A method and system for autonomous exploration of an unstructured unknown environment based on a quadruped robot

Abstract

The application discloses a method and system for autonomous exploration of a quadruped robot in an unstructured unknown environment, and belongs to the technical field of robot autonomous exploration. The method comprises the following steps: acquiring environment perception data, generating a global exploration target based on the environment perception data; establishing a crossing risk map according to the current local environment, searching for a local path in the crossing risk map; coordinating the global exploration target and the local path through a multi-level decision fusion mechanism to generate an exploration path; and enabling the quadruped robot to execute exploration according to the exploration path, and updating the environment map and adjusting the exploration target in real time to form an autonomous exploration closed loop. The application solves the problems of low exploration efficiency, poor safety, and path planning not meeting kinematic constraints of a traditional quadruped robot in an unknown complex terrain, and can realize efficient, safe and autonomous exploration in line with the motion characteristics of the quadruped robot, and is suitable for scenes such as disaster rescue, terrain survey and military reconnaissance.

Description

Technical Field

[0001] This application belongs to the field of autonomous exploration technology for robots, and in particular relates to a method and system for autonomous exploration of unstructured unknown environments based on quadruped robots. Background Technology

[0002] With the rapid development of robotics technology, quadruped robots are increasingly being used in complex terrain environments, such as disaster relief, terrain surveying, and military reconnaissance. However, achieving autonomous exploration in completely unknown environments still faces many challenges: 1. Exploration efficiency issues: Traditional boundary-based exploration methods only consider local views and lack an overall understanding of the structure of large areas, resulting in redundant exploration paths and low efficiency, especially in complex terrain where they are prone to getting stuck in local optima.

[0003] 2. Safety Issues: When quadruped robots move in unstructured terrain, the geometric features of the terrain and the robot's kinematic constraints must be fully considered. Traditional path planning methods such as A* and RRT often only consider obstacle avoidance while ignoring the risks of terrain traversal, which may lead to the robot becoming unstable or tipping over.

[0004] 3. Adaptability of motion constraints: Quadruped robots have complex kinematic characteristics and gait constraints. Traditional methods of separating global and local planning are difficult to generate paths that simultaneously satisfy exploration efficiency and motion feasibility.

[0005] Currently, academia and industry have proposed various solutions. Existing methods for exploring unknown environments typically separate global exploration target generation from local path execution. The global layer focuses more on the efficiency of unknown area coverage, while the local layer focuses more on obstacle avoidance or local drivability. However, when quadruped robots face unstructured terrain, relying solely on grid occupancy or a single terrain indicator is insufficient to accurately reflect the robot's actual traversal risk. Furthermore, global exploration targets in complex terrain may induce the robot into high-risk areas, making it difficult to balance exploration efficiency and movement safety. However, existing technologies still have the following shortcomings: 1. A single method cannot simultaneously address the dual requirements of exploration efficiency and safety; 2. There is a lack of mechanisms to effectively integrate global exploration strategies with local path planning; 3. Insufficient adaptability to the motion constraints specific to quadruped robots. Summary of the Invention

[0006] The purpose of this application is to overcome the shortcomings of the prior art and provide a method and system for autonomous exploration of unstructured unknown environments based on quadruped robots. This method and system can integrate large-area perception and risk perception path planning, and solve problems such as low exploration efficiency, poor safety, and path planning that does not meet the motion constraints of quadruped robots in the prior art.

[0007] The objective of this application is achieved through the following technical solution: A method for autonomous exploration of unstructured unknown environments using a quadruped robot, the method comprising: Acquire environmental perception data, and generate a global exploration target based on the environmental perception data; A traversal risk map is created based on the current local environment, and a local path is searched within the traversal risk map. An exploration path is generated by coordinating the global exploration objective and the local path through a multi-level decision fusion mechanism. The quadruped robot is instructed to follow the exploration path and update the environmental map and adjust the exploration target in real time, thus forming an autonomous exploration closed loop.

[0008] Furthermore, the acquisition of environmental perception data and the generation of a global exploration target based on the environmental perception data include: The raw point cloud data is acquired and, after preprocessing, combined with the robot's pose information to construct a local navigability map. The frontier unknown areas are extracted based on the local accessibility map, and clustering is performed based on the regional connectivity. The unknown regions after clustering are evaluated for area and risk level is calculated. The exploration order of the unknown regions is sorted according to preset rules to generate a global exploration target sequence.

[0009] Furthermore, the traversal risk map includes converting local environmental factors into a risk representation map applicable to the motion characteristics of a quadruped robot. The local environmental factors include terrain geometry, robot body stability, environmental uncertainty, and path continuity. The risk representation map applicable to the motion characteristics of a quadruped robot includes representations of terrain accessibility, body stability, and motion continuity.

[0010] Furthermore, the process of coordinating the global exploration objective and the local path through a multi-level decision fusion mechanism to generate the exploration path includes: The multi-level decision-making fusion mechanism includes global exploration, local risk planning, and motion control; The global exploration includes: Obtain the area, accessibility, and exploration rewards of the unknown region, and generate a global exploration target point sequence based on the area, accessibility, and exploration rewards of the unknown region; The local risk planning generates a local safe path to the current target based on the traversal risk map; The motion control includes: The local safety path is converted into control commands for the quadruped robot.

[0011] Furthermore, the step of converting the local safe path into control instructions for the quadruped robot includes: Within each decision cycle, a comprehensive evaluation value of candidate paths is calculated based on global exploration gains, local safe path risks, and the robot's current state, thereby determining the optimal path and converting the optimal path into quadruped robot motion control commands.

[0012] Furthermore, the real-time updating of the environment map and adjustment of exploration objectives to form an autonomous exploration closed loop includes: The quadruped robot continuously acquires new environmental point clouds during its exploration process; The local and global maps are updated in real time based on the new environmental point cloud. If a change in obstacles, a failure of the path, or the completion of the objective is detected, global planning and local path planning will be retried. The planning parameters are adaptively optimized based on path execution time, robot stability, and energy consumption.

[0013] On the other hand, this application also provides a system for autonomous exploration of unstructured unknown environments based on quadruped robots, the system comprising: An environmental sensing module acquires environmental sensing data. A global exploration module, which generates a global exploration target based on the environmental perception data; The path planning module establishes a crossing risk map based on the current local environment and searches for local paths in the crossing risk map. The decision fusion module coordinates the global exploration objective and the local path through a multi-level decision fusion mechanism to generate an exploration path; The map update module enables the quadruped robot to explore along the exploration path, and updates the environmental map and adjusts the exploration target in real time, forming an autonomous exploration closed loop.

[0014] Furthermore, the global exploration module includes: An unknown region extraction unit extracts the leading unknown region; A regional assessment unit, which calculates the area, accessibility, and security level of each region; The target point generation unit generates a global exploration target sequence based on a large-area perception strategy.

[0015] Furthermore, the path planning module includes: A terrain unit division unit, wherein the terrain unit division unit divides a local planning area into multiple terrain units; A node risk calculation unit calculates the node crossing risk based on slope, roughness, elevation difference, obstacle distance, and perception uncertainty. An edge weight calculation unit calculates the edge weights of the risk graph based on distance cost, terrain change cost, attitude change cost, and path continuity cost. The path search unit searches for executable paths that meet risk requirements in the traversal risk map.

[0016] Furthermore, the decision fusion and control module includes: A terrain complexity evaluation unit calculates terrain complexity based on slope complexity, roughness complexity, obstacle density, average risk of the risk map, map uncertainty, and robot posture stability. A dynamic weight adjustment unit adjusts the exploration efficiency weight and the safety weight according to the terrain complexity; A path selection unit, which selects the final execution path from candidate paths based on adjusted weights; The motion control unit translates the final execution path into speed, heading, and gait control commands for the quadruped robot.

[0017] The beneficial effects of this application are as follows: (1) The exploration efficiency of this application is significantly improved: the large area perception strategy reduces repeated exploration paths and improves the coverage efficiency of unknown areas.

[0018] (2) The safety of the planned path in this application is greatly enhanced: the path planning based on the risk map fully considers the terrain geometric risks and robot motion constraints, effectively reducing the risk of safety accidents such as overturning and jamming.

[0019] (3) The application has good motion adaptability: it is designed for the motion characteristics of quadruped robots, and the generated path conforms to the gait and dynamic constraints of quadruped robots.

[0020] (4) This application has strong real-time performance and computing power adaptability: it adopts a hierarchical management and incremental update mechanism to ensure real-time planning capabilities in large-scale environments.

[0021] (5) This application has high robustness: it supports multi-level decision fusion and closed-loop feedback mechanism so that the system can adapt to dynamic changes in the environment. Attached Figure Description

[0022] Figure 1 This is an overall architecture block diagram of the autonomous exploration system for unstructured unknown environments based on quadruped robots provided in the embodiments of this application; Figure 2 This is a global planning flowchart of the large-area perception module in an embodiment of this application; Figure 3 This is a schematic diagram of the risk map construction and path planning in an embodiment of this application; Figure 4 This is a flowchart of the multi-level decision fusion mechanism in the embodiments of this application. Detailed Implementation

[0023] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0024] Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0025] In the field of autonomous exploration technology for robots, existing technologies still have the following shortcomings: 1. A single method cannot simultaneously address the dual requirements of exploration efficiency and safety; 2. There is a lack of mechanisms to effectively integrate global exploration strategies with local path planning; 3. Insufficient adaptability to the motion constraints specific to quadruped robots.

[0026] To address the aforementioned technical problems, the following embodiments of a method and system for autonomous exploration of unstructured unknown environments based on quadruped robots are proposed in this application.

[0027] 1. Environmental perception and data preprocessing: The system acquires the raw point cloud output from the lidar and simultaneously obtains IMU data and robot pose output from the SLAM system.

[0028] Outlier removal, voxel downsampling, and ground segmentation are performed on the original point cloud to filter out invalid point clouds with a distance exceeding a preset threshold. A local accessibility map is constructed by combining robot pose information, and terrain accessibility evaluation results are generated based on slope, elevation difference, and ground roughness.

[0029] 2. Large-area perception and global exploration target generation: Unknown frontier regions are extracted based on local accessibility maps, and clustering is performed based on regional connectivity. The clustered unknown regions are evaluated for area, accessibility analysis, and risk level calculation; unknown regions with larger areas and lower risks are prioritized as exploration targets; a global exploration target sequence is generated, and the current optimal exploration target is output.

[0030] 3. Risk Map Construction and Local Path Planning: In this embodiment, the risk map construction and local path planning are used to generate local motion paths for quadruped robots in unstructured unknown environments that balance safety, accessibility, and path efficiency. The risk map is not based solely on the obstacle occupancy state as the basis for path planning, but rather comprehensively considers factors such as terrain geometry, robot body stability, environmental uncertainty, and path continuity, transforming the local environment into a risk representation map suitable for the quadruped robot's motion characteristics.

[0031] Specifically, the system uses the robot's current pose as the starting point for local planning, and the current exploration target point output by the global exploration module as the endpoint constraint for local planning. A local terrain evaluation window is then established within a preset radius around the robot. Let the robot's current pose be: ; in, This indicates the robot's current pose state; This indicates the robot's lateral position coordinates in the global map coordinate system; This represents the robot's vertical position coordinates in the global map coordinate system; This indicates the robot's current heading angle. The global exploration target point is: ; in, Indicates the current location of the global exploration target point; This represents the horizontal coordinate of the target point in the map coordinate system; This represents the vertical coordinate of the target point in the map coordinate system. The local planning area is defined as: ; in, This indicates the local planning area at the current moment. Indicates the local planning radius. It represents any spatial location in a local map.

[0032] The system first performs rasterization or voxelization on the local point cloud map, dividing the local planning area into several terrain units: ; in, Let N represent the i-th terrain unit, and N represent the total number of terrain units in the local area.

[0033] For each terrain unit The system uses the 3D point cloud set within the unit. Calculate terrain features such as height, slope, roughness, elevation difference, obstacle distance, and point cloud confidence. The average terrain height is expressed as: ; in, This represents the average height of the i-th terrain unit; This represents the set of three-dimensional point clouds contained within the i-th terrain unit; This indicates the number of points in the point cloud set; Represents any point in a point cloud set; Represents point cloud points The height coordinates.

[0034] Terrain slope is calculated using local ground normal vectors. Let the terrain unit be... The local fitting plane normal vector is: ; in, This represents the local ground fitting plane normal vector corresponding to the i-th terrain unit; , , Let x, y, and z represent the components of the normal vector in the x, y, and z directions, respectively. Then, the slope risk characteristic is expressed as: ; in, This represents the slope risk characteristic of the i-th terrain unit; Represents the vertical component of the local ground normal vector; when The closer the value is to 1, the closer the terrain is to a horizontal plane, and the lower the slope risk; when The smaller the value, the greater the slope inclination and the higher the risk of slope inclination.

[0035] Ground roughness is calculated by the degree of dispersion of point cloud height within a terrain unit: ; in, This represents the surface roughness of the i-th terrain unit; This represents the average elevation of the area. Topographic elevation differences are represented as follows: ; in, This represents the maximum height change within the i-th terrain unit; This represents the function for calculating the maximum value. This represents the function for calculating the minimum value. The obstacle approach risk is represented as: ; in, This indicates the obstacle approach risk for the i-th terrain unit; Representing terrain units Distance to the nearest obstacle or impassable area This represents the distance attenuation coefficient. This formula indicates that the closer one is to an obstacle, the higher the risk of approaching the obstacle.

[0036] To avoid instability in risk assessment due to sparse point clouds or sensor occlusion, the system further introduces a perception uncertainty factor: ; in, This represents the environmental perception uncertainty of the i-th terrain unit; Indicates the number of point clouds within a terrain unit. This represents the threshold number of point clouds used to determine perceptual sufficiency. When the number of point clouds is insufficient... An increase in the number of such numbers indicates insufficient environmental information in the area, thus increasing the risk of crossing the area.

[0037] After obtaining the basic terrain features, the system constructs node crossing risks based on the quadruped robot's motion characteristics. Unlike wheeled robots, quadruped robots in complex terrain need to determine not only whether an area is occupied by obstacles, but also whether that area is suitable as a center of movement for the robot. Therefore, this embodiment decomposes node risks into terrain geometry risks, posture stability risks, and perception uncertainty risks.

[0038] For terrain units Its comprehensive node risk is defined as: ; in, Indicates the node risk value. Let represent the weighting coefficients for slope, roughness, elevation difference, obstacle proximity, and perceived uncertainty, respectively, and satisfy the following: ; This is the normalized risk mapping function.

[0039] In one alternative implementation, the risk mapping function takes a piecewise saturated form: ; Where x can represent any terrain feature such as slope, roughness, elevation difference, or uncertainty. and These represent the low-risk threshold and high-risk threshold corresponding to the feature, respectively.

[0040] After comprehensively considering terrain risks, the node crossing risk is updated as follows: ; in, This indicates that the robot's body passes through a risky area. Let these represent the weights of terrain risk and body traversal risk, respectively, and satisfy the following conditions: ; Body through risk This is used to determine whether the robot's body is likely to collide with obstacles while passing through the area, or whether changes in terrain might cause the robot's posture to exceed the permissible range. It is defined as: ; in, This indicates the predicted pitch angle as the robot passes through the area. Indicates the predicted roll angle. and These represent the maximum permissible pitch and roll angles, respectively. This indicates the predicted clearance between the bottom of the fuselage and the terrain or obstacle. Indicates the minimum safe ground clearance. To prevent extremely small positive numbers from being divided by zero.

[0041] After completing the node risk calculation, the system constructs a traversal risk graph: ; in, This represents the current moment's traversal risk map; Represents the set of nodes in the risk graph. Denotes the set of edges. This represents the set of edge weights.

[0042] Each node Corresponding to one candidate passable terrain unit When two terrain units are spatially adjacent and satisfy the kinematic constraints of a quadruped robot, an edge is established between the corresponding nodes: ; The creation of edges must simultaneously satisfy the following constraints: ; ; ; ; in, Representing nodes respectively Spatial location, This indicates the maximum permissible distance for a single step or partial movement of the robot. Indicates the maximum permissible angle of change of heading. This indicates the height change between two nodes. This indicates the maximum height difference that the robot can cross. Indicates the curvature of the path segment. This indicates the maximum permissible curvature.

[0043] For edges that satisfy the constraints The system calculates the cost of traversing the area: ; in, This represents the edge weight of the path from node i to node j; This represents the cost of distance between nodes. Indicates the cost of terrain change. This represents the cost of attitude change. Represents the cost of path continuity. For the corresponding weights.

[0044] The cost of terrain change is defined as: ; in, These represent the adjustment coefficients for slope variation and roughness variation, respectively.

[0045] The attitude change cost is defined as: ; The path continuity cost is defined as: ; in, Indicates that by node Pointing to node Path direction, This indicates the direction of the previous path segment. This item is used to suppress frequent sharp turns in the path and improve the gait stability of the quadruped robot during execution.

[0046] Thus, the system obtains a traversal risk map that simultaneously expresses terrain accessibility, body stability, and motion continuity. This risk map enables local path planning to move beyond relying solely on a two-dimensional occupancy grid and instead reflect the real-world traversal risks faced by the quadruped robot in complex terrain.

[0047] 4. Multi-level decision fusion and motion control: In this embodiment, multi-level decision fusion and motion control are used to coordinate the global exploration objective, the path planning results of the local traversal risk map, and the robot's underlying motion control commands, enabling the quadruped robot to adaptively adjust its exploration strategy according to the complexity of the terrain in an unknown environment. The core of this part is: when the environment is relatively flat, with few obstacles and low risk, the system increases the exploration efficiency weight, allowing the robot to prioritize and quickly cover unknown areas; when the environment has a large slope, high roughness, dense obstacles, or unstable robot posture, the system increases the safety weight, allowing the robot to prioritize low-risk paths.

[0048] The multi-level decision fusion mechanism in this embodiment includes a global exploration layer, a local path decision layer, and a motion control execution layer.

[0049] The global exploration layer is used to generate a sequence of global exploration target points based on the area of ​​unknown regions, regional accessibility, and exploration rewards. Let the global candidate target set be: ; in, Let m represent the m-th candidate target to be explored.

[0050] The local path decision layer receives the traversal risk map constructed in step 3 and the corresponding candidate local paths, and generates a local safe path to the current target within the risk map. Let the set of local candidate paths be: ; in, Represents a local set of candidate paths; Indicates the first There are 10 candidate paths; Q represents the number of candidate paths. The motion control execution layer is used to convert local paths into velocity, posture, and gait control commands that the quadruped robot can execute.

[0051] Within each decision cycle, the system calculates a comprehensive evaluation value for candidate paths based on global exploration gains, local path risks, and the robot's current state. ; in, Indicate candidate path The overall evaluation value, This indicates the evaluation item for exploration efficiency. Indicates safety evaluation items, This indicates the control executability evaluation item. These represent the dynamic weights of exploration efficiency, security, and control executability at the current moment.

[0052] The candidate path selection results are as follows: ; When the optimal path Once determined, the system converts it into motion control commands for the quadruped robot. Let the target point for path tracking be... The robot's current pose is The position error is: ; The heading error is: ; Based on path risk and terrain complexity, the system generates the expected linear velocity: ; in, This represents the robot's desired linear velocity; Indicates the maximum permissible speed of movement; Indicates the minimum speed of movement; Indicates the normalized terrain complexity. This represents the normalized risk value near the current tracking path point. The formula indicates that when the risk is high or the terrain complexity is high, the robot automatically reduces its travel speed; when the risk is low and the terrain is simple, the robot increases its travel speed to improve exploration efficiency.

[0053] The desired angular velocity is defined as: ; in, This indicates the heading control gain.

[0054] The final output speed control command is: ; in, Indicates gait pattern.

[0055] 5. Map updates and closed-loop feedback: The robot continuously acquires new environmental point clouds during the exploration process; Real-time updates of local and global maps; If a change in obstacles, a failure of the path, or the completion of the objective is detected, global planning and local path planning will be retried. The planning parameters are adaptively optimized based on path execution time, robot stability, and energy consumption.

[0056] Reference Figure 1,like Figure 1 The diagram shown is the overall architecture block diagram of the quadruped robot-based autonomous exploration system for unstructured unknown environments provided in this embodiment. The specific architecture of the system is as follows: S1. Environmental Perception and Map Building. The system receives raw perception data from LiDAR, IMU, and depth camera, and obtains robot pose information through the SLAM system. Subsequently, the raw point cloud is filtered, the ground is segmented, and voxel downsampling is performed to construct a local drivability map. The system evaluates the drivability of the terrain based on parameters such as slope, elevation difference, and roughness.

[0057] S2. Large-area perception and global exploration target generation. The system identifies frontier unknown areas based on local maps and performs clustering and structural analysis on these unknown areas. Subsequently, it generates an exploration value score based on area, accessibility, and risk level, prioritizing large-scale unknown areas as current exploration targets.

[0058] S3. Risk Map Construction and Local Path Planning. The system constructs a risk map based on the current local environment and calculates path risk weights by combining the quadruped robot's motion constraints. Subsequently, it searches the risk map for low-risk paths that satisfy both safety and motion feasibility, and then performs smooth optimization on the paths.

[0059] S4. Multi-level decision fusion and closed-loop feedback. The system coordinates the global exploration objective with local path planning and dynamically adjusts the weights of exploration efficiency and safety based on environmental complexity. During the robot's execution of the planned path, the system monitors the robot's status, terrain changes, and path execution effects in real time, and dynamically updates the map and planning parameters based on feedback results.

[0060] Reference Figure 2 ,like Figure 2 The diagram shown is a global planning flowchart of the large-area perception module in this embodiment. This step aims to improve the efficiency of exploring unknown environments through environmental structure understanding.

[0061] The system first extracts the frontier regions and clusters them based on region connectivity. Then, it calculates the area, distance, and risk level of each unknown region, and uses a comprehensive scoring function to calculate its exploration value.

[0062] During the target generation phase, the system prioritizes larger unknown areas as exploration targets to avoid decreased exploration efficiency caused by frequent target switching. Simultaneously, for smaller, more nearby areas, the system employs a local supplementary exploration mechanism to achieve fine-grained environmental coverage.

[0063] This step can improve the coverage of exploration in unknown environments and reduce duplicate paths.

[0064] Reference Figure 3 ,like Figure 3 The diagram shown is a schematic diagram of the risk map construction and path planning in this embodiment. This step is used to achieve safe path planning in complex terrain environments.

[0065] The system starts from the robot's current position and samples nodes in the passable area on a local map. Then, it generates node risk values ​​based on terrain slope, obstacle density, and ground roughness.

[0066] During the edge weight calculation stage, the system further introduces quadruped robot kinematic constraints, including step length constraints, turning radius constraints, and body stability constraints, to ensure that the planned path meets the robot's actual motion capabilities.

[0067] Subsequently, the system uses a graph search algorithm to generate low-risk paths and performs smooth optimization on the paths to improve path continuity and execution stability.

[0068] Reference Figure 4 ,like Figure 4 The diagram shown is a flowchart of the multi-level decision fusion mechanism in this embodiment. This step establishes a closed-loop coordination mechanism between global exploration and local path planning.

[0069] The system receives global exploration objectives, risk map path information, and robot status information, and dynamically adjusts the exploration efficiency weight and safety weight according to the environmental complexity.

[0070] During path execution, the system monitors the robot's posture, energy consumption, and path execution error in real time, and adaptively adjusts the planning parameters based on the feedback results.

[0071] When a path failure is detected or a significant change occurs in the environment, the system re-triggers global exploration and local path planning, thereby achieving continuous and stable exploration in complex dynamic environments.

[0072] Based on the above-described method for autonomous exploration of unstructured unknown environments by quadruped robots, this embodiment also provides a system for autonomous exploration of unstructured unknown environments by quadruped robots.

[0073] The system includes: The environmental perception module is used to acquire environmental point cloud, pose, and inertial data, and to construct a local map; The global exploration module is used to identify unknown areas and generate global exploration objectives; The path planning module is used to create a crossing risk map based on the current local environment and search for local paths in the crossing risk map. The decision fusion module is used to coordinate the global exploration objective and the local path through a multi-level decision fusion mechanism to generate an exploration path; The map update module enables the quadruped robot to perform exploration along the described exploration path, and updates the environmental map and adjusts the exploration target in real time, forming an autonomous exploration closed loop.

[0074] Furthermore, the global exploration module includes: An unknown region extraction unit extracts the leading unknown region; A regional assessment unit, which calculates the area, accessibility, and security level of each region; The target point generation unit generates a global exploration target sequence based on a large-area perception strategy.

[0075] As one implementation method, the path planning module includes: Topographic unit division unit is used to divide a local planning area into multiple topographic units; The node risk calculation unit is used to calculate the node crossing risk based on slope, roughness, elevation difference, obstacle distance, and perceived uncertainty. The edge weight calculation unit is used to calculate the edge weights of the risk graph based on distance cost, terrain change cost, attitude change cost, and path continuity cost. The path search unit is used to search for executable paths that meet the risk requirements in the traversal risk graph.

[0076] As one implementation method, the decision fusion and control module includes: The terrain complexity evaluation unit is used to calculate terrain complexity based on slope complexity, roughness complexity, obstacle density, average risk of the risk map, map uncertainty, and robot posture stability. The dynamic weight adjustment unit is used to adjust the exploration efficiency weight and safety weight according to the terrain complexity. The path selection unit is used to select the final execution path from the candidate paths based on the adjusted weights. The motion control unit is used to convert the final execution path into speed, heading, and gait control commands for the quadruped robot.

[0077] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for autonomous exploration of unstructured unknown environments using a quadruped robot, characterized in that, The method includes: Acquire environmental perception data, and generate a global exploration target based on the environmental perception data; A traversal risk map is created based on the current local environment, and a local path is searched within the traversal risk map. An exploration path is generated by coordinating the global exploration objective and the local path through a multi-level decision fusion mechanism. The quadruped robot is instructed to follow the exploration path and update the environmental map and adjust the exploration target in real time, thus forming an autonomous exploration closed loop.

2. The method for autonomous exploration of unstructured unknown environments based on quadruped robots as described in claim 1, characterized in that, The process of acquiring environmental perception data and generating a global exploration target based on the environmental perception data includes: The raw point cloud data is acquired and, after preprocessing, combined with the robot's pose information to construct a local navigability map. The frontier unknown areas are extracted based on the local accessibility map, and clustering is performed according to the regional connectivity. The unknown regions after clustering are evaluated for area and risk level is calculated. The exploration order of the unknown regions is sorted according to preset rules to generate a global exploration target sequence.

3. The method for autonomous exploration of unstructured unknown environments based on quadruped robots as described in claim 1, characterized in that, The crossing risk map includes a risk representation map that converts local environmental factors into a risk representation map applicable to the motion characteristics of a quadruped robot. The local environmental factors include terrain geometry, robot body stability, environmental uncertainty, and path continuity. The risk representation map applicable to the motion characteristics of a quadruped robot includes representations of terrain passability, body stability, and motion continuity.

4. The method for autonomous exploration of unstructured unknown environments based on quadruped robots as described in claim 1, characterized in that, The process of generating an exploration path by coordinating the global exploration objective and the local path through a multi-level decision fusion mechanism includes: The multi-level decision-making fusion mechanism includes global exploration, local risk planning, and motion control; The global exploration includes: Obtain the area, accessibility, and exploration rewards of the unknown region, and generate a global exploration target point sequence based on the area, accessibility, and exploration rewards of the unknown region; The local risk planning generates a local safe path to the current target based on the constructed risk map; The motion control includes: The local safety path is converted into control commands for the quadruped robot.

5. The method for autonomous exploration of unstructured unknown environments based on quadruped robots as described in claim 4, characterized in that, The control instructions for converting the local safe path into a quadruped robot include: Within each decision cycle, a comprehensive evaluation value of candidate paths is calculated based on global exploration gains, local safe path risks, and the robot's current state, thereby determining the optimal path and converting the optimal path into quadruped robot motion control commands.

6. The method for autonomous exploration of unstructured unknown environments based on quadruped robots as described in claim 1, characterized in that, The real-time updating of the environment map and adjustment of exploration objectives to form an autonomous exploration closed loop includes: The quadruped robot continuously acquires new environmental point clouds during its exploration process; The local and global maps are updated in real time based on the new environmental point cloud. If a change in obstacles, a failure of the path, or the completion of the objective is detected, global planning and local path planning will be retried. The planning parameters are adaptively optimized based on path execution time, robot stability, and energy consumption.

7. A system for autonomous exploration of unstructured unknown environments based on quadruped robots, characterized in that, The system includes: An environmental sensing module acquires environmental sensing data. A global exploration module, which generates a global exploration target based on the environmental perception data; The path planning module establishes a crossing risk map based on the current local environment and searches for local paths in the crossing risk map. The decision fusion module coordinates the global exploration objective and the local path through a multi-level decision fusion mechanism to generate an exploration path; The map update module enables the quadruped robot to explore along the exploration path, and updates the environmental map and adjusts the exploration target in real time, forming an autonomous exploration closed loop.

8. The system for autonomous exploration of unstructured unknown environments based on quadruped robots as described in claim 7, characterized in that, The global exploration module includes: An unknown region extraction unit extracts the leading unknown region; A regional assessment unit, which calculates the area, accessibility, and security level of each region; The target point generation unit generates a global exploration target sequence based on a large-area perception strategy.

9. The system for autonomous exploration of unstructured unknown environments based on quadruped robots as described in claim 7, characterized in that, The path planning module includes: A terrain unit division unit, wherein the terrain unit division unit divides a local planning area into multiple terrain units; A node risk calculation unit calculates the node crossing risk based on slope, roughness, elevation difference, obstacle distance, and perception uncertainty. An edge weight calculation unit calculates the edge weights of the risk graph based on distance cost, terrain change cost, attitude change cost, and path continuity cost. The path search unit searches for executable paths that meet risk requirements in the traversal risk map.

10. The system for autonomous exploration of unstructured unknown environments based on quadruped robots as described in claim 7, characterized in that, The decision fusion and control module includes: A terrain complexity evaluation unit calculates terrain complexity based on slope complexity, roughness complexity, obstacle density, average risk of the risk map, map uncertainty, and robot posture stability. A dynamic weight adjustment unit adjusts the exploration efficiency weight and the safety weight according to the terrain complexity; A path selection unit, which selects the final execution path from candidate paths based on adjusted weights; The motion control unit translates the final execution path into speed, heading, and gait control commands for the quadruped robot.

Images