achieve automatic gait switching and parameter optimization, the stability is evaluated through multi-index fusion, a targeted adjustment strategy is selected according to an instability mode, the adaptive gait control problem under the limited space and the complex terrain is solved, and the adaptive gait recognition accuracy is improved. And the movement efficiency and the stability are improved.

Description

Technical Field

[0001] This invention relates to the field of gait control technology, specifically to an adaptive gait control method, system, device, and medium for a quadruped robot in a confined space. Background Technology

[0002] Quadruped robots are increasingly used in confined spaces such as mine inspection, pipeline maintenance, and disaster relief, but existing control methods have significant shortcomings.

[0003] Traditional gait control often uses preset fixed patterns, with gait parameters remaining basically unchanged during operation or simply adjusted based on speed. This approach has poor adaptability in complex terrains. When the robot encounters different surfaces such as slopes, steps, or gravel, it cannot switch to the appropriate gait in time, making it prone to slipping, collisions, or even instability and falls.

[0004] In terms of terrain perception, current methods mainly rely on visual recognition or ground contact feedback. Visual methods are limited by lighting conditions and dust interference in confined spaces, resulting in unstable recognition accuracy and weak ability to identify features such as puddles and small pebbles. Relying solely on foot contact feedback is a passive detection method; by the time a problem is detected, the robot is already in a disadvantageous position, leading to delayed adjustments.

[0005] Stability assessments typically employ static stability margin criteria, focusing solely on geometric relationships and neglecting dynamic factors such as inertial forces, changes in ground friction, and sufficiency of joint torques during motion. This can lead to robots losing control during actual movement, even if they appear geometrically stable, due to dynamic imbalances or joint torque saturation.

[0006] These problems limit the reliable operation of quadruped robots in complex environments with limited space. Summary of the Invention

[0007] In view of the above-mentioned problems, the present invention provides an adaptive gait control method, system, device and medium for quadruped robots in a limited space.

[0008] Therefore, the technical problem solved by this invention is that existing quadruped robot gait control methods suffer from fixed gait patterns and an inability to adjust in real time according to terrain, resulting in low movement efficiency and poor stability in complex terrain. Terrain perception relies on a single sensor, which has low recognition accuracy in environments with insufficient lighting and dust interference in confined spaces, and its judgment of terrain features is lagging, making it impossible to adapt the gait in advance. Stability assessment only considers static geometric relationships, ignoring inertial forces, changes in ground friction, and limitations in joint actuation capabilities during dynamic motion, making it impossible to accurately predict instability risks and lacking a real-time adjustment mechanism.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an adaptive gait control method for a quadruped robot in a confined space, comprising, Collect support reaction force information of each leg, fuselage attitude angle information, and depth image information of the terrain ahead; extract contact force distribution features, attitude change features, and spatial structure features from the support reaction force information, attitude angle information, and depth image information respectively, and combine them into a terrain feature vector; The terrain feature vector is input into a classifier to obtain the terrain category; the corresponding basic gait pattern is obtained from the gait library according to the terrain category, and the initial gait parameters in the basic gait pattern are obtained; The terrain slope value and terrain roughness value are extracted from the terrain feature vector. The step parameter in the initial gait parameters is adjusted according to the terrain slope value, and the ground clearance parameter in the initial gait parameters is adjusted according to the terrain roughness value to obtain the optimized gait parameters. Based on the optimized gait parameters, a stability index is calculated, and the stability index is fused to obtain a comprehensive stability score. When the overall stability score is lower than the safety threshold, an adjustment strategy is determined based on the value of the stability index, and the optimized gait parameters are corrected according to the adjustment strategy. The foot trajectory of each leg is generated based on the corrected optimized gait parameters, and the foot trajectory is converted into joint angle commands through inverse kinematics and then executed for control. The process of feeding back the support reaction force information and attitude angle information obtained after the execution of control to the extraction process of terrain feature vectors.

[0010] As a preferred embodiment of the adaptive gait control method for a quadruped robot in a confined space according to the present invention, the step of extracting contact force distribution features includes: Zero drift compensation is performed on the collected support reaction force information of each leg to obtain the compensated support reaction force; The compensated support reaction force is filtered to obtain the filtered support reaction force; Based on the filtered support reaction force, the uniformity of force distribution between each leg is calculated to obtain the contact force distribution characteristics.

[0011] As a preferred embodiment of the adaptive gait control method for a quadruped robot in a confined space described in this invention, the method includes: extracting posture change features, including: Obtain the fuselage attitude angles at the current and historical moments; Based on the fuselage attitude angles at the current and historical moments, calculate the time rate of change of the attitude angles; The rate of change over time is used as the characteristic of the attitude change.

[0012] As a preferred embodiment of the adaptive gait control method for a quadruped robot in a confined space described in this invention, the method includes: extracting spatial structural features, including: The depth image information is segmented into ground regions to obtain ground depth data; Geometric modeling is performed on the ground depth data to obtain ground geometric parameters; Based on the ground geometric parameters, the slope and undulation characteristics of the terrain are extracted as the spatial structure features.

[0013] As a preferred embodiment of the adaptive gait control method for a quadruped robot in a confined space according to the present invention, the step of calculating the stability index based on the optimized gait parameters includes: Obtain the position and acceleration of the fuselage's center of mass; Calculate the horizontal coordinates of the zero torque point based on the center of mass position, center of mass acceleration, and total mass of the robot. Calculate the vertical distance from the zero moment point to each boundary of the supporting polygon, and take the minimum distance value as the first stability index; Calculate the distance from the projection point of the centroid onto the horizontal plane to each boundary of the supporting polygon, and take the minimum distance value as the second stability index; Obtain the current torque and maximum allowable torque of each joint, calculate the ratio of the current torque to the maximum allowable torque of each joint, calculate the torque margin of each joint based on the ratio, and take the minimum torque margin as the third stability index.

[0014] The beneficial effects of this preferred technical solution are as follows: dynamic motion balance is evaluated by calculating the zero-moment point position; geometric support margin is evaluated by the support polygon margin; and executive capacity reserve is evaluated by the joint moment margin. These three indicators comprehensively characterize gait stability from different dimensions. Compared with the traditional single evaluation method that only uses static stability margin, this solution can accurately identify different instability modes such as dynamic instability, insufficient support, and joint saturation.

[0015] As a preferred embodiment of the adaptive gait control method for a quadruped robot in a confined space according to the present invention, wherein: determining the adjustment strategy based on the value of the stability index includes: Determine whether the first stability index is less than the first threshold; If the first stability index is less than the first threshold, then a timing adjustment strategy is selected, and the gait cycle increment is calculated based on the difference between the first stability index and the first threshold, and the current gait cycle is increased by the gait cycle increment. Determine whether the second stability index is less than the second threshold; If the second stability index is less than the second threshold, then the configuration adjustment strategy is selected, and the support width increment is calculated based on the difference between the second stability index and the second threshold, thereby increasing the support span between the four legs. Determine whether the third stability index is less than the third threshold; If the third stability index is less than the third threshold, then a load adjustment strategy is selected to reduce the stepping parameter and the ground clearance parameter in the optimized gait parameters.

[0016] The beneficial effects of this preferred technical solution are as follows: It triggers a timing adjustment strategy based on the first stability index to reduce motion speed; it triggers a configuration adjustment strategy based on the second stability index to expand the support area; and it triggers a load adjustment strategy based on the third stability index to reduce joint burden, thus achieving a precise match between instability modes and adjustment measures. Compared to traditional uniform adjustment methods, this solution adopts differentiated strategies for different causes of instability, avoiding the loss of motion efficiency due to over-adjustment and the safety hazards caused by under-adjustment, ensuring stability while also considering motion performance.

[0017] As a preferred embodiment of the adaptive gait control method for a quadruped robot in a confined space according to the present invention, the step of generating the foot trajectory of each leg based on the corrected optimized gait parameters includes: The swing phase duration and support phase duration of each leg are determined based on the corrected optimized gait parameters. For the leg in the swing phase, the starting and ending positions of the foot are determined according to the stepping parameters in the optimized gait parameters, and the maximum ground clearance of the foot is determined according to the ground clearance parameters in the optimized gait parameters. A foot trajectory is generated during the swing phase. The vertical component of the foot trajectory is interpolated parabolically so that the foot reaches the maximum height off the ground at the midpoint of the swing phase. The horizontal component is interpolated at a constant speed from the starting position to the ending position.

[0018] This invention provides an adaptive gait control system for a quadruped robot in a confined space.

[0019] To solve the above technical problems, the present invention provides the following technical solution: an adaptive gait control system for a quadruped robot in a limited space, comprising: a multimodal terrain perception module, used to collect support reaction force information of each leg, body posture angle information and depth image information of the terrain in front, and extract contact force distribution features, posture change features and spatial structure features from the support reaction force information, posture angle information and depth image information respectively, and combine them into a terrain feature vector; The terrain classification and recognition module is used to input the terrain feature vector into the classifier to obtain the terrain category; The gait library and selection module is used to obtain the corresponding basic gait pattern from the gait library according to the terrain category, and to obtain the initial gait parameters in the basic gait pattern; The gait parameter optimization module is used to extract terrain slope value and terrain roughness value from the terrain feature vector, adjust the step parameter in the initial gait parameters according to the terrain slope value, and adjust the ground clearance parameter in the initial gait parameters according to the terrain roughness value to obtain optimized gait parameters; The stability assessment module is used to calculate stability indices based on the optimized gait parameters and fuse the stability indices to obtain a comprehensive stability score. The gait adjustment decision module is used to determine an adjustment strategy based on the value of the stability index when the comprehensive stability score is lower than the safety threshold, and to correct the optimized gait parameters according to the adjustment strategy. The underlying execution control module is used to generate foot trajectories for each leg based on the corrected optimized gait parameters, convert the foot trajectories into joint angle commands through inverse kinematics and execute the control, and feed back the support reaction force information and attitude angle information measured after the control is executed to the multimodal terrain perception module.

[0020] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the adaptive gait control method for a quadruped robot in a limited space.

[0021] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the adaptive gait control method for a quadruped robot in a confined space.

[0022] The beneficial effects of this invention are as follows: by integrating multimodal information from foot force sensors, inertial measurement units, and vision cameras, it is possible to comprehensively perceive terrain features from three dimensions: force distribution, attitude change, and spatial structure. This overcomes the limitations of single sensors in complex environments and improves the accuracy and robustness of terrain judgment.

[0023] A mapping relationship between terrain categories and gait patterns was established, enabling automatic gait switching. Gait parameters were optimized in real time based on terrain slope and roughness, allowing the robot to select the most suitable movement mode for different terrains and improving its terrain adaptability. A triple-index evaluation of stability was conducted, using zero-moment point, support polygon, and joint torque, comprehensively considering dynamic balance, geometric support, and execution capability to accurately determine gait stability. When insufficient stability was detected, targeted adjustment strategies were selected based on the specific instability mode, achieving a shift from passive response to active maintenance. By using real-time sensor feedback information for terrain perception in the next cycle, a closed-loop control was formed, enabling the robot to continuously adapt to terrain changes and ensuring safe and stable operation in complex environments within confined spaces. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 The above is a flowchart of an adaptive gait control method for a quadruped robot in a confined space, provided as an embodiment of the present invention. Detailed Implementation

[0026] To make the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0027] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides an adaptive gait control method for a quadruped robot in a confined space, comprising: Step 1: Collect information on the reaction force of each leg, the attitude angle of the fuselage, and the depth image of the terrain ahead; extract contact force distribution features, attitude change features, and spatial structure features from the reaction force information, attitude angle information, and depth image information, and combine them into a terrain feature vector; Step 2: Input the terrain feature vector into the classifier to obtain the terrain category; according to the terrain category, obtain the corresponding basic gait pattern from the gait library, and obtain the initial gait parameters in the basic gait pattern; Step 3: Extract the terrain slope value and terrain roughness value from the terrain feature vector, adjust the step parameter in the initial gait parameters according to the terrain slope value, and adjust the ground clearance parameter in the initial gait parameters according to the terrain roughness value to obtain the optimized gait parameters; Step 4: Calculate the stability index based on the optimized gait parameters, and fuse the stability indexes to obtain a comprehensive stability score; Step 5: When the overall stability score is lower than the safety threshold, determine the adjustment strategy based on the value of the stability index, and correct the optimized gait parameters according to the adjustment strategy; Step 6: Generate foot trajectory for each leg based on the corrected optimized gait parameters, convert the foot trajectory into joint angle commands through inverse kinematics, and execute the control. Step 7: Feed back the support reaction force information and attitude angle information obtained after the execution of control to the terrain feature vector extraction process.

[0028] When quadruped robots operate in confined spaces such as mines and utility tunnels, the terrain conditions are complex and varied, including flat ground, slopes, steps, and gravel, with insufficient lighting and severe dust interference. Traditional gait control uses fixed gait patterns, and gait parameters remain largely unchanged during operation. This makes it difficult to adjust in time when encountering terrain changes, leading to slippage, collisions, and even instability. Single-mode visual perception has low accuracy in confined spaces, struggling to identify features such as water accumulation and fine gravel, while ground contact feedback is a passive detection method with delayed adjustments. Existing stability assessments often use static stability margins, considering only geometric relationships and ignoring dynamic factors such as inertial forces, changes in ground friction, and joint torque saturation during movement, making it impossible to accurately predict instability risks. To address these issues, this embodiment integrates multimodal perception of foot force, posture, and vision to establish an adaptive mapping from terrain to gait. It employs a multi-index fusion stability assessment and selects targeted adjustment strategies based on instability patterns, forming a closed-loop control system of perception-decision-execution to achieve terrain adaptation and stability assurance.

[0029] This embodiment constructs a complete adaptive gait control process through steps one through seven. Step one collects multi-source sensor information and extracts terrain feature vectors, solving the problem of inaccurate identification by a single sensor in complex environments and providing comprehensive and accurate terrain information for subsequent decision-making. Step two identifies terrain categories based on terrain features and matches them with basic gait patterns, achieving coarse-grained adaptive gait selection. Step three finely adjusts gait parameters based on terrain slope and roughness, completing a two-stage optimization from coarse selection to fine adjustment, ensuring precise matching between gait parameters and terrain features. Step four calculates multi-dimensional stability indices and fuses scores, achieving a comprehensive and accurate assessment of gait stability. Step five selects adjustment strategies and corrects gait parameters based on specific instability modes, transforming passive response into active maintenance, and adjusting promptly to avoid instability when stability is insufficient. Step six generates foot trajectories and executes control, transforming the gait parameters of the decision layer into the joint movements of the underlying layer. Step seven feeds back the executed sensor information to the perception stage, forming a closed-loop control that enables the system to continuously adapt to terrain changes. The entire process organically integrates terrain perception, gait decision-making, stability assurance, and execution control, solving the problem of adaptive control in complex terrain within a limited space.

[0030] Example 2, an embodiment of the present invention, provides an adaptive gait control method for a quadruped robot in a confined space based on the previous embodiment, comprising: Step 1: Collect support reaction force information for each leg, fuselage attitude angle information, and depth image information of the terrain ahead; extract contact force distribution features, attitude change features, and spatial structure features from the support reaction force information, attitude angle information, and depth image information, and combine them into a terrain feature vector, including the following steps A1-A9: A1: Perform zero-drift compensation on the collected support reaction force information of each leg to obtain the compensated support reaction force; A2: Filter the compensated support reaction force to obtain the filtered support reaction force; A3: Based on the filtered support reaction force, calculate the uniformity of force distribution between each leg to obtain the contact force distribution characteristics.

[0031] A4: Obtain the fuselage attitude angles at the current and historical moments; A5: Calculate the time rate of change of the attitude angle based on the fuselage attitude angle at the current moment and the historical moment; A6: The time change rate is used as the attitude change feature.

[0032] A7: Perform ground region segmentation on the depth image information to obtain ground depth data; A8: Perform geometric modeling on the ground depth data to obtain ground geometric parameters; A9: Based on the ground geometric parameters, extract the slope and undulation features of the terrain as the spatial structure features.

[0033] In this embodiment of the application, in step A2, the filtering process is carried out by: using a second-order low-pass filter to filter the compensated support reaction force. The filter adopts the Barthzinsky design method, and the cutoff frequency is set to 50 Hz. The filter coefficients are calculated based on the sampling frequency of 1000 Hz. The support reaction force measurement values ​​at the current time and the previous two time points are weighted and summed with the filtered output values ​​at the previous two time points to obtain the filtered support reaction force at the current time. This effectively removes high-frequency noise in the sensor measurement and retains the low-frequency component that reflects the actual ground contact.

[0034] In an optional implementation, in step A2, the filtering process can be performed by: using a Kalman filter to filter the compensated support reaction force, establishing a state-space model of the support reaction force, predicting the current state of the support reaction force based on the system dynamic model, correcting it by combining the actual measured value, calculating the Kalman gain through the state estimation covariance matrix, and optimally fusing the predicted value and the measured value, so as to track the dynamic changes of the support reaction force while suppressing measurement noise.

[0035] In another optional implementation, in step A2, the filtering process can also be performed by: using median filtering to filter the compensated support reaction force, setting the sliding window length to 5 sampling points, extracting the support reaction force measurement values ​​at each time within the window, sorting them by value and taking the median value as the filtered output at the current time, which effectively suppresses pulse interference and outliers, and is suitable for terrain contact scenarios with sudden impacts.

[0036] In this embodiment of the application, in step A8, the geometric modeling is performed by: modeling the ground depth data using the least squares plane fitting method, substituting the three-dimensional coordinates of each point in the ground point cloud into the plane equation, constructing an objective function for the sum of squared errors of the plane parameters, solving the coefficients of the plane equation by minimizing the sum of squared errors, obtaining the optimal fitting plane, and extracting the plane normal vector as the ground geometric parameters. This method has high computational efficiency and the fitting accuracy meets the requirements of real-time control.

[0037] In an optional implementation, in step A8, geometric modeling can be performed by: modeling the ground depth data using the RANSAC robust plane fitting method, randomly selecting three ground points to construct candidate planes, calculating the distances from the remaining points to the candidate planes, counting the number of inliers with distances less than a threshold, and selecting the plane with the most inliers as the optimal fitting result after multiple iterations. This can effectively eliminate interference from outliers such as gravel and obstacles.

[0038] In another alternative implementation, in step A8, geometric modeling can also be performed by: modeling the ground depth data using a surface fitting method, representing the ground as a quadratic surface, solving the surface parameters using the least squares method, which can describe complex terrain with large undulations, and extracting the tangent plane of the surface at the robot's current position as local ground geometric parameters.

[0039] In this embodiment of the application, in step two, the classifier classifies the terrain feature vectors by: using a support vector machine classifier to classify the terrain feature vectors, selecting the radial basis function as the kernel function, collecting feature samples of five typical terrains through the offline training phase, labeling the corresponding terrain categories, solving the support vectors and Lagrange multipliers, calculating the kernel function values ​​of the input feature vectors and each support vector during online runtime, and determining the terrain category by weighted summation and sign function.

[0040] In an optional implementation, in step two, the classifier can classify the terrain feature vector by using a random forest classifier, constructing multiple decision trees to form a forest, with each tree growing based on a randomly selected subset of features and training samples, and inputting the feature vector into each decision tree for classification voting during online operation, selecting the category with the most votes as the final classification result.

[0041] In another alternative implementation, in step two, the classifier can also classify the terrain feature vectors by: using a neural network classifier to classify the terrain feature vectors, constructing a multilayer perceptron containing an input layer, a hidden layer and an output layer, training the network weights through the backpropagation algorithm, inputting the feature vectors into the network for forward propagation during online operation, and the activation values ​​of each node in the output layer representing the probability of the corresponding terrain category, and selecting the category with the highest probability as the recognition result.

[0042] Step four: Calculating stability indices based on the optimized gait parameters, and fusing the stability indices to obtain a comprehensive stability score, includes the following steps B1-B5: B1: Obtain the position and acceleration of the fuselage's center of mass; B2: Calculate the horizontal coordinates of the zero torque point based on the center of mass position, center of mass acceleration, and total robot mass; B3: Calculate the vertical distance from the zero moment point to each boundary of the supporting polygon, and take the minimum distance value as the first stability index; B4: Calculate the distance from the projection point of the centroid onto the horizontal plane to each boundary of the supporting polygon, and take the minimum distance value as the second stability index; B5: Obtain the current torque and maximum allowable torque of each joint, calculate the ratio of the current torque to the maximum allowable torque of each joint, calculate the torque margin of each joint based on the ratio, and take the minimum torque margin as the third stability index.

[0043] Step 5: When the overall stability score is lower than the safety threshold, an adjustment strategy is determined based on the value of the stability index, and the optimized gait parameters are corrected according to the adjustment strategy, including the following steps C1-C6: C1: Determine whether the first stability index is less than the first threshold; C2: If the first stability index is less than the first threshold, then select the timing adjustment strategy, calculate the gait cycle increment based on the difference between the first stability index and the first threshold, and increase the current gait cycle by the gait cycle increment; C3: Determine whether the second stability index is less than the second threshold; C4: If the second stability index is less than the second threshold, then select the configuration adjustment strategy, calculate the support width increment based on the difference between the second stability index and the second threshold, and increase the support span between the four legs; C5: Determine whether the third stability index is less than the third threshold; C6: If the third stability index is less than the third threshold, then select the load adjustment strategy to reduce the stepping parameter and the ground clearance parameter in the optimized gait parameters.

[0044] In this embodiment of the application, in step five, the adjustment strategy is determined by: establishing a rule-based decision-making mechanism, matching the corresponding adjustment strategy type according to the comparison results of the first stability index, the second stability index, the third stability index and their respective thresholds; when the first stability index is lower than the threshold, the timing adjustment strategy is selected to extend the gait cycle and reduce the movement speed; when the second stability index is lower than the threshold, the configuration adjustment strategy is selected to increase the support span and expand the support area; when the third stability index is lower than the threshold, the load adjustment strategy is selected to reduce the step length and step height and reduce the joint load; if multiple indicators exceed the limit at the same time, the corresponding strategies are executed in order of priority.

[0045] In an optional implementation, in step five, the adjustment strategy can be determined by: establishing a decision-making mechanism based on fuzzy reasoning, fuzzifying the values ​​of each stability index into high, medium, and low levels, performing reasoning based on a preset fuzzy rule base, outputting the membership degree of the adjustment strategy, and selecting the strategy with the highest membership degree for execution, which can handle decision uncertainty in the critical state of the index.

[0046] In another optional implementation, the adjustment strategy in step five can also be determined by: establishing a decision mechanism based on cost function optimization, defining a comprehensive cost function that includes stability indicators and motion performance indicators, searching for the combination of adjustment strategies that minimizes the cost function, and taking into account motion efficiency while ensuring stability, thereby achieving multi-objective optimization decision-making.

[0047] Step Six: Generate foot trajectories for each leg based on the corrected optimized gait parameters, convert these foot trajectories into joint angle commands using inverse kinematics, and execute the control, including the following steps D1-D3: D1: Determine the swing phase duration and support phase duration of each leg based on the corrected optimized gait parameters; D2: For the leg in the swing phase, determine the starting and ending positions of the foot according to the stepping parameters in the optimized gait parameters, and determine the maximum ground clearance of the foot according to the ground clearance parameters in the optimized gait parameters; D3: Generate a foot trajectory during the swing phase duration. The vertical component of the foot trajectory is interpolated parabolically so that the foot reaches the maximum height off the ground at the midpoint of the swing phase. The horizontal component is interpolated at a constant speed from the starting position to the ending position.

[0048] Example 3, an embodiment of the present invention, provides an adaptive gait control method for a quadruped robot in a confined space based on the previous embodiment, comprising: In step one, six-dimensional force sensors installed at the ends of the four legs collect the reaction force and torque information of each leg. Each sensor outputs force components in three directions and torque components in three directions, with a sampling frequency set to 1000 Hz to capture the dynamic contact process. Simultaneously, an inertial measurement unit installed at the fuselage center of mass outputs three-axis acceleration, three-axis angular velocity, and attitude quaternions at a frequency of 500 Hz. A binocular vision camera at the front of the fuselage acquires terrain images at a rate of 30 frames per second, with a baseline distance of 0.12 meters and a field of view of 90 degrees.

[0049] For the foot force sensor data, zero drift compensation is first performed. While the robot is stationary, the outputs of each sensor are recorded as the zero-point reference value. During operation, this reference value is subtracted from the real-time measurement values ​​to obtain the true support reaction force. The support reaction force vector of the i-th leg is defined as: (1) In the formula, , , These represent the support reaction force components of the i-th leg at its foot end in the x, y, and z directions in the body coordinate system. To eliminate high-frequency noise, a second-order low-pass filter is used to filter the force signal. (2) In the formula, The filtered support reaction force at time k. The value measured at the current moment. These are the forward coefficients of the filter. The feedback coefficients are determined using the Butterworth filter design method based on a cutoff frequency of 50 Hz and a sampling frequency of 1000 Hz. =0.0201、 =0.0402、 =0.0201、 =-1.5610、 =0.6414.

[0050] For inertial measurement unit data, the pitch and roll angles of the fuselage are extracted as attitude features. Pitch angle The roll angle reflects the degree of tilt of the fuselage in the forward and backward direction. Reflecting the left and right tilt, its value can be directly obtained from the attitude calculation results of the inertial measurement unit. The fuselage attitude vector is defined as... Meanwhile, the acceleration of the computer's center of mass is used for subsequent dynamic stability analysis.

[0051] For visual data, a depth map of the foreground scene is generated using a binocular stereo matching algorithm, followed by ground segmentation to extract terrain regions. A plane fitting method is used to detect slope and height variations in the foreground terrain. The ground point cloud in the depth map is then fitted with a least-squares plane to obtain the ground plane equation. The slope of the terrain is represented by the inclination angle of the normal vector (a, b, c). After extracting terrain features, the multimodal sensing information is integrated into a comprehensive terrain feature vector. This vector contains information such as foot force distribution pattern, fuselage attitude angle, and terrain slope, and serves as the input to the terrain classification module.

[0052] In step two, based on the preprocessed sensor data from step one, key features characterizing terrain features are extracted. First, the force distribution pattern at the ends of the four legs is analyzed, and the imbalance of force at the ends is calculated. The imbalance index of force at the ends is defined as: (3) In the formula, Let be the vertical support reaction force of the i-th leg. This is the average value of the vertical support reaction forces of the four legs. This indicates the degree of unevenness in the distribution of force at the foot. On a flat surface... The value approaches zero, but increases on slopes or uneven ground. Extract the fuselage pitch angle. and roll angle The rate of change is defined as: (4) In the formula, The sampling time interval is set to 0.002 seconds. The rate of change of attitude angle reflects the degree of terrain undulation; the attitude angle changes gradually when walking on flat ground, and changes rapidly when encountering steps or potholes. Roughness features of the terrain ahead are extracted from the visual depth map and represented by the standard deviation of the depth values. (5) In the formula, Let j be the depth value of the j-th ground point. Where N is the average depth and N is the number of ground points. The standard deviation of depth is denoted by . Flat ground has a small standard deviation of depth, while gravel or stepped terrain has a large standard deviation of depth.

[0053] Construct a terrain feature vector by combining the extracted features as follows: (6) In the formula, each feature component has been defined in the preceding equation. This feature vector is input into a terrain classifier for terrain type identification. The terrain classifier uses a support vector machine algorithm to classify terrain into five categories: flat land, uphill, downhill, steps, and obstacles. The decision function of the classifier is: (7) In the formula, For Lagrange multipliers, For training sample labels, The kernel function is expressed as a radial basis function, where b is the bias term and M is the number of support vectors. The classifier outputs terrain category labels. These correspond to five terrain types. This terrain category information is then passed to the gait selection module in the third step.

[0054] In step three, based on the terrain category identified in step two, a corresponding basic gait pattern is selected from a pre-set gait library. The gait library contains five gaits optimized for different terrains: fast gait on flat ground, uphill crawling gait, downhill braking gait, step-crossing gait, and obstacle-avoiding gait. Each gait defines parameters such as leg swing sequence, duty cycle, baseline stride length, and stride height.

[0055] For a fast gait on flat ground, a diagonal trotting motion is adopted, with both diagonal legs swinging simultaneously. The duty cycle is set to 0.5, meaning 50% of the time is in the support phase and 50% in the swing phase. The baseline stride length is set to 0.30 meters and the stride height to 0.08 meters. This gait is fast, energy efficient, and suitable for rapid movement on flat terrain.

[0056] For uphill crawling, a static and stable crawling gait is adopted, with only one leg swinging at a time while the other three legs provide support. The duty cycle is increased to 0.75, the stride length is shortened to 0.20 meters, and the stride height is increased to 0.12 meters to adapt to the slope. This gait ensures that the robot maintains tripod support at all times on the slope, resulting in high stability, and the larger stride height prevents the feet from sliding on the slope.

[0057] For downhill braking gait, the crawling gait is also adopted, but the body's center of gravity is adjusted by shifting it backward to increase the reaction force of the rear leg. The duty cycle is set to 0.70, the stride length to 0.18 meters, and the stride height to 0.10 meters. The body pitch angle is controlled to be higher in the front and lower in the back to reduce the tendency to lean forward when going downhill.

[0058] For the step-crossing gait, a special two-stage gait is adopted. First, the front leg is raised high to step onto the step. After the front leg is stable, the body's center of gravity shifts forward, and then the rear leg is raised to step onto the step. The stride height of this gait is dynamically adjusted according to the step height, which is obtained through visual distance measurement. Then, set the swing leg step height to Leave a 5-centimeter safety margin for the meter.

[0059] For obstacle avoidance gait, the leg coordination pattern of a diagonal trot is maintained, but the direction of movement is adjusted, and an obstacle avoidance path is planned based on the visually recognized obstacle position. The stride length and stride height parameters of this gait are the same as those of the fast gait on flat ground, and obstacle avoidance is achieved by changing the fuselage heading angle.

[0060] The gait selection module outputs the corresponding set of basic gait parameters based on the terrain category C. ,in For gait period, Duty cycle, As the reference step size, This serves as the baseline step height. These basic parameters will be further optimized and adjusted in the fourth step based on the specific characteristics of the terrain.

[0061] In step four, the basic gait parameters selected in step three are standard values ​​designed for typical terrain. However, actual terrain is diverse, requiring fine-tuning of the gait parameters based on specific terrain characteristics to achieve optimal performance. The gait parameter optimization module receives the terrain feature parameters extracted in step two and the basic gait parameters from step three, and calculates the optimal gait parameters using an optimization algorithm.

[0062] For sloping terrain, the step size needs to be adjusted according to the slope angle. The steeper the slope, the smaller the step size should be to ensure stability. The step size adjustment factor is defined as: (8) In the formula, The terrain slope angle (obtained from the feature extraction in Formula 6) This is the step size adjustment coefficient. When the slope is 0 degrees, the coefficient is 1 to maintain the baseline step size; when the slope reaches 45 degrees, the coefficient is 0.5, halving the step size. The optimized step size is... ,in This is the baseline step size determined in the third step.

[0063] Step height needs to be adjusted according to terrain roughness; rough terrain requires a larger step height to avoid foot collisions with obstacles. The step height adjustment factor is defined as follows: (9) In the formula, The terrain roughness (calculated using formula 5) The maximum reference value for surface roughness is set to 0.15 meters. This is the step height adjustment coefficient. When the terrain smoothness and roughness are close to zero, the coefficient is approximately 1; when the roughness reaches its maximum value, the coefficient is 3, doubling the step height. The optimized step height is... .

[0064] The body attitude angle needs to be adjusted to compensate for the terrain slope, keeping the fuselage relatively horizontal to maintain center of gravity stability. The desired pitch angle is defined as: (10) In the formula, The slope angle of the terrain. This represents the desired pitch angle. A coefficient of 0.6 indicates partial slope compensation; full compensation would result in excessive leg joint angles exceeding the range of motion, hence the partial compensation strategy. The body maintains a slight forward lean when going uphill and a slight backward lean when going downhill.

[0065] Based on the above adjustments, the optimized gait parameter set is as follows: Among them, the gait period T and the duty cycle Keeping the baseline values ​​unchanged, the step size L, step height H, and desired pitch angle remain the same. These are the adjusted values. These optimized parameters are passed to the fifth step for stability evaluation to ensure that the adjusted gait still meets the stability requirements.

[0066] In step five, gait stability assessment is a crucial step in ensuring the robot's safe movement. The system employs a multi-index fusion method to comprehensively evaluate stability, mainly including three aspects: zero-moment point position assessment, support polygon margin assessment, and joint moment margin assessment.

[0067] The zero-moment point is an important indicator for evaluating dynamic stability; it is defined as the point where the net torque exerted by the robot on the ground is zero. When the zero-moment point is located inside the supporting polygon, the robot is in a stable state. To calculate the location of the zero-moment point, the acceleration of the robot's center of mass is first obtained. (Measured by the inertial measurement unit), position of the center of mass (Calculated using forward kinematics), the total mass of the robot is m. The coordinates of the zero-torque point on the horizontal plane are: (11) (12) In the formula, , Let be the horizontal coordinate of the zero-moment point, and g be the gravitational acceleration, taken as 9.8 m / s². Calculate the minimum distance from the zero-moment point to the boundary of the supporting polygon, defined as the stability margin: (13) In the formula, Let be the perpendicular distance from the zero-moment point to the j-th boundary. This represents the minimum distance, i.e., the stability margin. When The robot is stable when the zero torque point is inside the supporting polygon. The zero-moment point exceeds the supporting polygon robot's instability. A safety threshold is set. meters, when If the stability is insufficient, the gait needs to be adjusted.

[0068] The support polygon margin assesses stability geometrically, characterized by calculating the distance from the centroid projection point to the support polygon boundary. The centroid's projection coordinates on the horizontal plane are... Calculate the distance from the point to each side of the supporting polygon, and take the minimum value as the support margin. The calculation method for this indicator is similar to that of Formula 13, the difference being that the evaluation object changes from the zero moment point to the centroid projection point. A support margin threshold is set. rice.

[0069] Joint torque margin assessment determines whether each joint is close to torque saturation. Calculate the current torque for each joint. (Measured via inverse dynamics or a torque sensor) and the maximum permissible torque The ratio of the joint torque margin is defined as: (14) In the formula, Let be the torque margin of the j-th joint. When When the torque is close to zero, it indicates that the joint torque is close to saturation. The minimum margin among all joints is taken. As an overall joint margin indicator, a joint margin threshold is set. That is, when the joint torque exceeds 80% of the maximum torque, the margin is considered insufficient.

[0070] Based on the three stability indicators, the overall stability score is defined as follows: (15) In the formula, The weighting coefficients are determined based on the instability risk characteristics of different terrains. On flat ground, the most important weight for the zero-moment point index is set to [value missing]. =0.5、 =0.3、 =0.2, the weight of the support polygon margin on the slope is adjusted to be more important. =0.3、 =0.5、 =0.2. When the overall stability score If the current gait is deemed unstable, the sixth step of the gait adjustment decision is triggered. The stability assessment results and the values ​​of each sub-indicator are transmitted to the decision module to determine the adjustment strategy.

[0071] In step six, when the stability assessment in step five detects insufficient stability, the gait adjustment decision module formulates an adjustment strategy based on the specific instability mode. The system establishes a rule-based decision tree and takes targeted measures based on the exceeding of limits for different stability indicators.

[0072] When zero torque margin Less than the threshold but with support margin Normally, this indicates insufficient dynamic stability but acceptable geometric stability. In this case, a strategy of reducing movement speed is adopted. By extending the gait period T to reduce the acceleration of the center of mass, the adjusted gait period is: (16) In the formula, T is the current gait cycle. To adjust the cycle, and Defined by Equation 13. The smaller the zero-moment margin, the greater the increase in period, up to a maximum of 30%. The increased period leads to a decrease in motion speed, but the decrease in the acceleration of the center of mass causes the zero-moment point to move towards the center of the supporting polygon, improving stability.

[0073] When support margin When the value is below the threshold, it indicates that the center of gravity is too close to the support boundary, requiring adjustment of body posture to expand the support area. A strategy of increasing the support span is adopted, increasing the distance between the left and right legs by abducting the hip joint. The support span adjustment amount is defined as: (17) In the formula, The increase in width is measured in meters. and This was obtained from a stability assessment. The more severe the support margin deficiency, the greater the increase in support width, up to a maximum increase of 8 cm. Simultaneously, the body height is reduced by flexing all leg joints to lower the center of gravity, by an amount that... By reducing the height of the centroid, the area of ​​the supporting polygon can be increased, thereby improving stability.

[0074] When joint torque margin When the values ​​are below the threshold, it indicates that the torque of some joints is close to saturation, requiring a reduction in joint load. A strategy of reducing stride length and stride height is adopted to decrease the dynamic load on the swing leg movement. The adjusted stride length and stride height are as follows: (18) In the formula, L and H are the gait parameters optimized in the fourth step. and These are the adjusted parameters. Reducing both step length and step height by 20% simultaneously lowers joint angular velocity and angular acceleration, thus reducing the required joint torque.

[0075] When multiple stability indicators simultaneously exceed their limits, it indicates that the robot is in a state of severe instability, and an emergency stop strategy should be implemented. Immediately halt the current gait, quickly land all swinging legs to establish quadrupedal support, adjust the robot's posture to a crouching position to lower the center of gravity, and wait for stability to recover before replanning the gait. The emergency stop is triggered by the following conditions: (19) In the formula, The comprehensive stability score is defined by Formula 15. When the score is below 0.5, it is considered a dangerous state and an emergency stop is triggered.

[0076] Adjust the decision module to output new gait parameters and body posture target These instructions are then passed to the underlying control module in step seven for execution. Simultaneously, the adjustment results are fed back to the stability assessment module, which re-evaluates the adjustment effect in the next control cycle, thus forming a closed-loop regulation.

[0077] In step seven, the underlying execution control module receives the gait parameters and posture target output from step six, generates the desired motion trajectory of each leg foot, calculates the joint angle command through inverse kinematics, and finally realizes trajectory tracking through joint servo control.

[0078] For the leg in the swing phase, the foot trajectory is planned using cubic spline curves. The starting point of the swing phase is defined as... The destination is The horizontal displacement is determined by the step size L, and the vertical displacement is determined by the step height H. The vertical component of the oscillating phase trajectory takes the form of a parabola. (20) In the formula, t is the time within the oscillation phase. Let z(t) be the duration of the swing phase, z(t) be the foot height at time t, and H be the step height parameter. The trajectory reaches its maximum height H at the midpoint of the swing phase, and the starting and ending heights are... A smooth parabola is formed. The horizontal component moves at a constant velocity. (twenty one) In the formula, The foot is at a horizontal position, and L is the step length. The lateral component y(t) remains constant or is adjusted according to steering requirements.

[0079] For the leg in the support phase, the foot is stationary relative to the ground, but due to the movement of the fuselage, the foot exhibits relative motion in the body coordinate system. The trajectory of the supporting leg's foot in the body coordinate system is calculated by the changes in fuselage speed and attitude, maintaining contact between the foot and the ground.

[0080] Based on the planned foot trajectory, the desired angles of each joint are calculated using inverse kinematics. For each leg, a kinematic chain is established from the fuselage to the foot, including three degrees of freedom: pitch and lateral swing at the hip joint and pitch at the knee joint. The inverse kinematics are solved analytically, based on the foot position. Calculate the three joint angles The specific solution of inverse kinematics is based on geometric relationships, and the joint angles are calculated using trigonometric functions.

[0081] After obtaining the desired joint angle, a virtual model control algorithm is used to achieve joint control. The virtual model treats the leg as an equivalent virtual spring-damped system, and the joint control torque is calculated as follows: (twenty two) In the formula, The control torque of the j-th joint, For the desired joint angle, From the current practical perspective, For the desired angular velocity, This is the actual angular velocity. For proportional gain, This is the differential gain. The gain parameter is determined based on the joint type and load characteristics; for the hip joint, it is taken as... =150 N·m / rad =3 N·m·s / rad, knee joint taken =100 N·m / rad =2 N·m·s / rad. This control law achieves compliant tracking of the desired trajectory while also possessing a certain degree of damping to absorb terrain impacts.

[0082] The joint controller outputs torque commands to the joint motor driver, which then executes torque control to move the joint along the desired trajectory. The actual joint angle and angular velocity are measured by an encoder and fed back to the controller, forming position control. Simultaneously, the support reaction force information measured by the foot force sensor is fed back to the first-step perception module, initiating a new cycle of terrain perception and gait adjustment. The entire system constitutes a complete perception-decision-execution process.

[0083] Example 4 is an embodiment of the present invention, which provides an adaptive gait control method for a quadruped robot in a limited space. To verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0084] To verify the effectiveness of this patented method, a typical confined space scenario was built on a simulation platform for testing. The simulation scenario simulated an underground utility tunnel environment, 50 meters long and 2 meters wide, including a 10-meter flat section, an 8-meter uphill section (15-degree slope), an 8-meter downhill section (12-degree slope), a stepped area (3 steps, 0.15 meters high), a 10-meter gravel road surface, and a 4-meter waterlogged area, among other terrain features. A robot model was built using quadruped robot dynamics simulation software, with a body weight of 30 kg and a leg length of 0.5 meters, employing the same algorithm parameters as the actual controller. The simulation control cycle was 1 millisecond, and the running time was 300 seconds, recording data such as the robot's motion trajectory, gait parameter changes, stability indicators, and energy consumption. Three control methods were compared: the traditional fixed gait method (using a single diagonal trot), the rule-based gait switching method (switching gait according to preset rules without optimizing parameters), and the adaptive gait control method of this patent. The experimental results are shown in Tables 1, 2, and 3. Table 1 Comparison of Success Rates for Different Terrain Sections

[0085] Table 2. Adaptive Adjustment Effect of Gait Parameters

[0086] Table 3 Comparison and Analysis of Stability Indicators

[0087] This patented method demonstrates a significantly higher success rate than traditional methods across various terrains, achieving an overall success rate of 97%, with only a few failures occurring on extreme gravel surfaces. The fixed gait method suffers from a high failure rate on steps and gravel surfaces due to frequent instability caused by mismatched gait parameters. While the rule-switching method can select a suitable gait, failures still occur due to unoptimized parameters.

[0088] In terrain transition areas, this method can quickly identify terrain changes and adjust gait parameters. The adjustment range of stride length and stride height conforms to the calculation results of Formulas 8 and 9, and the stability score is improved by an average of 0.25 after adjustment, ensuring that the robot smoothly transitions to new terrain.

[0089] This patented method demonstrates superior performance across all stability metrics, with an average stability score of 1.62 significantly higher than the safety threshold of 1.0, and a minimum score of 1.05 also within the safe range. The fixed gait method, on the other hand, achieved a minimum stability score of only 0.45 and experienced 23 instability events, while this method only exhibited 2 minor instabilities with rapid recovery.

[0090] Example 5 is an embodiment of the present invention. This embodiment provides an adaptive gait control system for a quadruped robot in a limited space, including: a multimodal terrain perception module, used to collect support reaction force information of each leg, body posture angle information and depth image information of the terrain in front, and extract contact force distribution features, posture change features and spatial structure features from the support reaction force information, posture angle information and depth image information respectively, and combine them into a terrain feature vector; The terrain classification and recognition module is used to input the terrain feature vector into the classifier to obtain the terrain category; The gait library and selection module is used to obtain the corresponding basic gait pattern from the gait library according to the terrain category, and to obtain the initial gait parameters in the basic gait pattern; The gait parameter optimization module is used to extract terrain slope value and terrain roughness value from the terrain feature vector, adjust the step parameter in the initial gait parameters according to the terrain slope value, and adjust the ground clearance parameter in the initial gait parameters according to the terrain roughness value to obtain optimized gait parameters; The stability assessment module is used to calculate stability indices based on the optimized gait parameters and fuse the stability indices to obtain a comprehensive stability score. The gait adjustment decision module is used to determine an adjustment strategy based on the value of the stability index when the comprehensive stability score is lower than the safety threshold, and to correct the optimized gait parameters according to the adjustment strategy. The underlying execution control module is used to generate foot trajectories for each leg based on the corrected optimized gait parameters, convert the foot trajectories into joint angle commands through inverse kinematics and execute the control, and feed back the support reaction force information and attitude angle information measured after the control is executed to the multimodal terrain perception module.

[0091] This embodiment also provides an electronic device applicable to an adaptive gait control method for a quadruped robot in a confined space, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the adaptive gait control method for a quadruped robot in a confined space as proposed in the above embodiment.

[0092] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements an adaptive gait control method for a quadruped robot in a confined space as proposed in the above embodiments.

[0093] The storage medium proposed in this embodiment and the method for implementing adaptive gait control of a quadruped robot in a limited space proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0094] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An adaptive gait control method for a quadruped robot in a confined space, characterized in that: include, Collect support reaction force information of each leg, fuselage attitude angle information, and depth image information of the terrain ahead; extract contact force distribution features, attitude change features, and spatial structure features from the support reaction force information, attitude angle information, and depth image information respectively, and combine them into a terrain feature vector; The terrain feature vector is input into a classifier to obtain the terrain category; the corresponding basic gait pattern is obtained from the gait library according to the terrain category, and the initial gait parameters in the basic gait pattern are obtained; The terrain slope value and terrain roughness value are extracted from the terrain feature vector. The step parameter in the initial gait parameters is adjusted according to the terrain slope value, and the ground clearance parameter in the initial gait parameters is adjusted according to the terrain roughness value to obtain the optimized gait parameters. Based on the optimized gait parameters, a stability index is calculated, and the stability index is fused to obtain a comprehensive stability score. When the overall stability score is lower than the safety threshold, an adjustment strategy is determined based on the value of the stability index, and the optimized gait parameters are corrected according to the adjustment strategy. The foot trajectory of each leg is generated based on the corrected optimized gait parameters, and the foot trajectory is converted into joint angle commands through inverse kinematics and then executed for control. The process of feeding back the support reaction force information and attitude angle information obtained after the execution of control to the extraction process of terrain feature vectors.

2. The adaptive gait control method for a quadruped robot in a confined space as described in claim 1, characterized in that: The extraction of contact force distribution features includes: Zero drift compensation is performed on the collected support reaction force information of each leg to obtain the compensated support reaction force; The compensated support reaction force is filtered to obtain the filtered support reaction force; Based on the filtered support reaction force, the uniformity of force distribution between each leg is calculated to obtain the contact force distribution characteristics.

3. The adaptive gait control method for a quadruped robot in a confined space as described in claim 2, characterized in that: Extracting pose change features, including: Obtain the fuselage attitude angles at the current and historical moments; Based on the fuselage attitude angles at the current and historical moments, calculate the time rate of change of the attitude angles; The rate of change over time is used as the characteristic of the attitude change.

4. The adaptive gait control method for a quadruped robot in a confined space as described in claim 3, characterized in that: Extracting spatial structural features, including: The depth image information is segmented into ground regions to obtain ground depth data; Geometric modeling is performed on the ground depth data to obtain ground geometric parameters; Based on the ground geometric parameters, the slope and undulation characteristics of the terrain are extracted as the spatial structure features.

5. The adaptive gait control method for a quadruped robot in a confined space as described in claim 4, characterized in that: The calculation of the stability index based on the optimized gait parameters includes: Obtain the position and acceleration of the fuselage's center of mass; Calculate the horizontal coordinates of the zero torque point based on the center of mass position, center of mass acceleration, and total mass of the robot. Calculate the vertical distance from the zero moment point to each boundary of the supporting polygon, and take the minimum distance value as the first stability index; Calculate the distance from the projection point of the centroid onto the horizontal plane to each boundary of the supporting polygon, and take the minimum distance value as the second stability index; Obtain the current torque and maximum allowable torque of each joint, calculate the ratio of the current torque to the maximum allowable torque of each joint, calculate the torque margin of each joint based on the ratio, and take the minimum torque margin as the third stability index.

6. The adaptive gait control method for a quadruped robot in a confined space as described in claim 5, characterized in that: The step of determining the adjustment strategy based on the value of the stability index includes: Determine whether the first stability index is less than the first threshold; If the first stability index is less than the first threshold, then a timing adjustment strategy is selected, and the gait cycle increment is calculated based on the difference between the first stability index and the first threshold, and the current gait cycle is increased by the gait cycle increment. Determine whether the second stability index is less than the second threshold; If the second stability index is less than the second threshold, then the configuration adjustment strategy is selected, and the support width increment is calculated based on the difference between the second stability index and the second threshold, thereby increasing the support span between the four legs. Determine whether the third stability index is less than the third threshold; If the third stability index is less than the third threshold, then a load adjustment strategy is selected to reduce the stepping parameter and the ground clearance parameter in the optimized gait parameters.

7. The adaptive gait control method for a quadruped robot in a confined space as described in claim 6, characterized in that: The step of generating the foot trajectory of each leg based on the corrected optimized gait parameters includes: The swing phase duration and support phase duration of each leg are determined based on the corrected optimized gait parameters. For the leg in the swing phase, the starting and ending positions of the foot are determined according to the stepping parameters in the optimized gait parameters, and the maximum ground clearance of the foot is determined according to the ground clearance parameters in the optimized gait parameters. A foot trajectory is generated during the swing phase. The vertical component of the foot trajectory is interpolated parabolically so that the foot reaches the maximum height off the ground at the midpoint of the swing phase. The horizontal component is interpolated at a constant speed from the starting position to the ending position.

8. An adaptive gait control system for a quadruped robot in a confined space, employing the adaptive gait control method for a quadruped robot in a confined space as described in any one of claims 1 to 7, characterized in that, include: The multimodal terrain perception module is used to collect support reaction force information of each leg, fuselage attitude angle information and depth image information of the terrain in front, and extract contact force distribution features, attitude change features and spatial structure features from the support reaction force information, attitude angle information and depth image information respectively, and combine them into a terrain feature vector. The terrain classification and recognition module is used to input the terrain feature vector into the classifier to obtain the terrain category; The gait library and selection module is used to obtain the corresponding basic gait pattern from the gait library according to the terrain category, and to obtain the initial gait parameters in the basic gait pattern; The gait parameter optimization module is used to extract terrain slope value and terrain roughness value from the terrain feature vector, adjust the step parameter in the initial gait parameters according to the terrain slope value, and adjust the ground clearance parameter in the initial gait parameters according to the terrain roughness value to obtain optimized gait parameters; The stability assessment module is used to calculate stability indices based on the optimized gait parameters and to fuse the stability indices to obtain a comprehensive stability score. The gait adjustment decision module is used to determine an adjustment strategy based on the value of the stability index when the comprehensive stability score is lower than the safety threshold, and to correct the optimized gait parameters according to the adjustment strategy. The underlying execution control module is used to generate foot trajectories for each leg based on the corrected optimized gait parameters, convert the foot trajectories into joint angle commands through inverse kinematics and execute the control, and feed back the support reaction force information and attitude angle information measured after the control is executed to the multimodal terrain perception module.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the adaptive gait control method for a quadruped robot in a confined space according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the adaptive gait control method for a quadruped robot in a confined space according to any one of claims 1 to 7.

Images