Underground cable intelligent laying robot autonomous obstacle avoidance control method, device, equipment and medium

Abstract

The invention relates to an autonomous obstacle avoidance control method, device and equipment for an underground cable intelligent laying robot and a medium. The method comprises the following steps: acquiring underground environment data through a multi-source sensor array, performing multi-modal fusion on the position, shape and material information of an obstacle in the underground environment data, and generating an environment sensing model; classifying and identifying the obstacles based on the environment perception model, and outputting a classification result containing threat levels; wherein categories in classification identification comprise rigid obstacles, flexible obstacles and dynamic obstacles; according to a classification result and a robot kinematics constraint condition, generating an obstacle avoidance feasibility boundary; and generating a self-adaptive obstacle avoidance strategy based on the obstacle avoidance feasibility boundary, wherein the self-adaptive obstacle avoidance strategy is used for indicating the laying robot to output a control instruction. By adopting the method, the intelligent laying robot can make high-precision obstacle avoidance decisions on complex and diversified obstacles in an unstructured underground space.

Description

Technical Field

[0001] The present invention belongs to the field of automation control technology, and in particular relates to an autonomous obstacle avoidance control method, device, equipment and medium for an intelligent underground cable laying robot. Background Art

[0002] Underground cable laying, a crucial component of modern urban infrastructure construction, is directly linked to the safety and stability of power transmission. With the acceleration of urbanization and the increasing complexity of underground spaces, the development of intelligent laying robots has become a key technological tool for improving construction efficiency and quality. These robots are required to autonomously complete cable laying tasks in narrow and complex underground environments. Their autonomous obstacle avoidance capabilities are crucial for ensuring operational safety and continuity, making their research of significant importance.

[0003] However, current research and application in this area remain significantly limited. Many laying robots rely on simple obstacle avoidance rules or pre-set paths, making them difficult to adapt to the diverse types of obstacles and dynamically changing construction conditions in underground environments. This results in poor obstacle avoidance and can even cause equipment damage or interruptions to cable laying.

[0004] Against this backdrop, the key challenges facing research are becoming increasingly apparent. The complexity of underground environments requires robots to flexibly adjust their action strategies to accommodate different obstacles. However, existing technologies often lack adaptive maneuverability when faced with diverse obstacles. This deficiency further leads to another key issue: the difficulty in balancing optimal path planning with the continuity of cable laying during obstacle avoidance. When path planning fails to fully account for the robot's motion characteristics and physical limitations, it can lead to unstable obstacle avoidance movements, compromising cable laying quality. Summary of the Invention

[0005] Based on this, it is necessary to provide an autonomous obstacle avoidance control method, device, equipment and medium for an underground cable intelligent laying robot to address the above technical problems, which can enable the intelligent laying robot to make high-precision obstacle avoidance decisions for complex and diverse obstacles in unstructured underground spaces.

[0006] In a first aspect, the present application provides an autonomous obstacle avoidance control method for an intelligent underground cable laying robot, comprising:

[0007] Collect underground environmental data through a multi-source sensor array, perform multimodal fusion on the location, shape, and material information of obstacles in the underground environmental data, and generate an environmental perception model;

[0008] Obstacles are classified and identified based on the environmental perception model, and classification results including threat levels are output. The categories included in the classification include rigid obstacles, flexible obstacles, and dynamic obstacles.

[0009] Generate the obstacle avoidance feasibility boundary based on the classification results and the robot's kinematic constraints;

[0010] An adaptive obstacle avoidance strategy is generated based on the obstacle avoidance feasibility boundary, and the adaptive obstacle avoidance strategy is used to instruct the laying robot to output control instructions.

[0011] In one embodiment, underground environmental data is collected through a multi-source sensor array, and the location, shape, and material information of obstacles in the underground environmental data are fused multimodally to generate an environmental perception model, including:

[0012] The 3D point cloud data of the obstacle is obtained and noise is filtered out to generate denoised point cloud data;

[0013] Analyze the motion characteristics of obstacles based on the acquired millimeter-wave radar echo signal and generate velocity vector data;

[0014] Identify the thermal radiation characteristics of the obstacle surface through infrared thermal imaging sensors and generate material type identification;

[0015] The denoised point cloud data, velocity vector data and material type identification are temporally and spatially aligned to generate an environmental perception model.

[0016] In one embodiment, obstacles are classified and identified based on the environment perception model, and classification results including threat levels are output, including:

[0017] The environment perception model is input into the 3D object detection network to extract the geometric feature vectors of obstacles, where the 3D object detection network is the PointNet++ network;

[0018] Based on the material type identification, the physical deformation characteristics of the obstacle are obtained, wherein the physical deformation characteristics are used to determine the deformation coefficient threshold of the obstacle;

[0019] Combining geometric feature vectors with physical deformation characteristics, a clustering algorithm is used to classify obstacle type labels and obtain classification results.

[0020] The collision time prediction value is calculated based on the motion characteristics of dynamic obstacles to determine the threat level of the obstacles.

[0021] In one embodiment, generating an obstacle avoidance feasibility boundary based on the classification results and the robot's kinematic constraints includes:

[0022] According to the joint torque limit of the laying robot and the motion parameters of the cable release mechanism, the kinematic constraint equations are constructed;

[0023] Generate safe avoidance zones for each obstacle based on threat level, including:

[0024] According to the circumscribed geometric shape of the rigid obstacle, the first safety distance is expanded outward through GIS buffer analysis to generate a rigid avoidance area, which is then determined as a safe avoidance area.

[0025] and / or,

[0026] According to the spatial distribution of the seepage influence range of the flexible obstacle, the second safety distance is contracted inward through GIS buffer analysis to generate a flexible avoidance area, which is then determined as the safe avoidance area.

[0027] In the space defined by the kinematic constraint equations, a random sampling algorithm is used to generate a set of reachable poses of the robot.

[0028] Spatial analysis is performed on the safe avoidance area and the set of reachable poses to generate the obstacle avoidance feasibility boundary.

[0029] In one embodiment, generating an adaptive obstacle avoidance strategy based on an obstacle avoidance feasibility boundary includes:

[0030] Based on the dynamic window method, the velocity space sampling of the reachable pose set is performed to generate a set of candidate obstacle avoidance paths, where the range of the velocity space sampling is limited by the robot's maximum acceleration and the kinematic constraint equations;

[0031] Construct a multi-objective optimization function, the objective functions of which include path complexity, cable tension fluctuation and joint motion smoothness;

[0032] Optimize the candidate obstacle avoidance path set based on the multi-objective optimization function to generate an optimized obstacle avoidance path set;

[0033] The optimized obstacle avoidance path set is smoothed to obtain a continuous and executable obstacle avoidance trajectory, which is used to generate an adaptive obstacle avoidance strategy.

[0034] In a second aspect, the present application also provides an autonomous obstacle avoidance control device for an intelligent underground cable laying robot, comprising:

[0035] The obstacle perception module is used to collect underground environmental data through a multi-source sensor array, perform multimodal fusion on the location, shape and material information of obstacles in the underground environmental data, and generate an environmental perception model;

[0036] The recognition and classification module is used to classify and identify obstacles based on the environmental perception model and output classification results including threat levels. The categories in the classification and identification include rigid obstacles, flexible obstacles, and dynamic obstacles.

[0037] The obstacle avoidance boundary generation module is used to generate the obstacle avoidance feasibility boundary based on the classification results and the robot's kinematic constraints;

[0038] The adaptive control module is used to generate an adaptive obstacle avoidance strategy based on the obstacle avoidance feasibility boundary, and the adaptive obstacle avoidance strategy is used to instruct the laying robot to output control instructions.

[0039] In a third aspect, the present application also provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor implements the above-mentioned autonomous obstacle avoidance control method of the intelligent underground cable laying robot when executing the computer program.

[0040] In a fourth aspect, the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned autonomous obstacle avoidance control method of the intelligent underground cable laying robot is implemented.

[0041] The autonomous obstacle avoidance control method, device, equipment, and medium for the intelligent underground cable laying robot collects underground environmental data through a multi-source sensor array, integrates the location, shape, and material information of obstacles, constructs a high-precision environmental perception model, and accurately extracts the multi-dimensional characteristics of unstructured underground space. Based on this model, obstacles are classified and identified by their rigidity, flexibility, and dynamic attributes, and classification results with threat levels are output to accurately identify the risk levels of obstacles in different physical states. The robot's kinematic constraints are combined to generate an obstacle avoidance feasibility boundary, quantify the robot's operable space in complex environments, and provide a physical constraint basis for decision-making. Based on the boundary, an adaptive obstacle avoidance strategy is generated, and control instructions are dynamically optimized to achieve accurate obstacle avoidance of diverse obstacles. This technical solution improves the robot's obstacle avoidance accuracy and adaptability to diverse obstacles in complex underground environments through a closed-loop mechanism of perception fusion, intelligent classification, constraint modeling, and dynamic decision-making, effectively ensuring the continuity of cable laying and construction safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0043] Figure 1 A schematic flow chart of the autonomous obstacle avoidance control method for an intelligent underground cable laying robot provided by the present invention;

[0044] Figure 2 A schematic diagram of a multimodal obstacle perception method provided by the present invention;

[0045] Figure 3 This is a structural schematic diagram of the autonomous obstacle avoidance control device of the underground cable intelligent laying robot provided by the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0047] First, a brief introduction is given to the terms involved in the embodiments of this application.

[0048] Three-dimensional point cloud data refers to a discrete data set consisting of a large number of three-dimensional spatial coordinate points obtained by sensing devices such as laser radar (LiDAR), depth cameras or structured light scanning. Each data point contains its position information in the three-dimensional coordinate system and can be attached with attributes (such as reflection intensity, RGB color, normal vector, etc.). Its core feature is to accurately characterize the surface geometry and spatial distribution of the target object in an unstructured form, which is suitable for three-dimensional reconstruction and environmental perception of complex scenes. Compared with two-dimensional images or voxel data, three-dimensional point clouds can more accurately restore the three-dimensional geometric information of the real world, providing centimeter-level accuracy environmental perception capabilities for intelligent underground cable laying robots.

[0049] Millimeter-wave radar echo signals refer to electromagnetic wave signals reflected from targets and received by radar systems operating in the millimeter-wave frequency band (30-300GHz). By analyzing the frequency, phase, and time difference between the transmitted and reflected waves, they can obtain key information such as the target object's distance, radial velocity, azimuth, and scattering characteristics. This signal is formed by the frequency-modulated continuous wave (FMCW) generated by the transmitter after being reflected by the target. It contains multi-dimensional data such as time domain waveform, Doppler frequency shift, and polarization characteristics. In technical implementation, the echo signal is processed through mixing, Fourier transform, and constant false alarm rate (CFAR) detection to extract the target range-velocity two-dimensional spectrum, and the azimuth information is analyzed through beamforming algorithms to ultimately generate high-resolution spatial target features. Compared with traditional radar, millimeter-wave radar has higher ranging accuracy (up to centimeter level) and anti-interference ability due to its shorter wavelength. It is particularly suitable for real-time detection of dynamic obstacles (such as mobile construction equipment) in underground cable laying scenarios and penetrating identification of hidden obstacles under non-line-of-sight conditions (such as detecting buried pipelines through soil medium reflection), providing robots with robust environmental perception capabilities.

[0050] The PointNet++ network is a deep learning-based point cloud processing framework. It effectively addresses issues such as point cloud data disorder, non-uniformity, and missing local structure through a hierarchical feature learning mechanism combined with local area sampling, grouping, and multi-layer perceptron (MLP) for multi-scale feature aggregation. The network structure comprises a sampling layer (furthest point sampling), a grouping layer (local neighborhood construction based on sphere queries), and a feature encoding layer (point-by-point MLP and max pooling). It progressively captures local geometric details and global semantic information from the raw point cloud, and achieves multi-level feature fusion through skip connections, improving the ability to represent complex 3D scenes.

[0051] Based on the above explanation of terms, the implementation environment of the autonomous obstacle avoidance control method for the intelligent underground cable laying robot provided in the embodiment of the present application is explained. Schematically, the implementation environment includes: a multimodal sensor array, a terminal, and a processor. The processor, the multimodal sensor array, and the terminal are connected via network signals; the multimodal sensor array includes but is not limited to millimeter wave radars, lidars, infrared thermal imaging sensors, optical sensors, and inertial measurement units (IMUs); the processor can be a central processing unit, a multi-core processor, a neural network processor, or an artificial intelligence chip, etc., which are not limited here.

[0052] In combination with the above explanations of terms and implementation environments, the application scenarios of the embodiments of this application are explained. The autonomous obstacle avoidance control method for the intelligent underground cable laying robot provided in the embodiments of this application can be applied to, but not limited to, the following scenarios:

[0053] This technical solution can be applied to cable laying operations in complex underground utility corridors. The robot collects environmental data within the corridor using a multi-source sensor array, generates an environmental perception model, and classifies and identifies rigid, flexible, and dynamic obstacles, assessing their threat level. Based on the obstacle avoidance feasibility boundary, it generates an adaptive obstacle avoidance strategy, guiding the robot to safely and efficiently complete cable laying tasks within the corridor's confined spaces, effectively preventing equipment damage and construction delays caused by collisions.

[0054] In new underground cable tunnel construction, this technical solution helps robots navigate complex obstacles within the tunnel. By integrating multi-source sensor data to build a precise environmental perception model, it can classify and identify areas of accumulated water (dynamic obstacles), temporary support structures (rigid obstacles), and flexible pipelines left over from construction. An adaptive strategy based on obstacle avoidance boundary generation enables the robot to autonomously plan its path within the unstructured environment during the construction phase, ensuring the continuity of cable laying operations, improving construction efficiency, and reducing the risk of human intervention.

[0055] In underground mine cable laying applications, this technology solution helps robots operate safely within the mine's complex underground environment. Leveraging a multimodal fusion environmental perception model, the robot can identify obstacles within the mine, such as rigid ore piles, flexible conveyor belts, and dynamic transport vehicles. By generating a feasible obstacle avoidance boundary and implementing adaptive obstacle avoidance strategies, the robot can autonomously avoid obstacles within the mine's narrow and volatile environment, ensuring smooth cable laying operations while avoiding potential safety hazards caused by collisions and improving the automation level of mine electrical installation.

[0056] Illustratively, the autonomous obstacle avoidance control method for the intelligent underground cable laying robot provided in the embodiment of the present application can also be applied to other application scenarios. It is only used as an example here and is not limited to the specific application scenario.

[0057] In an exemplary embodiment, Figure 1 As shown, a method for autonomous obstacle avoidance control of an intelligent underground cable laying robot is provided. This embodiment uses the method as an example of a terminal in the aforementioned implementation environment. It is understandable that the method can also be applied to a server, or to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps 101 to 104:

[0058] Step 101: Collect underground environmental data through a multi-source sensor array, perform multimodal fusion on the position, shape and material information of obstacles in the underground environmental data, and generate an environmental perception model.

[0059] Specifically, a multi-source sensor array can include lidar, ultrasonic sensors, and visual sensors, distributed across the robot to provide comprehensive environmental awareness. For example, lidar emits laser pulses and measures reflection time to obtain precise distance and location information about obstacles. Ultrasonic sensors can detect obstacles at close range and are particularly suitable for obstacle detection in complex environments. Visual sensors can provide information on the shape and material of obstacles, helping the robot gain a more comprehensive understanding of its surroundings. Through the collaborative work of these sensors, the collected underground environmental data includes key information such as the location, shape, and material of obstacles.

[0060] Step 102 : classify and identify obstacles based on the environment perception model, and output classification results including threat levels; wherein the categories in the classification and identification include rigid obstacles, flexible obstacles, and dynamic obstacles.

[0061] Rigid obstacles are defined as those with a fixed physical structure, a very low deformation coefficient, and resistance to elastic or plastic deformation. Their geometric shape is stable, and collisions can easily damage equipment or block paths. Flexible obstacles are defined as those made of soft materials with high deformability or prone to state changes under external forces. Their primary threat comes from indirect risks such as physical seepage, collapse, or insufficient stability. Dynamic obstacles are those whose position or motion changes over time, requiring real-time trajectory tracking and prediction of potential collision risks. For example, classification and recognition algorithms can employ deep learning methods, such as using 3D convolutional neural networks to analyze features of environmental perception models to identify rigid, flexible, and dynamic obstacles. Alternatively, rule-based classification methods can be employed, such as setting classification rules based on the obstacle's material, shape, and motion characteristics. Threat level assessment is based on obstacle type, size, location, and motion. For example, a large rigid obstacle located in the robot's path is considered a higher threat. Classification results and threat level information provide key insights for developing obstacle avoidance strategies.

[0062] Step 103: Generate an obstacle avoidance feasibility boundary based on the classification results and the robot's kinematic constraints.

[0063] Specifically, the robot's kinematic constraints include maximum steering angle, minimum turning radius, maximum climbing angle, and speed limit. These parameters determine the range of paths the robot can safely traverse. The obstacle avoidance feasibility boundary is determined by calculating the safe distance between the obstacle and the robot and combining it with the robot's kinematic capabilities. For example, for rigid obstacles, the safe distance is calculated based on the obstacle size and the robot's minimum turning radius. For dynamic obstacles, their speed and direction must also be considered, and the safe distance is dynamically adjusted by predicting their trajectory. The resulting obstacle avoidance feasibility boundary provides clear spatial constraints for the robot's path planning.

[0064] Step 104 : generating an adaptive obstacle avoidance strategy based on the obstacle avoidance feasibility boundary, wherein the adaptive obstacle avoidance strategy is used to instruct the laying robot to output a control instruction.

[0065] Specifically, the adaptive obstacle avoidance strategy dynamically adjusts the robot's motion parameters by analyzing the environmental perception model and the obstacle avoidance feasibility boundary in real time. For example, if a rigid obstacle with a high threat level is detected ahead of the robot, the strategy instructs the robot to slow down and steer around it. However, if a flexible obstacle with a low threat level is encountered, the robot is allowed to pass at an appropriate speed. Control instructions include steering angle, speed adjustment, lift height, and propulsion force, ensuring the robot can safely and efficiently complete cable laying tasks in complex underground environments while avoiding equipment damage and construction delays caused by collisions or jams.

[0066] The autonomous obstacle avoidance control method, device, equipment, and medium for the intelligent underground cable laying robot collects underground environmental data through a multi-source sensor array, integrates the location, shape, and material information of obstacles, constructs a high-precision environmental perception model, and accurately extracts the multi-dimensional characteristics of unstructured underground space. Based on this model, obstacles are classified and identified by their rigidity, flexibility, and dynamic attributes, and classification results with threat levels are output to accurately identify the risk levels of obstacles in different physical states. The robot's kinematic constraints are combined to generate an obstacle avoidance feasibility boundary, quantify the robot's operable space in complex environments, and provide a physical constraint basis for decision-making. Based on the boundary, an adaptive obstacle avoidance strategy is generated, and control instructions are dynamically optimized to achieve accurate obstacle avoidance of diverse obstacles. This technical solution improves the robot's obstacle avoidance accuracy and adaptability to diverse obstacles in complex underground environments through a closed-loop mechanism of perception fusion, intelligent classification, constraint modeling, and dynamic decision-making, effectively ensuring the continuity of cable laying and construction safety.

[0067] like Figure 2 As shown, in one embodiment, underground environmental data is collected through a multi-source sensor array, and the location, shape, and material information of obstacles in the underground environmental data are multimodally fused to generate an environmental perception model, including:

[0068] Step 201 : 3D point cloud data of the obstacle is acquired and noise is filtered out to generate denoised point cloud data.

[0069] Specifically, the acquired three-dimensional point cloud data is subjected to noise filtering to generate denoised point cloud data. The noise filtering algorithm can adopt a statistical-based method, such as using the Random Sample Consensus (RANSAC) algorithm to remove outliers; or adopt a filtering-based method, such as smoothing the point cloud data through Gaussian filtering or median filtering. The denoised point cloud data can more accurately reflect the true shape and position of the obstacle and reduce the impact of noise on subsequent processing. For example, in a complex underground environment, the point cloud data may be interfered with by dust, water vapor, etc. The noise filtering process can effectively eliminate the optical scattering noise in the underground humid environment, so that the geometric restoration error of the obstacle surface contour is reduced to the centimeter level.

[0070] Step 202: Analyze the motion characteristics of the obstacle based on the acquired millimeter-wave radar echo signal to generate velocity vector data.

[0071] Specifically, millimeter-wave radar determines the speed and direction of an obstacle by measuring the frequency change (Doppler effect) of the echo signal. For example, a fast Fourier transform (FFT) can be used to perform spectral analysis on the echo signal to extract velocity information; or a tracking algorithm based on a Kalman filter can be used to estimate the obstacle's trajectory in real time. The generated velocity vector data contains the obstacle's direction and speed, providing a key basis for identifying dynamic obstacles. For example, for a moving underground water pipeline or construction vehicle, millimeter-wave radar can capture its motion status in real time, ensuring that the robot can respond to the obstacle in a timely manner.

[0072] Step 203: Identify the thermal radiation characteristics of the obstacle surface through an infrared thermal imaging sensor and generate a material type identifier.

[0073] Specifically, infrared thermal imaging sensors can detect infrared radiation emitted from the surface of obstacles. Objects of different materials have different thermal radiation characteristics. For example, metal pipes generally have higher thermal conductivity and their surface temperature distribution is relatively uniform; while plastic pipes have lower thermal conductivity and their surface temperature varies more significantly. By analyzing infrared thermal imaging data, a threshold-based method or machine learning algorithm (such as a support vector machine) can be used to classify and identify obstacle materials. The generated material type identification can help the robot better understand the physical properties of the obstacle. For example, rigid metal obstacles and flexible plastic obstacles require different treatment methods in obstacle avoidance strategies.

[0074] Step 204 : performing spatiotemporal alignment fusion on the denoised point cloud data, velocity vector data, and material type identifier to generate an environmental perception model.

[0075] Specifically, the spatiotemporal alignment fusion process includes two steps: time synchronization and spatial registration. Time synchronization ensures that data from different sensors are in the same time frame; spatial registration converts data in different coordinate systems into a unified global coordinate system. The fusion algorithm can use a probability model-based method, such as conditional random field (CRF) or Bayesian network, to jointly model multi-source data; or use a deep learning-based fusion method, such as using a 3D convolutional neural network to extract features from point cloud data, and combining velocity vectors and material type identification for multimodal fusion. The generated environmental perception model can comprehensively reflect the position, shape, motion state and material information of obstacles, providing a high-precision description of the environment for subsequent classification and obstacle avoidance decisions.

[0076] In one embodiment, obstacles are classified and identified based on the environment perception model, and classification results including threat levels are output, including:

[0077] The environment perception model is input into the 3D target detection network to extract the geometric feature vectors of obstacles, where the 3D target detection network is the PointNet++ network.

[0078] Specifically, the PointNet++ network first samples and groups point cloud data to form multi-level point sets. It then uses a multi-layer perceptron (MLP) to extract features from each point set. Finally, through a feature propagation mechanism, high-level features are fused into low-level features to generate a geometric feature vector containing information about the obstacle's shape, size, and position. For example, for a rigid obstacle, the network can extract its regular geometric shape features; for a flexible obstacle, it can capture its irregular morphological characteristics. The extracted geometric feature vector provides an important geometric basis for subsequent obstacle classification.

[0079] Based on the material type identification, the physical deformation characteristics of the obstacle are obtained, wherein the physical deformation characteristics are used to determine the deformation coefficient threshold of the obstacle.

[0080] Specifically, physical deformation characteristics are used to characterize the deformation coefficient threshold of an obstacle, that is, the possibility and degree of deformation of the obstacle when subjected to external force. For example, the deformation coefficient threshold of a rigid obstacle is relatively high, and usually only a slight deformation will occur when subjected to a large external force; while the deformation coefficient threshold of a flexible obstacle is relatively low, and it is easy to produce obvious deformation or diffusion under the influence of external force or environmental terrain. For example, the physical deformation characteristics of the obstacle can be determined based on the pre-established mapping relationship between different materials and deformation coefficient thresholds, combined with the material type identification in the environmental perception model. For example, for obstacles made of metal materials, the deformation coefficient threshold range is determined based on their material properties and structural characteristics by consulting the material mechanical properties database or experimental data. The determination of physical deformation characteristics helps the robot judge whether the obstacle can be safely bypassed or whether a special obstacle avoidance strategy needs to be adopted during the obstacle avoidance process.

[0081] Combining geometric feature vectors with physical deformation characteristics, the obstacle type labels are divided through clustering algorithm to obtain classification results.

[0082] Specifically, clustering algorithms can classify obstacles into different categories based on the similarity of features. For example, the K-Means clustering algorithm is used to cluster obstacles into categories such as rigid obstacles, flexible obstacles, and dynamic obstacles based on the distribution of geometric feature vectors and physical deformation characteristics in the feature space. The K-Means algorithm randomly initializes cluster centers and calculates the distance between each obstacle feature vector and the cluster center to assign it to the category to which the nearest cluster center belongs, and iteratively updates the cluster center until convergence. In addition, density-based clustering algorithms, such as DBSCAN, can also be used to use the density distribution in the feature space to discover clusters of arbitrary shapes and effectively identify different types of obstacles. Using clustering algorithms to classify obstacle type labels not only improves classification efficiency, but also adapts to the diversity of obstacle types in different scenarios, providing an accurate classification basis for subsequent threat level assessment.

[0083] The collision time prediction value is calculated based on the motion characteristics of dynamic obstacles to determine the threat level of the obstacles.

[0084] Specifically, a Kalman filter can be used to predict the future trajectory of an obstacle based on the velocity vector acquired by the millimeter-wave radar. For example, acceleration observations can be integrated to correct trajectory prediction errors. Furthermore, the robot's maximum braking acceleration and steering capabilities are combined to calculate a safe avoidance time threshold. For example, a high-threat-level command can be generated for a dynamic obstacle approaching at high speed, triggering an emergency obstacle avoidance strategy.

[0085] In one embodiment, generating an obstacle avoidance feasibility boundary based on the classification results and the robot's kinematic constraints includes:

[0086] The kinematic constraint equations are constructed according to the joint torque limits of the laying robot and the motion parameters of the cable release mechanism.

[0087] Specifically, the joint torque limit determines the maximum torque that each joint of the robot can generate, thereby limiting the robot's range of motion and speed. The cable release mechanism's motion parameters include the cable release speed, acceleration, and maximum extension length, which affect the robot's posture and position adjustment capabilities during the laying process. Kinematic constraint equations quantify these constraints through mathematical relationships. For example, the forward kinematics and inverse kinematics equations in robotics can be combined with joint torque constraints to establish a model of the robot's reachable space. This model can accurately define the areas that the robot can safely reach in complex underground environments, providing a basic framework for generating the feasibility boundary for subsequent obstacle avoidance.

[0088] Generate safe avoidance zones for each obstacle based on threat level, including:

[0089] According to the circumscribed geometric shape of the rigid obstacle, the first safety distance is expanded outward through GIS buffer analysis to generate a rigid avoidance area, which is then determined as a safe avoidance area.

[0090] And / or, according to the spatial distribution of the seepage influence range of the flexible obstacle, the second safety distance is contracted inward through GIS buffer analysis to generate a flexible avoidance area, and the flexible avoidance area is determined as the safe avoidance area.

[0091] For example, GIS spatial analysis technology can be used to generate differentiated avoidance areas based on threat level and obstacle type. For example, a buffer zone analysis is performed on the circumscribed geometric shape of a rigid obstacle (such as a concrete wall), and a rigid avoidance area is generated by expanding the safety distance outward. Specifically, the safety distance is dynamically adjusted according to the robot body size and dynamic error tolerance (such as body radius + 10cm positioning error). For flexible obstacles (such as water seepage layers), the influence range is predicted according to the seepage diffusion model, and the flexible avoidance area is generated by shrinking the safety distance inward. For example, the shrinkage amount is set based on the product of the seepage rate and the robot response time. According to the dynamic diffusion of the threat range (seepage influence area) of the flexible obstacle over time, the robot is allowed to avoid in advance in the safe area where the seepage has not yet reached. This method can simultaneously avoid rock walls (rigid) and water-permeable soft soil (flexible) in mine tunnels, thereby improving the obstacle avoidance coverage rate.

[0092] In the space defined by the kinematic constraint equations, a random sampling algorithm is used to generate the set of reachable poses of the robot.

[0093] Specifically, random sampling algorithms can randomly generate a large number of pose samples within the robot's kinematic constraints. These samples represent the robot's possible positions and postures. For example, a rapid-exploring random tree (RRT) algorithm or a probabilistic roadmap (PRM) algorithm can be used. The RRT algorithm constructs a continuous path by gradually expanding random samples from an initial pose toward the target area. The PRM algorithm pre-generates a graph structure containing a large number of random pose samples and then searches for feasible paths within the graph. Alternatively, adaptive sampling strategies can be used to balance exploration and exploitation efficiency. For example, sampling density is increased in narrow and complex areas, while computational overhead is reduced in open and flat areas. Furthermore, a pre-collision detection mechanism can be introduced to eliminate pose points that interfere with obstacles. For example, in cable crossing areas, poses whose distance to existing pipelines is less than a safety threshold are excluded, generating a highly reliable set of reachable poses. These algorithms can efficiently explore the robot's reachable area in complex underground environments. The generated set of reachable poses provides a comprehensive set of pose options for subsequent determination of the obstacle avoidance feasibility boundary.

[0094] Spatial analysis is performed on the safe avoidance area and the set of reachable poses to generate the obstacle avoidance feasibility boundary.

[0095] Specifically, the spatial analysis process involves performing geometric operations on the safe avoidance area and the set of reachable poses. For example, by calculating the intersection or difference between the two, the boundaries of the area that the robot can reach while avoiding obstacles are determined. The obstacle avoidance feasibility boundary clearly defines the range within which the robot can travel while satisfying kinematic constraints and avoidance requirements, providing clear spatial constraints for the final obstacle avoidance strategy generation. For example, within the generated obstacle avoidance feasibility boundary, the robot can safely plan a laying path while avoiding various obstacles, ensuring the smooth progress of the cable laying operation.

[0096] In one embodiment, generating an adaptive obstacle avoidance strategy based on an obstacle avoidance feasibility boundary includes:

[0097] Based on the dynamic window method, velocity space sampling is performed on the reachable pose set to generate a set of candidate obstacle avoidance paths, where the range of velocity space sampling is limited by the robot's maximum acceleration and kinematic constraint equations.

[0098] Specifically, the dynamic window method samples the robot's velocity space, taking into account its dynamic characteristics, such as maximum acceleration and maximum velocity, as well as the reachable range defined by the kinematic constraint equations. For example, the velocity space can be divided into multiple windows, each corresponding to a different velocity and acceleration combination. Random or regular sampling is then performed within these windows to generate a series of possible motion trajectories. These trajectories start from the current position and, while satisfying velocity and acceleration constraints, explore the set of reachable poses of the robot within the obstacle avoidance feasibility boundary. The resulting candidate obstacle avoidance path set contains a variety of potential obstacle avoidance routes, providing a diverse range of options for subsequent optimization.

[0099] A multi-objective optimization function is constructed, and the objective functions of the multi-objective optimization function include path complexity, cable tension fluctuation and joint motion smoothness.

[0100] Specifically, path complexity reflects the tortuosity and execution difficulty of the path and can be expressed as a function of the path length or the number of turns in the path. Cable tension fluctuation measures the degree of change in cable tension during installation. Excessive fluctuation can lead to cable damage or uneven installation, and can be calculated by analyzing the force applied to the cable along the path. Joint motion smoothness focuses on the continuity and stability of the robot's joint motion and can be assessed by the degree of sudden change in joint velocity and acceleration. A multi-objective optimization function comprehensively considers these three objectives. For example, a weighted sum method can be used to combine them into a comprehensive optimization objective function, where the weight coefficients are determined based on the actual application requirements. By constructing such a multi-objective optimization function, the advantages and disadvantages of different obstacle avoidance paths can be comprehensively evaluated, ensuring that the ultimately selected path is not only feasible but also optimal across multiple key performance indicators.

[0101] The candidate obstacle avoidance path set is optimized based on the multi-objective optimization function to generate the optimized obstacle avoidance path set.

[0102] Exemplarily, the path optimization process can be implemented by a variety of algorithms. For example, a hierarchical optimization strategy can be used to perform multi-objective iterative optimization on the candidate obstacle avoidance path set. Specifically, in the first stage, the Pareto front screening is used to eliminate obviously inferior paths; in the second stage, the improved NSGA-II algorithm is introduced to increase the priority of cable tension constraints in the crossover and mutation operations; in the third stage, the frontier paths are sorted based on fuzzy comprehensive evaluation. Through the above technical methods, in the permeable soft soil area, the optimized obstacle avoidance path set can simultaneously meet the requirements of shortest detour distance (reducing seepage exposure time), minimum tension fluctuation (preventing cable sheath wear) and joint motion coherence (avoiding mechanical jamming). The paths in the optimized obstacle avoidance path set have been significantly improved in terms of path complexity, cable tension fluctuation amount and joint motion smoothness, providing high-quality path options for the final obstacle avoidance trajectory generation.

[0103] The optimized obstacle avoidance path set is smoothed to obtain a continuous and executable obstacle avoidance trajectory, which is used to generate an adaptive obstacle avoidance strategy.

[0104] Specifically, path smoothing aims to eliminate abrupt points and discontinuities in the path, making the robot's movement smoother and more natural. For example, the B-spline curve interpolation method can be used to fit the optimized discrete path points into a smooth continuous curve; or the path can be smoothed using a filter, such as a Gaussian filter or a moving average filter, to reduce the high-frequency noise in the path. The smoothed obstacle avoidance trajectory not only meets the robot's kinematic and dynamic constraints, but also ensures that the cable tension changes smoothly during the laying process, reducing mechanical stress on the cable and improving the quality and efficiency of cable laying. This enables the generated adaptive obstacle avoidance strategy to be dynamically adjusted according to the actual environment and the robot's state, guiding the robot to complete the cable laying task safely and efficiently in complex underground environments.

[0105] In summary, the autonomous obstacle avoidance control method for the intelligent underground cable laying robot provided in this application collects underground environmental data through a multi-source sensor array, integrates the geometric, material and motion characteristics of obstacles, and generates a high-precision environmental perception model; extracts the geometric characteristics of obstacles based on the PointNet++ network, quantifies the material deformation characteristics in combination with thermal radiation data, and accurately classifies rigid, flexible and dynamic obstacles through a semi-supervised clustering algorithm, and predicts the dynamic threat level based on the motion characteristics; further, combines the robot kinematic constraints with GIS spatial analysis technology to generate differentiated avoidance areas for rigid and flexible obstacles, and outputs the obstacle avoidance feasibility boundary through reachable posture sampling and topological operations; generates a candidate path set based on the dynamic window method, constructs a multi-objective optimization function of path complexity, cable tension fluctuation and motion smoothness, and outputs a continuously executable adaptive obstacle avoidance strategy after hierarchical optimization and trajectory smoothing. The above technical solution can make high-precision obstacle avoidance decisions in unstructured underground spaces, improve the safety, efficiency and intelligence level of underground cable laying operations, and reduce the need for manual intervention and construction risks.

[0106] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0107] Based on the same inventive concept, the embodiments of the present application also provide an autonomous obstacle avoidance control device 10 for an intelligent underground cable laying robot for implementing the autonomous obstacle avoidance control method for an intelligent underground cable laying robot. The implementation solution provided by this device is similar to the implementation solution described in the above method. Therefore, the specific limitations of one or more embodiments of the autonomous obstacle avoidance control device 10 for an intelligent underground cable laying robot provided below can be found in the limitations of the autonomous obstacle avoidance control method for an intelligent underground cable laying robot above, and will not be repeated here.

[0108] In an exemplary embodiment, Figure 3 As shown, an autonomous obstacle avoidance control device 10 for an intelligent underground cable laying robot is provided, comprising:

[0109] The obstacle perception module 11 is used to collect underground environmental data through a multi-source sensor array, perform multimodal fusion on the position, shape and material information of obstacles in the underground environmental data, and generate an environmental perception model.

[0110] The identification and classification module 12 is used to classify and identify obstacles based on the environmental perception model and output classification results including threat levels; wherein the categories in the classification and identification include rigid obstacles, flexible obstacles and dynamic obstacles.

[0111] The obstacle avoidance boundary generation module 13 is used to generate an obstacle avoidance feasibility boundary based on the classification results and the robot kinematic constraints.

[0112] The adaptive control module 14 is used to generate an adaptive obstacle avoidance strategy based on the obstacle avoidance feasibility boundary, and the adaptive obstacle avoidance strategy is used to instruct the laying robot to output control instructions.

[0113] In one embodiment, the obstacle perception module 11 includes:

[0114] The point cloud acquisition unit is used to obtain the three-dimensional point cloud data of the obstacle and perform noise filtering on the three-dimensional point cloud data to generate denoised point cloud data.

[0115] The motion analysis unit is used to analyze the motion characteristics of obstacles based on the acquired millimeter-wave radar echo signal and generate velocity vector data.

[0116] The material recognition unit is used to identify the thermal radiation characteristics of the obstacle surface through an infrared thermal imaging sensor and generate a material type identification.

[0117] The spatiotemporal fusion unit is used to perform spatiotemporal alignment and fusion of denoised point cloud data, velocity vector data, and material type identification to generate an environmental perception model.

[0118] In one embodiment, the recognition and classification module 12 includes:

[0119] The feature extraction unit is used to input the environment perception model into the three-dimensional object detection network to extract the geometric feature vectors of obstacles, where the three-dimensional object detection network is the PointNet++ network.

[0120] The deformation modeling unit is used to obtain the physical deformation characteristics of the obstacle based on the material type identification, wherein the physical deformation characteristics are used to determine the deformation coefficient threshold of the obstacle.

[0121] The clustering classification unit is used to combine the geometric feature vector and the physical deformation characteristics, divide the obstacle type labels through the clustering algorithm, and obtain the classification results.

[0122] The threat assessment unit is used to calculate the collision time prediction value based on the motion characteristics of the dynamic obstacle and determine the threat level of the obstacle.

[0123] In one embodiment, the obstacle avoidance boundary generation module 13 includes:

[0124] The constraint modeling unit is used to construct kinematic constraint equations according to the joint torque limits of the laying robot and the motion parameters of the cable release mechanism.

[0125] The avoidance area generation unit is used to generate safe avoidance areas for each obstacle based on the threat level, including:

[0126] The rigid avoidance subunit is used to expand the first safety distance outward according to the circumscribed geometric shape of the rigid obstacle through GIS buffer analysis, generate a rigid avoidance area, and determine the rigid avoidance area as a safe avoidance area.

[0127] The flexible avoidance subunit is used to shrink the second safety distance inward according to the spatial distribution of the seepage influence range of the flexible obstacle through GIS buffer analysis, generate a flexible avoidance area, and determine the flexible avoidance area as a safe avoidance area.

[0128] The pose sampling unit is used to generate a set of reachable poses of the robot using a random sampling algorithm within the space defined by the kinematic constraint equations.

[0129] The boundary generation unit is used to perform spatial analysis on the safe avoidance area and the reachable pose set to generate the obstacle avoidance feasibility boundary.

[0130] In one embodiment, the adaptive control module 14 includes:

[0131] The path generation unit is used to perform velocity space sampling on the reachable pose set based on the dynamic window method to generate a candidate obstacle avoidance path set, where the range of velocity space sampling is limited by the robot's maximum acceleration and kinematic constraint equations.

[0132] The optimization modeling unit is used to construct a multi-objective optimization function, where the objective functions of the multi-objective optimization function include path complexity, cable tension fluctuation and joint motion smoothness.

[0133] The path optimization unit is used to optimize the candidate obstacle avoidance path set based on the multi-objective optimization function to generate an optimized obstacle avoidance path set.

[0134] The trajectory generation unit is used to perform path smoothing on the optimized obstacle avoidance path set to obtain a continuously executable obstacle avoidance trajectory, which is used to generate an adaptive obstacle avoidance strategy.

[0135] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the autonomous obstacle avoidance control method of an intelligent underground cable laying robot as described above are implemented.

[0136] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0137] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0138] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. An autonomous obstacle avoidance control method for an intelligent underground cable laying robot, characterized in that: The method comprises: Collect underground environmental data through a multi-source sensor array, perform multimodal fusion on the location, shape, and material information of obstacles in the underground environmental data, and generate an environmental perception model; Classify and identify the obstacles based on the environmental perception model, and output a classification result including a threat level; wherein the categories in the classification and identification include rigid obstacles, flexible obstacles, and dynamic obstacles; generating an obstacle avoidance feasibility boundary based on the classification results and the robot's kinematic constraints; An adaptive obstacle avoidance strategy is generated based on the obstacle avoidance feasibility boundary, and the adaptive obstacle avoidance strategy is used to instruct the laying robot to output a control instruction.

2. The method according to claim 1, characterized in that The method collects underground environmental data through a multi-source sensor array, performs multimodal fusion on the position, shape, and material information of obstacles in the underground environmental data, and generates an environmental perception model, including: The three-dimensional point cloud data of the obstacle is obtained, and noise filtering is performed on the three-dimensional point cloud data to generate denoised point cloud data; Analyzing the motion characteristics of the obstacle based on the acquired millimeter-wave radar echo signal to generate velocity vector data; Identify the thermal radiation characteristics of the obstacle surface through infrared thermal imaging sensors and generate material type identification; The denoised point cloud data, the velocity vector data and the material type identifier are temporally and spatially aligned and fused to generate the environmental perception model.

3. The method according to claim 2, characterized in that The classifying and identifying the obstacles based on the environment perception model and outputting a classification result including a threat level includes: Inputting the environment perception model into a three-dimensional object detection network to extract the geometric feature vector of the obstacle, wherein the three-dimensional object detection network is a PointNet++ network; obtaining a physical deformation characteristic of the obstacle based on the material type identifier, wherein the physical deformation characteristic is used to determine a deformation coefficient threshold of the obstacle; Combining the geometric feature vector and the physical deformation characteristics, dividing the obstacle type labels using a clustering algorithm to obtain the classification result; A collision time prediction value is calculated according to the motion characteristics of the dynamic obstacle to determine the threat level of the obstacle.

4. The method according to claim 1, wherein Generating an obstacle avoidance feasibility boundary based on the classification result and the robot kinematic constraints includes: Constructing kinematic constraint equations based on the joint torque limits of the laying robot and the motion parameters of the cable release mechanism; Generating a safe avoidance area for each obstacle based on the threat level includes: According to the circumscribed geometric shape of the rigid obstacle, a first safety distance is expanded outward through GIS buffer analysis to generate a rigid avoidance area, and the rigid avoidance area is determined as the safe avoidance area; and / or, According to the spatial distribution of the seepage influence range of the flexible obstacle, the second safety distance is contracted inwardly through GIS buffer analysis to generate a flexible avoidance area, and the flexible avoidance area is determined as the safe avoidance area; In the space defined by the kinematic constraint equations, a random sampling algorithm is used to generate a set of reachable poses of the robot; A spatial analysis is performed on the safe avoidance area and the set of reachable postures to generate the obstacle avoidance feasibility boundary.

5. The method according to claim 4, characterized in that Generating an adaptive obstacle avoidance strategy based on the obstacle avoidance feasibility boundary includes: Performing velocity space sampling on the reachable pose set based on a dynamic window method to generate a candidate obstacle avoidance path set, wherein the range of the velocity space sampling is limited by the robot's maximum acceleration and kinematic constraint equations; Constructing a multi-objective optimization function, wherein the objective functions of the multi-objective optimization function include path complexity, cable tension fluctuation and joint motion smoothness; Performing path optimization on the candidate obstacle avoidance path set based on the multi-objective optimization function to generate an optimized obstacle avoidance path set; Path smoothing is performed on the optimized obstacle avoidance path set to obtain a continuously executable obstacle avoidance trajectory, and the obstacle avoidance trajectory is used to generate the adaptive obstacle avoidance strategy.

6. An autonomous obstacle avoidance control device for an intelligent underground cable laying robot, characterized in that: The device comprises: The obstacle perception module is used to collect underground environmental data through a multi-source sensor array, perform multimodal fusion on the location, shape and material information of obstacles in the underground environmental data, and generate an environmental perception model; an identification and classification module, configured to classify and identify the obstacles based on the environmental perception model and output a classification result including a threat level; wherein the categories in the classification and identification include rigid obstacles, flexible obstacles, and dynamic obstacles; An obstacle avoidance boundary generation module is used to generate an obstacle avoidance feasibility boundary based on the classification results and the robot kinematic constraints; An adaptive control module is used to generate an adaptive obstacle avoidance strategy based on the obstacle avoidance feasibility boundary, and the adaptive obstacle avoidance strategy is used to instruct the laying robot to output control instructions.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images