Robot gait self-adaptive adjustment method and system combined with topographic feature recognition

Abstract

The invention discloses a robot gait self-adaptive adjustment method and system combined with topographic feature recognition, and relates to the technical field of robot gaits, and the method comprises the steps: collecting a topographic image in front of a robot advancing path, carrying out the deblurring processing, and extracting ground feature data through a lightweight neural network; matching the ground characteristic data with a preset gait control parameter database to obtain a gait control parameter set, and generating a control signal through a proportional-differential control method according to the gait control parameter set; correcting the control signal based on a reinforcement learning algorithm, and generating a target control instruction for driving a joint actuator; in the process of executing the target control instruction, collecting state feedback data in real time, and comparing and optimizing the feedback data with the currently executed target control instruction to realize feedback adjustment; the problems of insufficient terrain recognition precision caused by image blurring, response lag caused by sensing and control decoupling and lack of adaptive ability are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of robot gait technology, and more specifically, to a robot gait adaptive adjustment method and system that incorporates terrain feature recognition. Background Technology

[0002] As the application of robotics, especially mobile robots and humanoid robots, continues to expand in complex environments, the requirements for robots' environmental adaptability are increasing. In complex, unstructured terrains, such as gravel, slopes, and mud, traditional fixed gait planning methods, lacking environmental perception and adaptation mechanisms, often fail to guarantee gait stability and movement efficiency, easily leading to problems such as falls, slips, or excessive foot impact, severely limiting the practicality and reliability of robots.

[0003] Several methods attempt to improve adaptability by acquiring ground contact feedback through pre-installed sensors or by adjusting gait control parameters using rule bases. However, these methods generally suffer from the following technical limitations:

[0004] In real-world environments, imaging problems such as motion blur and jitter blur are common, leading to a decrease in the quality of image feature extraction, which in turn affects the accuracy and robustness of terrain recognition.

[0005] Furthermore, existing control systems typically decouple sensing from control, failing to form a closed-loop feedback control system driven by ground features. This results in insufficient targeting of control strategies, low response frequency, poor adaptability to sudden terrain changes, and difficulty in achieving efficient regulation.

[0006] To address the above problems, this invention proposes a solution. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a robot gait adaptive adjustment method and system that combines terrain feature recognition. By integrating image deblurring processing, lightweight neural network terrain feature extraction, reinforcement learning-driven control correction, and fuzzy logic fusion regulation mechanism, a closed-loop gait adjustment system integrating perception, control, and feedback is constructed, effectively solving the problems of insufficient terrain recognition accuracy caused by image blurring, response lag caused by decoupling of perception and control, and lack of adaptive capability.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] In a first aspect, this application provides a robot gait adaptive adjustment method combining terrain feature recognition. The method includes: acquiring a terrain image ahead of the robot's path and performing deblurring processing; extracting ground characteristic data from the blurred image using a lightweight neural network; matching the ground characteristic data with a preset gait control parameter database to obtain a gait control parameter set; generating a control signal based on the gait control parameter set using a proportional-derivative control method; correcting the control signal based on a reinforcement learning algorithm and generating a target control command to drive the joint actuators; and during the execution of the target control command, acquiring real-time state feedback data and comparing and optimizing the feedback data with the currently executed target control command to achieve feedback adjustment.

[0010] In one embodiment, a terrain image ahead of the robot's path is acquired and deblurred. Specifically, each frame of the terrain image is converted into a single-channel grayscale image, and a sharpness index is calculated. If the sharpness index is lower than a preset blur threshold, the current image is determined to be a blurry image, and deblurring is performed. Edge detection is performed on the blurry image to obtain an edge intensity map and a direction map. The direction map is divided into several direction intervals, and the number of pixels in each direction is counted to construct an edge direction distribution histogram. If the proportion of pixels in the main direction exceeds a preset direction concentration threshold, it is determined to be linear motion blur; otherwise, it is nonlinear jitter blur. Deblurring is performed on both linear motion blur and nonlinear jitter blur.

[0011] In one embodiment, deblurring is performed on linear motion blur and nonlinear jitter blur, specifically as follows: A two-dimensional Fourier transform is performed on the blurred image based on linear motion blur to obtain the blur direction and spacing, and the blur length is inferred from the spacing; a linear motion blur kernel is constructed as a linear blur point spread function based on the blur direction and blur length; a non-blind deconvolution algorithm is used to restore the blurred image based on the point spread function to obtain the blurred image; based on nonlinear jitter blur, the blurred image is set as the observed image, and an estimated clear image and a nonlinear blur kernel are initialized, and an optimization objective function is constructed; based on the optimization objective function, an alternating minimization strategy is used to optimize the image and blur kernel until a preset number of iterations is reached to obtain the optimized nonlinear blur kernel; the nonlinear blur kernel is used as the blur kernel for non-blind deconvolution to perform deblurring processing to obtain the clear image.

[0012] In one embodiment, ground feature data is extracted from the blurred image using a lightweight neural network. Specifically, the blurred image is preprocessed to obtain a standardized image; the standardized image is used as input to construct a lightweight convolutional neural network model for preliminary ground feature recognition, outputting a ground feature classification map; the confidence level in the ground feature classification map is obtained, and based on a preset confidence threshold, low-confidence image regions in the ground feature classification map are extracted to construct corresponding candidate region mask maps; based on the candidate region mask maps, corresponding high-resolution image sub-blocks are extracted from the terrain image, and these sub-blocks are used as input to a preset refined recognition model; the output of the refined recognition model is fused with the ground feature classification map to generate a ground feature semantic segmentation map, which is then converted into structured ground feature data.

[0013] In one embodiment, ground characteristic data is matched with a preset gait control parameter database to obtain a gait control parameter set. A control signal is then generated based on the gait control parameter set using a proportional-derivative (PDA) control method. Specifically, the ground characteristic data is linearly normalized to obtain a ground feature vector; the Euclidean distance between the current ground feature vector and the ground feature vector in the database is calculated using the K-nearest neighbor algorithm, and the gait control parameter set corresponding to the minimum distance is selected as the output result; the current state information of the robot is obtained, and a control correction term is calculated based on the error between the matched control parameter set and the state information using the PDA control method; finally, a control signal is generated based on the control correction term.

[0014] In one embodiment, the control signal is modified based on a reinforcement learning algorithm to generate a target control command to drive the joint actuator. Specifically, this involves: constructing a reinforcement learning model, taking ground feature vectors, state information, and control signals as input states; using adjustment factors of the control signals as output actions; constructing a reward function, which includes vibration spectrum indices, gait symmetry metrics, and contact surface stability indices; aiming to maximize the reward function, outputting adjustment factors of the control signals based on the current input states through the reinforcement learning model; and fusing the adjustment factors with the control signals using a fuzzy logic reasoning method to generate a target control command to drive the joint actuator.

[0015] In one embodiment, a target control command is generated by fusing adjustment factors and control signals based on a fuzzy logic reasoning method. Specifically, the adjustment factors of the control signals output by the reinforcement learning model are used as input variables of the fuzzy controller to construct corresponding membership functions, mapping the continuous values ​​of each adjustment factor to fuzzy states. Based on the fuzzy states, fuzzy logic reasoning is performed using a preset empirical rule base to generate corresponding fuzzy control outputs. The centroid method is used to convert the fuzzy control outputs into numerical correction quantities. The numerical correction quantities are fused with the control signals to generate target control commands, which include target joint angle values, target joint angular velocity values, target joint torque values, and target joint damping values. The target control commands are physically constrained and verified, and the verified target control commands are output to drive the joint actuators.

[0016] In one embodiment, during the execution of the target control command, state feedback data is collected in real time, and the feedback data is compared and optimized with the currently executed target control command to achieve feedback adjustment. Specifically, the following steps are taken: State feedback data is acquired during the execution of the target control command by the joint actuator. The state feedback data includes actual joint angle values, actual joint angular velocity values, actual joint torque values, foot-ground contact force values, and acceleration values. The state feedback data and the target control command are numerically compared, and errors are calculated accordingly. Based on the errors and the state feedback data, quantitative feedback evaluation indicators are constructed, including trajectory tracking error indicators, gait stability indicators, and contact impact indicators. The quantitative feedback evaluation indicators are combined into a multi-dimensional state vector, which is then transmitted as input data to the reinforcement learning model to obtain the optimized adjustment factor for the next control cycle. The optimized adjustment factor is used to generate the target control command for the next control cycle, thus achieving closed-loop feedback adjustment.

[0017] Secondly, this application provides a robot gait adaptive adjustment method and system that combines terrain feature recognition, the system comprising:

[0018] The image recognition module is used to acquire terrain images ahead of the robot's path and perform deblurring processing. It then uses a lightweight neural network to extract ground feature data from the blurred images.

[0019] The control signal generation module is used to match ground characteristic data with a preset gait control parameter database to obtain a gait control parameter set, and generate a control signal based on the gait control parameter set using a proportional-derivative control method.

[0020] The control signal correction module is used to correct the control signal based on a reinforcement learning algorithm and generate target control commands to drive the joint actuator.

[0021] The real-time feedback adjustment module is used to collect status feedback data in real time during the execution of the target control command, and compare and optimize the feedback data with the currently executed target control command to achieve feedback adjustment.

[0022] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0023] 1. By introducing fuzzy detection and classification mechanisms into the acquisition of images along the robot's forward path, combined with differentiated defuzzification strategies for linear and nonlinear fuzzy types, image clarity and structural discernibility are significantly improved, resolving the impact of image blurring on terrain recognition accuracy during dynamic movement. Simultaneously, a lightweight convolutional neural network is used for preliminary ground feature extraction, and confidence analysis and region masking strategies guide a high-precision recognition model to perform fine-grained perception of low-confidence areas, achieving high-precision recognition of key terrain areas under computationally limited conditions. Through multi-scale feature fusion and spatial pyramid pooling, the semantic modeling capabilities are enhanced, ultimately outputting structured ground feature data. This provides high-quality, quantifiable environmental perception support for subsequent path planning and gait adaptive control, significantly improving the robot's adaptability and stability in complex, unstructured terrain.

[0024] 2. By constructing a mapping database between ground features and gait control parameters, and combining K-nearest neighbor matching and proportional-derivative control to generate basic control signals, a reinforcement learning algorithm is introduced to achieve adaptive optimization based on environmental feedback. Furthermore, fuzzy logic reasoning is integrated to enhance the fine-tuning capability of the control signals. Simultaneously, a closed-loop feedback mechanism is constructed to correct control parameters in real time, thus forming an integrated gait adjustment system of "perception-matching-control-optimization-feedback". This scheme not only improves the robot's stability and environmental adaptability in complex terrains, but also significantly enhances the robustness, coordination, and energy efficiency of the control system through multi-dimensional reward function evaluation and dynamic policy updates, demonstrating strong generalization ability and engineering practical value. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the robot gait adaptive adjustment method that combines terrain feature recognition, provided in an embodiment of this application.

[0026] Figure 2 This is a schematic diagram of the robot gait adaptive adjustment system that combines terrain feature recognition, provided in an embodiment of this application.

[0027] Figure 3 The edge direction distribution histogram provided for the embodiments of this application. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0029] Reference Figure 1 As shown in the diagram, the robot gait adaptive adjustment method combining terrain feature recognition provided by the present invention includes the following steps:

[0030] S1: Collect terrain images ahead of the robot's path and perform deblurring processing. Extract ground characteristic data from the blurred images using a lightweight neural network.

[0031] The system collects terrain images of the path in front of the mobile platform by means of visual sensors installed at the front end of the mobile platform. These visual sensors include RGB cameras, LiDAR, depth cameras, or binocular stereo vision modules. The acquisition frequency (e.g., 5~15 Hz) is dynamically adjusted according to the robot's travel speed.

[0032] In this embodiment, the terrain image ahead of the robot's path is acquired and deblurred, specifically as follows:

[0033] Each frame of the terrain image is converted into a single-channel grayscale image, and the sharpness index is calculated based on the image gradient variance method of the Laplacian operator.

[0034] The specific formula for calculating the clarity index is as follows:

[0035]

[0036] In the formula, For clarity indicators, Let i be the value of the i-th pixel in the Laplacian response image. denoted as the average pixel value, and N as the total number of pixels.

[0037] If the sharpness index is lower than the preset blur threshold, the current image is determined to be a blurry image and deblurring is performed; otherwise, it is directly sent to the subsequent neural network for ground feature extraction.

[0038] Edge detection processing is performed on the blurred image. The edge detection uses the Sobel operator to calculate the gradient in the horizontal and vertical directions to obtain the edge intensity map and the direction map.

[0039] like Figure 3As shown, the orientation map is divided into several orientation intervals, the number of pixels in each orientation is counted, and an edge orientation distribution histogram is constructed.

[0040] If the proportion of pixels in the main direction exceeds the preset direction concentration threshold, the image is determined to have a clear main direction and is confirmed as linear motion blur; otherwise, it is non-linear jitter blur.

[0041] Perform deblurring operations on linear motion blur and nonlinear jitter blur.

[0042] Linear motion blur refers to the uniform stretching of pixels in an image along a fixed direction during exposure, resulting in a trailing effect. Nonlinear jitter blur, on the other hand, refers to image blur caused by high-frequency, irregular, minute vibrations of the image acquisition device during exposure. Its motion trajectory lacks clear directionality and exhibits nonlinear and non-uniform characteristics.

[0043] Furthermore, deblurring operations are performed on linear motion blur and nonlinear jitter blur, specifically as follows:

[0044] A two-dimensional Fourier transform is performed on the blurred image based on linear motion blur to obtain the blur direction and the spacing between the periodic grooves in the spectrum, and the blur length is inferred from the spacing. The blur direction is the main direction angle.

[0045] A linear motion fuzzy kernel is constructed based on the fuzzy direction and fuzzy length as a linear fuzzy point diffusion function;

[0046] Based on the point spread function, a non-blind deconvolution algorithm is used to restore the blurred image to obtain the blurred image.

[0047] Based on nonlinear jitter blur, the blurred image is set as the observed image, the estimated sharp image and the nonlinear blur kernel are initialized, and an optimization objective function is constructed;

[0048] Based on the objective function, an alternating minimization strategy is used to optimize the image and the blur kernel until a preset number of iterations is reached to obtain the final optimized nonlinear blur kernel. The optimization includes fixing the current blur kernel and optimizing the estimated image so that it approximates the observed image while preserving the edges, and fixing the current image estimation result and optimizing the blur kernel so that it more accurately characterizes the blur characteristics in the current image.

[0049] The nonlinear blur kernel is used as the blur kernel for non-blind deconvolution to deblur the blurred image and obtain a clear image.

[0050] The specific calculation formula for the optimization objective function is as follows:

[0051]

[0052] In the formula, To optimize the objective function, To observe the image, To estimate a clear image, It is a nonlinear fuzzy kernel. For image gradient, For regularization parameters, It is the square of the L2 norm. It is a norm.

[0053] The point spread function (PSF) is a function that describes how a point light source is diffused into a certain shape and pattern during the imaging process in an image acquisition system. It is used to characterize the blurring characteristics of the system and is the core parameter for image blurring modeling.

[0054] It should be noted that deblurring effectively identifies and corrects image blurring issues generated by the acquisition robot during dynamic movement. Through blur detection and classification, and matching different deblurring operations for refined processing, the clarity and structural discernibility of the terrain images are significantly improved. This lays a high-quality input foundation for subsequent ground feature extraction, enhancing the overall recognition accuracy and robustness of the system.

[0055] Furthermore, ground feature data is extracted from the blurred image using a lightweight neural network, specifically as follows:

[0056] The blurred image is preprocessed to obtain a standardized image. The preprocessing includes image size normalization, contrast enhancement, histogram equalization, and region cropping to improve the accuracy of feature extraction by the subsequent neural network.

[0057] A lightweight convolutional neural network model for preliminary ground feature identification is constructed based on standardized images as input, outputting a ground feature classification map. The model employs an encoder-decoder structure, comprising an input preprocessing layer, a lightweight feature extraction backbone network, a feature enhancement module, a decoding module, and a classification output layer. The feature extraction backbone network is a lightweight convolutional neural network, preferably using a lightweight CNN structure based on MobileNetV3 or ShuffleNetV2. Each basic convolutional block in the network contains the following structural units: depthwise separable convolution, batch normalization, Swish activation function, channel attention mechanism, and residual connections. Within this backbone network structure... Semantic information at different scales is extracted from feature maps at different depths and fused with the decoding path through skip connections in the subsequent decoding stage to improve edge structure and spatial positioning accuracy. The feature enhancement module preferably adopts a spatial pyramid pooling structure to supplement the global context perception capability of large-scale terrain structures. The decoding module performs progressive upsampling and recovery of low-resolution semantic feature maps, combines them with shallow features from the encoding stage to improve the quality of detail recovery, and performs fusion processing through a lightweight convolution module to output a prediction map with the same size as the original image. The output layer is a pixel-by-pixel ground feature classification map, which uses the Softmax activation function to output multi-class probabilities for subsequent cascaded recognition.

[0058] Obtain the confidence level in the ground feature classification map, and based on the preset confidence level threshold, extract the image regions with low confidence levels in the ground feature classification map to construct the corresponding candidate region mask map;

[0059] In the lightweight neural network, the final classification layer usually uses the Softmax activation function to transform the prediction result of each pixel into a probability distribution of each category. The maximum category probability value of each pixel is taken as the confidence of that pixel. Pixels with confidence below a preset threshold are marked as low-confidence image regions.

[0060] Based on the candidate region mask map, the corresponding high-resolution image sub-blocks are extracted from the terrain image, and the sub-blocks are used as input to the preset refined recognition model;

[0061] The refined model is a medium-depth high-precision neural network used to further identify ground feature boundaries and fine-grained categories in a local area. It preferably adopts a pruned and optimized EfficientNet, ResNet or Swin-Transformer structure to control computational complexity while ensuring accuracy.

[0062] The high-precision local recognition results output by the refined recognition model are fused with the ground feature classification map to generate a ground feature semantic segmentation map that balances global consistency and local accuracy.

[0063] The fusion process employs a region-level update strategy: the results of a refined recognition model are used to overwrite the original results for the regions specified in the mask image, while the predictions of the lightweight convolutional neural network model remain unchanged for the non-masked regions.

[0064] The semantic segmentation map is converted into structured ground feature data.

[0065] Ground characteristic data refers to a set of parameterized data extracted from visual images through neural network model analysis, which can characterize the physical morphology of the ground, the state of the surface structure, and its mechanical impact on moving objects. It is mainly used to support robots and other moving objects to achieve path planning, gait adaptive control, and stability optimization in natural or semi-structured environments. The ground characteristic data includes local slope, surface roughness, surface type, and obstacle boundary distance.

[0066] The candidate region mask is a binary map with the same size as the original image. It is used to indicate pixel regions with low confidence in the ground feature classification map output by the lightweight neural network. Its function is to identify regions with high prediction uncertainty of the lightweight model and use them as candidate regions for subsequent high-precision model to perform fine recognition, so as to achieve local optimization of resource utilization.

[0067] High-resolution image patches refer to local image patches extracted from the original input image that correspond to low-confidence regions in the candidate region mask, retaining their original resolution. These patches serve as input to a high-precision recognition model, used to further extract higher-order feature information such as feature boundaries and fine-grained categories within a local area.

[0068] It should be noted that by introducing a two-stage architecture of lightweight convolutional neural networks and regional-level refined recognition models, high-precision perception of key areas is achieved while ensuring overall inference efficiency, overcoming the shortcomings of insufficient accuracy or excessive computational burden of traditional single models in resource-constrained environments. At the same time, multi-scale feature fusion and spatial pyramid pooling mechanisms are adopted to enhance the semantic modeling capability of complex ground structures, improving the recognition accuracy of edge contours and fine-grained terrain differences. The final output structured ground characteristic data has higher semantic interpretability and parameterized expression capability, which can directly support the path planning and dynamic control tasks of the robot system, significantly enhancing its environmental adaptability and stability in natural scenes.

[0069] S2, the ground characteristic data is matched with the preset gait control parameter database to obtain the gait control parameter set, and the control signal is generated according to the gait control parameter set through the proportional-derivative control method.

[0070] The gait control parameter database is a pre-constructed data structure mapping multiple ground characteristics to corresponding gait control parameter sets. It supports the rapid search of the most suitable gait control strategy set for the current terrain based on ground features during actual walking. Its core function is to establish a mapping relationship from ground feature space to gait control strategy space, enabling the robot to dynamically adjust its gait control mode according to ground changes, thereby improving stability and environmental adaptability. The gait control parameter database is constructed by selecting representative ground attribute parameters as components of ground feature vectors, based on the robot's target application scenario (such as outdoors, factories, and residences). Common parameters include: slope, friction, hardness, and unevenness. Multiple sets of ground feature vectors are obtained by collecting sensor data from multiple different ground types in a simulation environment. For each set of ground feature vectors, parameters are manually designed and fine-tuned based on expert experience and existing motion control data to generate the corresponding gait control parameter set.

[0071] In this embodiment, ground characteristic data is matched with a preset gait control parameter database to obtain a gait control parameter set, and a control signal is generated based on the gait control parameter set using a proportional-derivative control method, specifically:

[0072] The ground characteristic data is linearly normalized to obtain the ground feature vector that represents the ground characteristics;

[0073] The Euclidean distance between the current ground feature vector and the ground feature vector in the database is calculated based on the K-nearest neighbor algorithm, and the gait control parameter set corresponding to the minimum distance is selected as the output result. The gait control parameter set includes stride length, stride frequency, leg lift height, angular trajectory array of each joint, and expected angular velocity array, expected torque value and expected damping value of each joint.

[0074] Obtain the current state information of the robot, including the current position angle and angular velocity of each joint, the posture angle of the robot body, and the position of the center of gravity;

[0075] Based on the proportional-derivative control method, a control correction term is calculated according to the error between the matched control parameter set and the state information. Then, a control signal for controlling each joint actuator is generated based on the control correction term. The control signal includes the target angle, target angular velocity, target torque, and target damping.

[0076] S3, the control signal is corrected based on the reinforcement learning algorithm, and the target control command for driving the joint actuator is generated.

[0077] Specifically, by modifying the control signal through reinforcement learning, adaptive optimization based on environmental feedback and historical experience can be achieved. Reinforcement learning constructs a reward function oriented towards actual motion performance, continuously adjusting joint control parameters based on multi-dimensional sensor inputs. This makes the control strategy more robust and generalizable in the face of different terrain disturbances, load changes, or nonlinear interferences. Simultaneously, this modification mechanism can significantly improve the robot's walking stability, motion coordination, and energy efficiency, thereby enhancing the overall intelligence and adaptability of the gait control system.

[0078] In this embodiment, the control signal is modified based on a reinforcement learning algorithm, and a target control command for driving the joint actuator is generated, specifically as follows:

[0079] Construct a reinforcement learning model, taking ground feature vectors, state information, and control signals as input states;

[0080] The target angle, target angular velocity, target torque, and target damping adjustment factors in the control signal are used as output actions;

[0081] Construct a reward function, which includes a vibration spectrum index, a gait symmetry measure, and a contact surface stability index;

[0082] With the goal of maximizing the reward function, an adjustment factor for the control signal output by the reinforcement learning model is generated based on the current input state.

[0083] The adjustment factor and control signal are fused using a fuzzy logic reasoning method to generate target control commands to drive the joint actuator.

[0084] The reward function is calculated using the following formula:

[0085]

[0086] In the formula, For the reward function, As a measure of gait symmetry, For the stability index of the contact surface, For vibration spectrum indicators, , , These are the weighting coefficients.

[0087] The vibration spectrum index refers to the spectral entropy of the robot joint vibration signal. The higher the entropy value, the more severe the control jitter. An angle / acceleration data of the controlled joint is selected using a high-frequency IMU or a high-precision encoder, and sampled within a preset time window to obtain a discrete sequence. A Fast Fourier Transform (FFT) is performed on the discrete sequence to obtain the amplitude spectrum for frequency distribution. The energy of each frequency point on the frequency distribution is normalized to form a probability distribution, and the Shannon entropy formula is applied to the probability distribution to obtain the spectral entropy, thus yielding the vibration spectrum index.

[0088] Gait symmetry refers to the relative deviation of key gait parameters (such as stride length, ground contact duration, and swing period) between the left and right legs. It is calculated by obtaining the key gait parameters of the left and right legs, dividing the absolute value of the difference between these parameters by the sum of the key gait parameters for both legs, and then summing these values ​​to construct the symmetry loss function. and with 1- / 3 provides a measure of gait symmetry; the closer to 1, the more symmetrical the gait.

[0089] The contact surface stability index refers to the rate of change of the actual contact area between the robot's foot and the ground (e.g., whether the forefoot or heel is raised). By deploying m×n pressure sensors on the sole of the foot, a pressure matrix is ​​collected between two points. The set of contact points in the pressure matrix that exceed a preset pressure threshold is obtained. The rate of change of the contact area (i.e., the maximum and minimum contact areas divided by the maximum area) is calculated by taking the maximum and minimum contact areas during the support period. Subtracting the rate of change of the contact area from 1 yields the contact surface stability index; a higher index indicates more stable contact.

[0090] It is worth noting that by introducing vibration spectrum indices, gait symmetry measures, and contact surface stability indices to construct the reward function in reinforcement learning, a multi-dimensional and refined evaluation of the robot's gait control performance can be achieved. This not only captures overall stability and energy consumption but also identifies fine-grained dynamic performance. Vibration spectrum indices reflect structural resonance and perturbation response, gait symmetry measures reveal motion coordination and execution consistency, and contact surface stability reflects the reliability of the interaction between foot contact forces and the ground. Combining these three indices can effectively avoid problems such as local instability, high energy consumption, and gait imbalance, thereby improving the convergence speed, generalization ability, and execution stability of reinforcement learning strategies, and significantly enhancing the robot's adaptability to complex terrain and task changes.

[0091] Furthermore, based on fuzzy logic reasoning, the adjustment factor and control signal are fused to generate target control commands to drive the joint actuator, specifically:

[0092] The adjustment factor of the control signal output by the reinforcement learning model is used as the input variable of the fuzzy controller to construct the corresponding membership function. The continuous values ​​of each adjustment factor are mapped to fuzzy states, such as "low", "medium" and "high". The membership function adopts a Gaussian shape function.

[0093] Based on the fuzzy state, fuzzy logic reasoning is performed using a preset empirical rule base to generate corresponding fuzzy control outputs. The rule base contains the relationship between adjustment factor states and corresponding control signal adjustment suggestions. Example rules include: "If the target angle adjustment factor is high and the target joint torque adjustment factor is medium, then the target angle should be increased appropriately and the torque should be slightly adjusted." This rule directly guides the fuzzy adjustment direction and magnitude of the control signal.

[0094] The centroid method is used to convert the fuzzy control output into a numerical correction quantity;

[0095] The numerical correction value is fused with the control signal to generate a target control command. The fusion method includes weighted superposition or multiplicative modulation. The target control command includes target joint angle value, target joint angular velocity value, target joint torque value, and target joint damping value.

[0096] Physical constraint verification is performed on the target control commands to ensure that they comply with the safety range and dynamic limits of the actuator;

[0097] The verified target control command is output to drive the joint actuator.

[0098] It should be noted that the physical constraint verification includes limiting the maximum range and rate of change of angle, angular velocity, torque and damping to prevent sudden changes in control signals from causing system instability.

[0099] S4. During the execution of the target control command, real-time status feedback data is collected, and the feedback data is compared and optimized with the currently executed target control command to achieve feedback adjustment.

[0100] In this embodiment, during the execution of the target control command, status feedback data is collected in real time, and the feedback data is compared and optimized with the currently executed target control command to achieve feedback adjustment. Specifically:

[0101] Acquire state feedback data during the execution of target control commands by the joint actuator. The state feedback data includes actual joint angle value, actual joint angular velocity value, actual joint torque value, foot-ground contact force value, and acceleration value.

[0102] The error is calculated by comparing the status feedback data with the target joint angle, target joint angular velocity, and target joint torque values ​​set in the target control command.

[0103] Based on the error and state feedback data, a quantitative feedback evaluation index is constructed, which includes a trajectory tracking error index, a gait stability index, and a contact impact index.

[0104] Among them, the trajectory tracking error index is the sum of the squares of the errors of each joint angle value, the gait stability index is calculated from the deviation between the trunk acceleration modulus and the set expected acceleration range, and is used to measure the stability of the robot during walking, and the contact impact index is the peak value of the vertical acceleration signal when the foot first contacts the ground, which is directly obtained by the foot acceleration sensor.

[0105] The quantitative feedback evaluation indicators are combined into a multidimensional state vector, which is then transmitted as input data to the reinforcement learning model to obtain the optimization adjustment factor for the next control cycle.

[0106] The target control command for the next control cycle is generated by using an optimized adjustment factor, so that its control parameters are closer to the current system execution state. Specifically, the target joint angle, target joint angular velocity, and target joint torque value are corrected with the optimized adjustment factor to achieve closed-loop feedback regulation.

[0107] Reference Figure 2 As shown in the diagram, the robot gait adaptive adjustment system combining terrain feature recognition provided by this invention includes an image recognition module, a control signal generation module, a control signal correction module, and a real-time feedback adjustment module, with connections between the modules:

[0108] The image recognition module is used to acquire terrain images ahead of the robot's path and perform deblurring processing. It then uses a lightweight neural network to extract ground feature data from the blurred images.

[0109] The control signal generation module is used to match ground characteristic data with a preset gait control parameter database to obtain a gait control parameter set, and generate a control signal based on the gait control parameter set using a proportional-derivative control method.

[0110] The control signal correction module is used to correct the control signal based on a reinforcement learning algorithm and generate target control commands to drive the joint actuator.

[0111] The real-time feedback adjustment module is used to collect status feedback data in real time during the execution of the target control command, and compare and optimize the feedback data with the currently executed target control command to achieve feedback adjustment.

[0112] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0113] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0114] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0115] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0116] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A robot gait adaptive adjustment method combining terrain feature recognition, characterized in that, Includes the following steps: The terrain image ahead of the robot's path is collected and deblurred. Ground feature data is extracted from the blurred image using a lightweight neural network. Ground characteristic data is matched with a preset gait control parameter database to obtain a gait control parameter set, and a control signal is generated based on the gait control parameter set using a proportional-derivative control method; The control signal is corrected based on a reinforcement learning algorithm, and a target control command is generated to drive the joint actuator. During the execution of the target control command, status feedback data is collected in real time, and the feedback data is compared and optimized with the currently executed target control command to achieve feedback adjustment.

2. The robot gait adaptive adjustment method combining terrain feature recognition according to claim 1, characterized in that, The process of collecting terrain images ahead of the robot's path and performing deblurring is as follows: Each frame of the terrain image is converted into a single-channel grayscale image, and a sharpness index is calculated. If the sharpness index is lower than the preset blur threshold, the current image is determined to be a blurry image and deblurring is performed. Edge detection processing is performed on the blurred image to obtain the edge intensity map and orientation map; Divide the directional map into several directional intervals, count the number of pixels in each direction, and construct an edge direction distribution histogram; If the proportion of pixels in the main direction exceeds the preset direction concentration threshold, it is judged as linear motion blur; Otherwise, it is a non-linear jitter blur; Perform deblurring operations on linear motion blur and nonlinear jitter blur.

3. The robot gait adaptive adjustment method combining terrain feature recognition according to claim 2, characterized in that, The deblurring operation for linear motion blur and nonlinear jitter blur specifically involves: A two-dimensional Fourier transform is performed on the blurred image based on linear motion blur to obtain the blur direction and spacing, and the blur length is inferred from the spacing. A linear motion fuzzy kernel is constructed based on the fuzzy direction and fuzzy length as a linear fuzzy point diffusion function; Based on the point spread function, a non-blind deconvolution algorithm is used to restore the blurred image and obtain the blurred image. Based on nonlinear jitter blur, the blurred image is set as the observed image, the estimated sharp image and the nonlinear blur kernel are initialized, and an optimization objective function is constructed; Based on the objective function, an alternating minimization strategy is used to optimize the image and the blur kernel until a preset number of iterations is reached, resulting in an optimized nonlinear blur kernel. The nonlinear blur kernel is used as the blur kernel for non-blind deconvolution to perform deblurring and obtain a clear image.

4. The robot gait adaptive adjustment method combining terrain feature recognition according to claim 1, characterized in that, The extraction of ground characteristic data from the blurred image using a lightweight neural network specifically involves: The blurred image is preprocessed to obtain a standardized image; Using standardized images as input, a lightweight convolutional neural network model is constructed for preliminary identification of ground features, and the output is a ground feature classification map. Obtain the confidence level in the ground feature classification map, and based on the preset confidence level threshold, extract the image regions with low confidence levels in the ground feature classification map to construct the corresponding candidate region mask map; Based on the candidate region mask map, the corresponding high-resolution image sub-blocks are extracted from the terrain image, and the sub-blocks are used as input to the preset refined recognition model; The output of the refined recognition model is fused with the ground feature classification map to generate a ground feature semantic segmentation map, which is then converted into structured ground feature data.

5. The robot gait adaptive adjustment method combining terrain feature recognition according to claim 1, characterized in that, The process involves matching ground characteristic data with a preset gait control parameter database to obtain a gait control parameter set, and then generating a control signal based on the gait control parameter set using a proportional-derivative control method. Specifically: The ground characteristic data is linearly normalized to obtain the ground feature vector; The Euclidean distance between the current ground feature vector and the ground feature vector in the database is calculated based on the K-nearest neighbor algorithm, and the set of gait control parameters corresponding to the minimum distance is selected as the output result. The current state information of the robot is obtained, and the control correction term is calculated based on the error between the matched set of control parameters and the state information using the proportional-derivative control method. Control signals are generated based on control correction terms.

6. The robot gait adaptive adjustment method combining terrain feature recognition according to claim 1, characterized in that, The step of modifying the control signal based on the reinforcement learning algorithm and generating the target control command to drive the joint actuator is as follows: Construct a reinforcement learning model, taking ground feature vectors, state information, and control signals as input states; The adjustment factor of the control signal is used as the output action; Construct a reward function, which includes a vibration spectrum index, a gait symmetry measure, and a contact surface stability index; With the goal of maximizing the reward function, an adjustment factor for the control signal output by the reinforcement learning model is generated based on the current input state. The adjustment factor and control signal are fused using a fuzzy logic reasoning method to generate target control commands to drive the joint actuator.

7. The robot gait adaptive adjustment method combining terrain feature recognition according to claim 6, characterized in that, The fuzzy logic reasoning method fuses the adjustment factor and control signal to generate the target control command, specifically as follows: The adjustment factor of the control signal output by the reinforcement learning model is used as the input variable of the fuzzy controller to construct the corresponding membership function and map the continuous values ​​of each adjustment factor to the fuzzy state. Based on the fuzzy state, fuzzy logic reasoning is performed using a preset empirical rule base to generate corresponding fuzzy control outputs. The centroid method is used to convert the fuzzy control output into a numerical correction quantity; The numerical correction value is fused with the control signal to generate a target control command, which includes a target joint angle value, a target joint angular velocity value, a target joint torque value, and a target joint damping value. Perform physical constraint verification on the target control command, and output the verified target control command to drive the joint actuator.

8. The robot gait adaptive adjustment method combining terrain feature recognition according to claim 7, characterized in that, During the execution of the target control command, real-time status feedback data is collected, and the feedback data is compared and optimized with the currently executed target control command to achieve feedback adjustment. Specifically: Acquire state feedback data during the execution of target control commands by the joint actuator. The state feedback data includes actual joint angle value, actual joint angular velocity value, actual joint torque value, foot-ground contact force value, and acceleration value. The errors are calculated by comparing the status feedback data with the target control commands. Based on the error and state feedback data, a quantitative feedback evaluation index is constructed, which includes a trajectory tracking error index, a gait stability index, and a contact impact index. The quantitative feedback evaluation indicators are combined into a multidimensional state vector, which is then transmitted as input data to the reinforcement learning model to obtain the optimization adjustment factor for the next control cycle. The target control command for the next control cycle is generated by using the optimized adjustment factor, thereby realizing closed-loop feedback regulation.

9. A system using the robot gait adaptive adjustment method combining terrain feature recognition as described in any one of claims 1-8, characterized in that, It includes an image recognition module, a control signal generation module, a control signal correction module, and a real-time feedback adjustment module, and the modules are interconnected. The image recognition module is used to acquire terrain images ahead of the robot's path and perform deblurring processing. It then uses a lightweight neural network to extract ground feature data from the blurred images. The control signal generation module is used to match ground characteristic data with a preset gait control parameter database to obtain a gait control parameter set, and generate a control signal based on the gait control parameter set using a proportional-derivative control method. The control signal correction module is used to correct the control signal based on a reinforcement learning algorithm and generate target control commands to drive the joint actuator. The real-time feedback adjustment module is used to collect status feedback data in real time during the execution of the target control command, and compare and optimize the feedback data with the currently executed target control command to achieve feedback adjustment.

Images