A leg-foot type robot inspection navigation method and system under a narrow space

Abstract

The application relates to the field of robot inspection navigation, and discloses a leg-foot type robot inspection navigation method and system in a narrow space; the method acquires environment sensing data of a narrow inspection space and body state data of a four-foot robot, generates a continuous section constraint unit along an inspection channel; generates a dynamic posture envelope according to the body state data, a candidate body posture and a candidate gait, and calculates foot end landing safety windows; determines a target passing posture based on the section constraint unit, the dynamic posture envelope and the foot end landing safety windows; judges the point-of-view reachability of an inspection sensor to a to-be-inspected object, and generates a point-of-view compensation action when the point-of-view reachability is not available; generates an inspection navigation track according to the target passing posture, the foot end landing safety windows and the point-of-view compensation action; when a front section cannot be passed or does not meet a turning condition, generates a retreat path according to a safety residence point and a retreat posture. The method is suitable for narrow space inspection such as a pipe gallery, a cable trench and a device interlayer.

Description

Technical Field

[0001] This invention relates to the field of robot inspection and navigation, and specifically to a method and system for inspection and navigation of legged robots in confined spaces. Background Technology

[0002] Quadruped robots possess excellent terrain adaptability, enabling them to perform inspection tasks in confined spaces inaccessible to humans, such as pipe galleries, cable trenches, equipment mezzanines, and underground passages. Current quadruped robot navigation typically relies on sensors like LiDAR, depth cameras, and inertial measurement units to create environmental maps, combined with path planning, obstacle avoidance control, and gait control to complete inspection maneuvers. Compared to wheeled or tracked platforms, quadruped robots can traverse localized obstacles, adapt to uneven terrain, and can be equipped with inspection sensors such as visible light cameras and thermal imaging cameras to detect objects like instruments, valves, cable connectors, and support joints, thus demonstrating their application value in confined inspection scenarios.

[0003] However, confined spaces often contain constraints such as side walls, overhead components, cable trays, pipeline supports, low beams, and lateral protrusions. Existing navigation methods often focus on determining the direction of passage, planning the robot's central path, or judging passability based on ground elevation, easily simplifying the robot into a fixed outline or point mass for obstacle avoidance. For quadruped robots, their ability to navigate depends not only on whether their body shape can enter the passage, but also on the candidate body height, roll angle, pitch angle, yaw angle, and the sweep space of the swinging legs and the foot landing area under gait changes. If navigation is based solely on two-dimensional paths, ordinary point cloud maps, or ground elevation maps, problems may arise such as the path being deemed passable, but during actual walking, interference between the body and overhead components, collisions between the swinging legs and lateral protrusions, and the inability of the feet to obtain a stable support area.

[0004] Furthermore, inspection tasks require more than just robots to navigate confined spaces; they also demand that inspection sensors effectively observe the objects to be inspected. During low-profile passage, narrow-step passage, or lateral avoidance, the observation rays of the inspection sensors may be blocked by the robot body, side walls, cable trays, or pipeline supports. This results in the robot completing the passage but failing to obtain the necessary image, temperature, or acoustic data for inspection. Simultaneously, in some confined inspection sections, the robot may lack the space to turn around in place. If temporary obstacles appear ahead, the ground becomes slippery, or subsequent sections become impassable, existing methods of simple replanning or local obstacle avoidance are insufficient to ensure the robot's safe exit. Therefore, there is an urgent need for a quadruped robot inspection navigation method and system that can simultaneously consider cross-sectional space constraints, body posture envelope, swing leg sweep space, foot landing safety, inspection viewpoint accessibility, and retreat safety within confined spaces. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for inspection and navigation of legged robots in confined spaces, thereby solving the technical problems existing in the prior art.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution: A method for inspection and navigation of a legged robot in confined spaces includes the following steps: S1: Acquire environmental perception data of the confined inspection space where the quadruped robot is located, as well as the body state data of the quadruped robot; S2: Determine the extension direction of the inspection channel based on the environmental perception data, and generate multiple continuously arranged cross-sectional constraint units along the extension direction of the inspection channel. S3: Based on the body state data, candidate fuselage attitude and candidate gait, generate dynamic attitude envelopes corresponding to each cross-section constraint unit. The dynamic attitude envelopes include fuselage occupancy envelopes and swing leg sweep envelopes. S4: Calculate the foot landing safety window within each cross-sectional constraint element; S5: Generate candidate passage postures based on cross-sectional constraint units, dynamic attitude envelopes and foot landing safety windows, and determine the target passage posture from the candidate passage postures that meets the fuselage passage conditions, leg swing conditions and foot landing conditions. S6: Determine the viewpoint reachability of the inspection sensor for the object to be inspected based on the target's passage posture, and generate a viewpoint compensation action when the viewpoint reachability does not meet the conditions. S7: Generates an inspection navigation trajectory based on the target's passage posture, foot landing safety window, and viewpoint compensation action, and controls the quadruped robot to perform inspections according to the inspection navigation trajectory; S8: If there are no candidate passing postures that meet the conditions for fuselage passage, leg swing, and foot landing in the front section constraint unit, or if the front section constraint unit does not meet the turning condition, generate a retreat path based on the section constraint units that have been passed, the recorded safe stopping points, and the retreatable postures.

[0007] Preferably, the environmental perception data includes three-dimensional point cloud data, depth image data, and location information of the object to be inspected, and the body state data includes fuselage posture, joint state, and foot contact state. The cross-sectional constraint unit includes ground height, lateral clear width, top clear height, lateral protrusion position, and orientation of the object to be inspected, and is used to make spatial cross-judgments with the fuselage occupancy envelope, the swing leg sweep envelope, and the observation rays of the inspection sensors, respectively.

[0008] Preferably, multiple consecutively arranged cross-sectional constraint elements are generated along the extension direction of the inspection channel, including: The coordinates of the 3D point cloud data are unified and the point sets are divided to obtain the ground point set, side wall point set, top point set and protrusion point set; The extension direction of the inspection channel is determined based on the extended distribution of the side wall point set in the horizontal projection plane; Multiple cross-sectional planes are established along the extension direction of the inspection channel according to the set sampling interval. Each cross-sectional plane is perpendicular to the extension direction of the inspection channel. The sampling interval is set to be no greater than the single step length of the quadruped robot. Points within a preset thickness range before and after each cross-section plane are selected as point cloud data for the corresponding cross-section constraint unit. The preset thickness is determined by the point cloud sampling density and the set sampling interval. The ground height, lateral clearance, top clearance, and position of lateral protrusions are determined based on the point cloud data of the corresponding cross-sectional constraint unit. The object to be inspected, located between two adjacent cross-sectional planes, is projected onto the corresponding cross-sectional constraint unit to obtain the orientation of the object to be inspected.

[0009] Preferably, the generation of the dynamic attitude envelope includes: The basic envelope of the quadruped robot is established based on its body dimensions, the installation shape of the inspection sensors, and the boundary of its shape when its legs are folded up. Based on the candidate fuselage height, candidate roll angle, candidate pitch angle and candidate yaw angle, the basic fuselage envelope is transformed to obtain the fuselage occupancy envelope; The supporting leg and swinging leg are determined based on joint status and candidate gait, and the swing start point, swing end point and swing trajectory of each swinging leg are determined. The swing leg sweep envelope is generated based on the swing trajectory of each swing leg and the outer boundary of the corresponding leg link. The fuselage occupancy envelope and the swing leg sweep envelope are mapped to the coordinate system of the corresponding section constraint element to form the dynamic attitude envelope corresponding to that section constraint element.

[0010] Preferably, the calculation of the foot-landing safety window includes: The candidate foot placement area is determined based on the foot contact state and candidate gait; Areas in the candidate landing area that overlap with side walls, lateral protrusions, and bottom obstacles are eliminated; Calculate the ground slope and elevation difference based on the point cloud height changes within the candidate landing area; The foot support area is determined by the overlap area between the projected area of ​​the foot contact surface and the ground contact area within the candidate foot landing area. The area where the ground slope does not exceed the set slope threshold, the height difference does not exceed the set height difference threshold, the foot support area is not less than the set support area threshold, the distance between the foot boundary and the side wall and lateral protrusions is not less than the set lateral gap threshold, and the distance between the swing leg sweep envelope and the side wall and lateral protrusions is not less than the set swing gap threshold is defined as the foot landing safety window. Among them, the set support area threshold is determined by the foot contact surface area and the foot contact stability condition; the set lateral clearance threshold is determined by the minimum avoidance distance from the foot boundary to the side wall and lateral protrusions; the set swing clearance threshold is determined by the minimum avoidance distance from the outer boundary of the swing leg sweep envelope to the side wall and lateral protrusions; and the set slope threshold and set height difference threshold are jointly determined by the center of gravity projection range, leg workspace and joint limits of the quadruped robot in static support state.

[0011] Preferably, determining the target passing posture from the candidate passing postures that satisfies the fuselage passing conditions, leg swing conditions, and foot landing conditions includes: Multiple candidate flight attitudes are generated within the same cross-sectional constraint unit. Each candidate flight attitude includes candidate fuselage height, candidate roll angle, candidate pitch angle, candidate yaw angle, and candidate gait. Interference judgment is made between the fuselage occupancy envelope corresponding to each group of candidate passage attitudes and the lateral clear width, top clear height and lateral protrusion position of the cross-sectional constraint unit; Interference judgment is performed between the swing leg sweep envelope corresponding to each group of candidate passage postures and the side walls, top, lateral protrusions and bottom obstacles in the cross-sectional constraint unit; Match the foot landing point corresponding to each group of candidate passage postures with the foot landing safety window; When the fuselage occupancy envelope does not interfere, the swing leg sweep envelope does not interfere, and the landing point of each foot is within the corresponding foot landing safety window, the candidate passage posture is determined as the target passage posture.

[0012] Preferably, the accessibility of the inspection sensor to the object to be inspected is determined based on the target's passage posture, and a viewpoint compensation action is generated when the viewpoint accessibility condition is not met, including: Determine the sensor pose of the inspection sensor within the corresponding cross-sectional constraint unit based on the target's passage posture; Based on the sensor pose and the orientation of the object to be inspected, an observation ray is generated pointing from the inspection sensor to the object to be inspected. Determine whether the observed ray intersects with the sidewall boundary, top boundary, lateral protrusion boundary, or fuselage occupancy envelope boundary in the corresponding cross-sectional constraint unit, and calculate the observation distance between the inspection sensor and the object to be inspected. When the observation ray intersects with the side wall boundary, top boundary, lateral protrusion boundary or fuselage occupancy envelope boundary, or when the observation distance is not within the set observation distance range, the corresponding compensation sensor poses for the pause observation action, fuselage height adjustment action, fuselage yaw adjustment action and retrospective observation action are generated respectively. When the observation ray corresponding to the pose of the compensation sensor does not intersect with the side wall boundary, top boundary, lateral protrusion boundary and fuselage occupancy envelope boundary, and the corresponding observation distance is within the set observation distance range, the action corresponding to the pose of the compensation sensor is determined as the viewpoint compensation action. The observation distance range is determined by the effective imaging distance of the inspection sensor, the focusing distance, the minimum identifiable size of the object to be inspected, the image resolution, and the sensor calibration parameters.

[0013] Preferably, the stopping and observing action is to control the quadruped robot to stop moving on the front or rear side of the corresponding cross-sectional constraint unit and maintain the current body posture for observation; The aircraft height adjustment procedure involves adjusting the aircraft height while keeping the foot landing point within the foot landing safety window before conducting observation. The fuselage yaw adjustment action is to adjust the fuselage yaw angle and then observe it, provided that neither the fuselage occupancy envelope nor the swing leg sweep envelope interferes with the corresponding cross-sectional constraint element. The retrospective observation action involves controlling the quadruped robot to pass through the corresponding cross-sectional constraint unit, and then using the sensor pose located behind the corresponding cross-sectional constraint unit to observe the object to be inspected.

[0014] Preferably, the generation of the inspection navigation trajectory includes: The target passage attitude sequence through each cross-section constraint unit is determined according to the arrangement order of the cross-section constraint units. The continuity judgment of the target's passage attitude between adjacent cross-section constraint units is performed. The continuity judgment includes the change in fuselage altitude, the change in roll angle, the change in pitch angle, the change in yaw angle, and the number of gait switching. Eliminate target passage posture sequences that do not meet the continuity judgment; Establish a connection between the target passage posture sequence that meets the continuity judgment and includes viewpoint compensation actions and the corresponding object to be inspected; The inspection navigation trajectory is generated based on the associated target passage posture sequence, foot landing safety window and viewpoint compensation action. The allowable ranges for changes in fuselage height, roll angle, pitch angle, yaw angle, and number of gait switching are determined by the quadruped robot's attitude control parameters, joint speed limits, joint angle limits, inspection speed, and gait library switching rules.

[0015] A legged robot inspection and navigation system for confined spaces includes: The multi-source sensing module is used to acquire environmental perception data of the confined inspection space where the quadruped robot is located, as well as the body state data of the quadruped robot. The cross-section constraint mapping module is used to determine the extension direction of the inspection channel based on environmental perception data, and generate multiple continuously arranged cross-section constraint units along the extension direction of the inspection channel. The dynamic attitude envelope generation module is used to generate dynamic attitude envelopes corresponding to each section constraint unit based on the body state data, candidate fuselage attitude and candidate gait. The dynamic attitude envelope includes fuselage occupancy envelope and swing leg sweep envelope. The foot safety window calculation module is used to calculate the foot safety window within each cross-sectional constraint unit; The joint planning module is used to generate candidate passage postures based on cross-sectional constraint units, dynamic attitude envelopes and foot landing safety windows, and to determine the target passage posture from the candidate passage postures that meets the fuselage passage conditions, leg swing conditions and foot landing conditions. The inspection viewpoint compensation module is used to determine the viewpoint accessibility of the inspection sensor to the object to be inspected based on the target's passage posture, and to generate viewpoint compensation actions when the viewpoint accessibility does not meet the conditions. The navigation control module is used to generate an inspection navigation trajectory based on the target's walking posture, foot landing safety window, and viewpoint compensation action, and to control the quadruped robot to perform inspections according to the inspection navigation trajectory. The retreat control module is used to generate a retreat path based on the passed section constraint units, the recorded safe stopping points, and the retreatable postures when there are no candidate passing postures that meet the fuselage passing conditions, leg swing conditions, and foot landing conditions in the forward section constraint unit, or when the forward section constraint unit does not meet the turning conditions.

[0016] In summary, the present invention has the following main beneficial effects: Compared with existing technologies, this application generates continuously arranged cross-sectional constraint units along the inspection channel extension direction of the narrow inspection space, and incorporates ground height, lateral clearance, top clearance, lateral protrusion position, and orientation of the object to be inspected into the same cross-sectional constraint system. This enables the quadruped robot to no longer rely solely on two-dimensional paths or ordinary ground elevation information for navigation, but to perform segmented constraint analysis on spatial constraints such as side walls, top components, cable trays, and pipeline supports in narrow spaces. This achieves the purpose of identifying areas with limited body, limited leg swing, and limited passage posture in narrow spaces in advance, reducing the risk of the robot getting stuck, colliding, or unable to continue moving forward after entering narrow sections.

[0017] This application generates a dynamic posture envelope based on the body state data, candidate body postures, and candidate gaits. It also calculates the foot-landing safety window within each cross-sectional constraint unit, enabling joint judgment of the body occupancy envelope, swing leg sweep envelope, and foot-landing area under the same spatial constraints. This avoids simplifying the quadruped robot into a fixed contour or point mass for obstacle avoidance. Consequently, the system can filter out target walking postures from candidate walking postures that simultaneously satisfy the body passage conditions, leg swing conditions, and foot-landing conditions, thereby improving the quadruped robot's stability and gait feasibility in low, narrow spaces with many lateral protrusions.

[0018] This application, after the target's passage posture meets the passage conditions, continues to determine the accessibility of the inspection sensor's viewpoint to the object to be inspected. When the observation ray is blocked or the observation distance does not meet the conditions, a viewpoint compensation action is generated. At the same time, when the front section constraint unit cannot pass or does not meet the turning conditions, a retreat path is generated based on the safe dwell point and the retreatable posture. This makes the inspection navigation not only focus on whether the robot can pass, but also on whether the object to be inspected can be effectively observed and whether the robot can safely exit in the non-turnable section, thereby reducing the risk of missed detections, false detections, and loss of control in confined spaces. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] refer to Figure 1 A method for inspection and navigation of a legged robot in confined spaces, comprising the following steps: S1: Acquire environmental perception data of the confined inspection space where the quadruped robot is located, as well as the body state data of the quadruped robot; S2: Determine the extension direction of the inspection channel based on the environmental perception data, and generate multiple continuously arranged cross-sectional constraint units along the extension direction of the inspection channel. S3: Based on the body state data, candidate fuselage attitude and candidate gait, generate dynamic attitude envelopes corresponding to each cross-section constraint unit. The dynamic attitude envelopes include fuselage occupancy envelopes and swing leg sweep envelopes. S4: Calculate the foot landing safety window within each cross-sectional constraint element; S5: Generate candidate passage postures based on cross-sectional constraint units, dynamic attitude envelopes and foot landing safety windows, and determine the target passage posture from the candidate passage postures that meets the fuselage passage conditions, leg swing conditions and foot landing conditions. S6: Determine the viewpoint reachability of the inspection sensor for the object to be inspected based on the target's passage posture, and generate a viewpoint compensation action when the viewpoint reachability does not meet the conditions. S7: Generates an inspection navigation trajectory based on the target's passage posture, foot landing safety window, and viewpoint compensation action, and controls the quadruped robot to perform inspections according to the inspection navigation trajectory; S8: If there are no candidate passing postures that meet the conditions for fuselage passage, leg swing, and foot landing in the front section constraint unit, or if the front section constraint unit does not meet the turning condition, generate a retreat path based on the section constraint units that have been passed, the recorded safe stopping points, and the retreatable postures.

[0022] The confined space can be a utility tunnel, cable trench, equipment mezzanine, underground passage, equipment cabinet passage, or other inspection space with side walls, top components, pipeline supports, cable trays, valve boxes, or protruding supports. The quadruped robot is equipped with a lidar, depth camera, inspection sensors, inertial measurement unit, joint encoder, and foot contact detection components. The inspection sensors are one or more of visible light cameras, thermal imaging cameras, acoustic sensors, or local magnification cameras. The objects to be inspected include instruments, valves, cable joints, pipeline connections, support connections, leak observation points, or equipment nameplates.

[0023] After entering a confined inspection space, the quadruped robot acquires environmental perception data and body state data. Environmental perception data includes 3D point cloud data, depth image data, and the location information of the object to be inspected. Body state data includes body posture, joint states, and foot contact states. Body posture is obtained by an inertial measurement unit and a robot state estimator, and includes at least body height, roll angle, pitch angle, and yaw angle. Joint states are obtained by encoders at each leg joint, including at least the angles of the hip, knee, and ankle joints. Foot contact states are obtained by foot force sensors, joint torque estimates, or foot contact switches, used to determine the supporting leg and the swinging leg.

[0024] Environmental perception data is uniformly expressed in the robot's body coordinate system or the local coordinate system of the inspection channel. When 3D point cloud data comes from LiDAR and depth image data comes from a depth camera, both are transformed to the same coordinate system based on sensor extrinsic parameters. Sensor extrinsic parameters include rotation and translation relationships, which are determined by the calibration process after robot assembly. The location information of the object to be inspected comes from a pre-established map of the object to be inspected, or from the target identification results of instruments, valves, cable connectors, or nameplates during the robot's inspection process. For the object to be inspected obtained through target identification, its 3D orientation is determined based on the depth value of the target's circumscribed region in the depth image or the corresponding point cloud position.

[0025] In this embodiment, the confined inspection space is not directly represented as a regular two-dimensional grid map, nor is it simply represented as a ground elevation map. Instead, multiple continuously arranged cross-sectional constraint units are generated along the extension direction of the inspection channel. These cross-sectional constraint units are used to perform spatial intersection judgments with the quadruped robot's body occupancy envelope, the swing leg sweep envelope, and the observation rays of the inspection sensors, respectively.

[0026] Specifically, the 3D point cloud data undergoes coordinate unification and point set partitioning to obtain ground point sets, sidewall point sets, top point sets, and protrusion point sets. Point set partitioning is based on point cloud height, point cloud normal vector direction, and the point cloud's position relative to the inspection channel centerline. The ground point set consists of points located within the foot-accessible height range where the vertical component of the normal vector is greater than its horizontal component; the sidewall point set consists of points located on both sides of the inspection channel where the horizontal component of the normal vector is greater than its vertical component; the top point set consists of points located above the fuselage foundation envelope, forming the top clearance boundary; and the protrusion point set consists of points extending from the sidewall, top, or ground boundary into the inspection channel, reducing the lateral clearance or top clearance of the corresponding cross-sectional constraint unit. Protrusion point sets correspond to cable trays, pipelines, supports, valve boxes, low beams, or temporary storage items.

[0027] The extension direction of the inspection channel is determined based on the distribution of the sidewall point set within the horizontal projection plane. Specifically, the main direction of the sidewall point set's distribution within the horizontal projection plane is extracted, and the direction with the highest consistency in the sidewall point set's extension direction is determined as the extension direction of the inspection channel. When the sidewalls on site are discontinuous, the sidewall point set within a set distance range in front of the robot is selected for segmented calculation, and the continuity of the inspection channel's extension direction is corrected based on the angle between adjacent segment directions. The set distance range is determined by the effective detection range of the lidar and the robot's current inspection speed.

[0028] After determining the extension direction of the inspection channel, multiple cross-sectional planes are established along this extension direction at a set sampling interval. Each cross-sectional plane is perpendicular to the extension direction of the inspection channel. The sampling interval is determined by the single-step length of the current candidate gait in the quadruped robot gait library, and is not greater than the single-step length, so that the constraint units of adjacent cross-sections can cover changes in foot placement and changes in body posture.

[0029] Each cross-sectional plane is mapped to ground point sets, sidewall point sets, top point sets, and protrusion point sets to obtain each cross-sectional constraint unit. When mapping the cross-sectional plane to these sets, points within a preset thickness range before and after the cross-sectional plane are selected as the point cloud data for that constraint unit. The preset thickness is determined by the point cloud sampling density and a set sampling interval. This process ensures that even if the actual point cloud does not strictly fall on the cross-sectional plane, the geometric constraints of the corresponding constraint unit can be determined using the point cloud within the neighborhood of the cross-sectional plane.

[0030] Each cross-sectional constraint element includes ground height, lateral clearance, top clearance, location of lateral protrusions, and orientation of the object to be inspected. Ground height is determined by statistical values ​​of the heights of ground points within the cross-sectional plane's neighborhood; lateral clearance is determined by the minimum lateral distance between the point sets on both sides of the cross-section or the inner boundaries of lateral protrusions; top clearance is determined by the minimum vertical distance between the ground height of the cross-section and the lower boundary of the top point set or the top protrusion; the location of lateral protrusions is determined by the projected position of the protrusion point set relative to the cross-sectional plane. The orientation of the object to be inspected is obtained by projecting the object located between two adjacent cross-sectional planes onto the corresponding cross-sectional constraint element.

[0031] After generating the cross-sectional constraint elements, a dynamic attitude envelope is generated based on the quadruped robot's body dimensions, joint states, candidate body postures, and candidate gaits. Candidate body postures include candidate body height, candidate roll angle, candidate pitch angle, and candidate yaw angle. Candidate gaits are provided by a gait library supported by the robot controller, which records the swing leg, supporting leg, foot trajectory, stride length range, and body height range for each gait. Candidate gaits include normal gait, low-profile gait, narrow stride gait, short gait, lateral crossing gait, and backward gait.

[0032] The dynamic attitude envelope includes the body occupancy envelope and the swing leg sweep envelope. The body occupancy envelope is established based on the quadruped robot's body dimensions, the mounting shape of the inspection sensors, and the boundary of the legs in the folded-back state. When the inspection sensors protrude outside the body, the mounting shape of the inspection sensors is included in the body occupancy envelope. Subsequently, based on candidate body height, candidate roll angle, candidate pitch angle, and candidate yaw angle, the position and attitude of the body occupancy envelope within the cross-sectional constraint unit are transformed to obtain the actual body occupancy range under the candidate passage attitude.

[0033] The swing leg sweep envelope is determined based on joint states and candidate gait. First, the supporting leg and swing leg are determined based on foot contact states. Then, the swing leg's swing initiation point, swing end point, and swing trajectory are determined based on candidate gait. The swing trajectory is provided by the robot's gait controller or obtained by interpolating the swing initiation point, swing end point, and intermediate foot lift height. The spatial shape of the leg links is sampled along the swing trajectory, and the spatial regions traversed by the leg links during the sampling process are merged to form the swing leg sweep envelope.

[0034] The fuselage occupancy envelope and swing leg sweep envelope are mapped onto the coordinate system of the corresponding section constraint element to form a dynamic attitude envelope corresponding to that section constraint element. This dynamic attitude envelope changes with the candidate fuselage attitude and candidate gait, and is not equivalent to a fixed length, width, and height profile. Through this processing, the limitations imposed by low-profile components on the fuselage occupancy envelope, the limitations imposed by lateral protrusions on the swing leg sweep envelope, and whether the foot landing area can support the corresponding gait are determined.

[0035] The foot landing safety window is calculated within each cross-sectional constraint element. The foot landing safety window is a suitable landing area that simultaneously satisfies the requirements of foot support area, ground slope, height difference limit, lateral clearance, and swing leg sweep clearance. Specifically, the candidate landing area for each foot is determined based on the foot contact state and candidate gait. The candidate landing area is located within the corresponding foot working space and is constrained by the stride length range and lateral stride range of the candidate gait.

[0036] For each candidate landing area, first eliminate areas that overlap with side walls, lateral protrusions, and bottom obstacles; then calculate the ground slope and elevation difference based on the point cloud height changes within the candidate landing area; finally, determine whether the foot support area, ground slope, elevation difference, lateral clearance, and swing leg sweep clearance meet the requirements.

[0037] The foot support area is determined based on the overlap area between the projected area of ​​the foot contact surface and the contactable ground area within the candidate landing area. When the overlap area is not less than a set support area threshold, the candidate landing area is determined to meet the foot support area requirement. When the ground slope does not exceed a set slope threshold, the elevation difference does not exceed a set elevation difference threshold, the foot support area is not less than a set support area threshold, the distance between the foot boundary and the sidewall and lateral protrusions is not less than a set lateral clearance threshold, and the distance between the swing leg sweep envelope and the sidewall and lateral protrusions is not less than a set swing clearance threshold, the area is designated as a foot landing safety window.

[0038] The aforementioned thresholds are all determined by the robot's structural parameters, controller parameters, and sensor parameters. The support area threshold is determined by the foot contact surface area and foot contact stability conditions; the lateral clearance threshold is determined by the minimum avoidance distance from the foot boundary to the sidewall and lateral protrusions; the swing clearance threshold is determined by the minimum avoidance distance from the outer boundary of the swing leg's sweep envelope to the sidewall and lateral protrusions; the slope threshold and height difference threshold are jointly determined by the quadruped robot's center of gravity projection range, leg workspace, and joint limits in a static support state. These parameters are determined by the robot's structural parameter table, foot contact surface dimensions, joint limit parameters, gait library parameters, and the controller's allowable stable support conditions.

[0039] After obtaining the cross-sectional constraint element, dynamic attitude envelope, and foot landing safety window, candidate passing attitudes are generated. Each set of candidate passing attitudes includes candidate fuselage height, candidate roll angle, candidate pitch angle, candidate yaw angle, and candidate gait. For the same cross-sectional constraint element, available candidate gaits are first selected from the gait library, and then multiple sets of candidate passing attitudes are generated by combining the fuselage height range, roll angle range, pitch angle range, and yaw angle range.

[0040] Three types of judgments are performed on each group of candidate passage postures. The first type is the fuselage passage condition judgment, which involves interfering with the lateral clearance width, top clearance height, and lateral protrusion position of the cross-sectional constraint unit corresponding to the candidate passage posture; if the fuselage passage envelope spatially intersects with the sidewall, top, or lateral protrusion, the candidate passage posture does not meet the fuselage passage condition. The second type is the leg swing condition judgment, which involves interfering with the swing leg sweep envelope corresponding to the candidate passage posture with the sidewall, top, lateral protrusion, and bottom obstacles in the cross-sectional constraint unit; if spatial intersects, the candidate passage posture does not meet the leg swing condition. The third type is the foot landing condition judgment, which involves matching each foot landing point corresponding to the candidate passage posture with the corresponding foot landing safety window; if each foot landing point is located within the corresponding foot landing safety window, the foot landing condition is met.

[0041] When the fuselage occupancy envelope does not interfere, the swing leg sweep envelope does not interfere, and each foot landing point is within the corresponding foot landing safety window, the candidate passage posture is determined as the candidate passage posture that satisfies the fuselage passage condition, the leg swing condition, and the foot landing condition. If there are multiple candidate passage postures that meet the conditions, the selection is based on the continuity of postures between adjacent sections, the number of gait switching times, the foot landing safety window margin, and the accessibility of the inspection viewpoint.

[0042] For candidate passage postures that meet the conditions for fuselage passage, leg swing, and foot landing, the accessibility of the inspection sensor's viewpoint is further determined. First, the sensor pose within the corresponding cross-sectional constraint unit is determined based on the candidate passage postures. The sensor pose is jointly determined by the fuselage pose, sensor mounting extrinsic parameters, and candidate fuselage pose. Then, based on the sensor pose and the orientation of the object to be inspected, an observation ray is generated pointing from the inspection sensor to the object to be inspected.

[0043] The intersection judgment objects of the observation ray include the sidewall boundary, top boundary, lateral protrusion boundary, and fuselage occupancy envelope boundary in the cross-sectional constraint unit, as well as the fuselage occupancy envelope boundary under the current candidate passage attitude. It determines whether the observation ray intersects with the above intersection judgment objects and calculates the observation distance between the inspection sensor and the object to be inspected. When the observation ray does not intersect with the sidewall boundary, top boundary, lateral protrusion boundary, or fuselage occupancy envelope boundary, and the observation distance is within the set observation distance range, the viewpoint accessibility is determined to meet the inspection requirements. When the observation ray intersects with the sidewall boundary, top boundary, lateral protrusion boundary, or fuselage occupancy envelope boundary, or the observation distance is not within the set observation distance range, a viewpoint compensation action is generated.

[0044] The observation distance range is determined by the effective imaging distance, focusing distance, minimum identifiable size of the object to be inspected, image resolution, and sensor calibration parameters of the inspection sensor. For visible light cameras, the observation distance range is determined by the camera focal length, pixel size, actual size of the object to be inspected, and the lowest pixel percentage of the object in the image; for thermal imaging sensors, the observation distance range is determined by the thermal imaging resolution, temperature measurement distance, and size of the object to be inspected. The observation distance range is provided in the sensor calibration file or equipment parameter table.

[0045] The viewpoint compensation actions include at least one of the following: stationary observation, body height adjustment, body yaw adjustment, and retrospective observation. Stationary observation involves controlling the quadruped robot to stop moving in front of or behind the corresponding cross-sectional constraint unit while maintaining its current body posture for observation. Body height adjustment involves adjusting the body height for observation while ensuring the foot landing point remains within the foot landing safety window. Body yaw adjustment involves adjusting the body yaw angle for observation while ensuring that neither the body occupancy envelope nor the swing leg sweep envelope interferes with the corresponding cross-sectional constraint unit. Retrospective observation involves controlling the quadruped robot to pass through the corresponding cross-sectional constraint unit and then using the sensor pose located behind the constraint unit to observe the object to be inspected.

[0046] When generating viewpoint compensation actions, the poses of the compensation sensors corresponding to the aforementioned actions are calculated, and observation rays are regenerated based on these poses. When the observation ray corresponding to the compensation sensor pose does not intersect with the sidewall boundary, top boundary, lateral protrusion boundary, or fuselage occupancy envelope boundary, and the corresponding observation distance is within the set observation distance range, the action corresponding to that compensation sensor pose is determined as a viewpoint compensation action. If multiple compensation actions meet the conditions, the execution order is determined according to the action amplitude, dwell time, and impact on the continuity of subsequent cross-section candidate passage attitudes.

[0047] When generating the inspection navigation trajectory, the candidate passage posture sequence for each section constraint unit is determined according to the arrangement order of the section constraint units. The continuity of candidate passage postures between adjacent section constraint units is assessed. This continuity assessment includes changes in fuselage height, roll angle, pitch angle, yaw angle, and the number of gait transitions. The allowable ranges are determined by the quadruped robot controller's attitude control parameters, joint speed limits, joint angle limits, inspection speed, and gait library switching rules. If the candidate passage postures between adjacent section constraint units do not meet the above allowable ranges, the corresponding candidate passage posture sequence will not be included in the inspection navigation trajectory.

[0048] For candidate travel posture sequences that meet the continuity criteria and include viewpoint compensation actions, an association is established between the candidate travel posture sequence and the corresponding object to be inspected. The association includes the cross-sectional constraint unit where the object to be inspected is located, the position where the viewpoint compensation action is performed, the pose of the compensation sensor, and the corresponding observation ray. Subsequently, an inspection navigation trajectory is generated based on the associated candidate travel posture sequence, the foot landing safety window, and the viewpoint compensation action. The inspection navigation trajectory includes the robot's center trajectory, body posture sequence, gait sequence, foot landing point sequence, viewpoint compensation action sequence, and sensor observation time.

[0049] Before the quadruped robot enters the current narrow inspection section, it is determined whether the continuous cross-sectional constraint unit ahead meets the turning condition. The turning condition is: within the same cross-sectional constraint unit, there exists a candidate passage posture that allows the quadruped robot to complete the set yaw angle change, and the fuselage occupancy envelope and swing leg sweep envelope corresponding to the candidate passage posture do not interfere with the cross-sectional constraint unit.

[0050] The yaw angle variation is determined by the angle between adjacent channels in the inspection route, the yaw angle required for a turn-around maneuver, or the yaw angle required for a safe turn before retreating. When the inspection task requires the robot to complete a turn-around within the corresponding cross-sectional constraint unit, the yaw angle variation is set to correspond to the yaw angle required for the turn-around maneuver. When the inspection task requires the robot to enter an adjacent channel from the current channel, the yaw angle variation is set to correspond to the angle between the extension directions of the adjacent channels. When the inspection task only requires the robot to adjust the observation direction within the cross-section, the yaw angle variation is set to correspond to the fuselage yaw angle variation required for the inspection sensor to obtain the observation ray of the object to be inspected.

[0051] When a continuous section constraint unit does not meet the turning conditions, a safe stopping point is determined on the entrance side of that continuous section constraint unit. The section constraint unit corresponding to the safe stopping point simultaneously satisfies the fuselage passage condition, leg swing condition, foot landing condition, and turning condition. The safe stopping point, the candidate passage posture adopted when entering the continuous section constraint unit, and the reversible posture corresponding to each passed section constraint unit are recorded. The reversible posture is recorded in one-to-one correspondence with the passed section constraint units. When generating the reversal path, the corresponding reversible postures are called sequentially in the reverse order of the passed section constraint units.

[0052] When the robot detects that there are no candidate passage postures that satisfy the body passage conditions, leg swing conditions, and foot landing conditions in the cross-sectional constraint unit ahead within a narrow inspection section, or when the foot contact state indicates foot slippage, it calls the recorded safe stopping points and reversible postures to generate a reversal path. Foot slippage is determined jointly based on the foot contact state, the relative displacement of the foot, and the actual posture change of the body. When the foot is in contact and the displacement of the foot relative to the ground point set exceeds the allowable displacement determined by the foot contact surface size and positioning error, it is determined that the foot is slipping; or, when the robot executes a gait control command, if the actual posture change of the body is inconsistent with the expected posture change corresponding to the gait control command, and the inconsistency persists for multiple consecutive control cycles, it is determined that the foot contact state indicates foot slippage.

[0053] The retreat path is generated in the reverse order of the cross-sectional constraint units that have been passed, and the quadruped robot is controlled to exit the corresponding narrow inspection section by a retreatable posture until it reaches the safe stopping point or the turnable area on the entrance side.

[0054] During the inspection process, if the environmental perception data is updated, causing changes in the lateral clearance, top clearance, position of lateral protrusions, or orientation of the object to be inspected for any cross-sectional constraint unit, the candidate passage attitude screening and viewpoint accessibility judgment are re-executed. If no candidate passage attitude that meets the fuselage passage conditions, leg swing conditions, and foot landing conditions is found after re-screening, retrace control is triggered; if a candidate passage attitude that meets the conditions is found after re-screening, the inspection navigation trajectory is updated based on the updated cross-sectional constraint unit, dynamic attitude envelope, foot landing safety window, and viewpoint compensation action.

[0055] Example 2 This embodiment provides a quadruped robot inspection and navigation system for confined spaces. The system is deployed within the quadruped robot's computing unit, or it can be implemented collaboratively by the robot's computing unit and a remote inspection management terminal. The system includes a multi-source perception module, a cross-sectional constraint mapping module, a dynamic posture envelope generation module, a foothold safety window calculation module, a joint planning module, an inspection viewpoint compensation module, a navigation control module, and a retreat control module.

[0056] The multi-source perception module is used to acquire environmental perception data and the robot's body state data within the confined inspection space. It receives 3D point cloud data from the LiDAR, depth image data from the depth camera, position information of the object to be inspected from the inspection sensors or object recognition program, and body posture, joint state, and foot contact state from the inertial measurement unit, joint encoder, and foot contact detection components. The multi-source perception module outputs the environmental perception data to the cross-section constraint mapping module and the body state data to the dynamic posture envelope generation module and the joint planning module.

[0057] The cross-section constraint mapping module determines the extension direction of the inspection channel based on environmental perception data and generates multiple consecutively arranged cross-section constraint units along this direction. The module includes a point set partitioning unit, a channel direction determination unit, a cross-section plane establishment unit, and a cross-section constraint generation unit. The point set partitioning unit divides the 3D point cloud data into ground point sets, sidewall point sets, top point sets, and protrusion point sets. The channel direction determination unit determines the extension direction of the inspection channel based on the distribution of the sidewall point sets within the horizontal projection plane. The cross-section plane establishment unit establishes multiple cross-section planes along the inspection channel's extension direction at a set sampling interval. The cross-section constraint generation unit generates cross-section constraint units based on the correspondence between each cross-section plane and the ground point sets, sidewall point sets, top point sets, protrusion point sets, and the location information of the object to be inspected. The cross-section constraint mapping module outputs these cross-section constraint units to the dynamic attitude envelope generation module, the landing safety window calculation module, the joint planning module, and the inspection viewpoint compensation module.

[0058] The dynamic attitude envelope generation module generates dynamic attitude envelopes corresponding to each cross-sectional constraint unit based on the quadruped robot's body dimensions, joint states, candidate body postures, and candidate gaits. The module calls the robot's structural parameter table and gait library to determine the basic body envelope, candidate body postures, candidate gaits, supporting legs, swinging legs, and swing trajectories, generating the body occupancy envelope and swinging leg sweep envelope. The module then outputs the dynamic attitude envelopes to the joint planning module and the inspection viewpoint compensation module.

[0059] The foot landing safety window calculation module is used to calculate the foot landing safety window within each cross-sectional constraint unit. Based on the foot contact state and candidate gait, the module determines the candidate foot landing area for each foot. Then, based on the ground slope, elevation difference, foot support area, distance from the foot boundary to the sidewall and lateral protrusions, and distance from the swing leg sweep envelope to the sidewall and lateral protrusions, the module determines the foot landing safety window. The foot landing safety window calculation module outputs the foot landing safety windows to the joint planning module and navigation control module.

[0060] The joint planning module generates candidate passage postures based on cross-sectional constraint units, dynamic posture envelopes, and foot landing safety windows, and determines whether the quadruped robot meets the body passage conditions, leg swing conditions, and foot landing conditions under the candidate passage postures. The joint planning module performs interference checks on the body occupancy envelope and cross-sectional constraint units, and on the swing leg sweep envelope and cross-sectional constraint units, matching the foot landing point with the foot landing safety window. The joint planning module outputs the candidate passage postures that meet the body passage conditions, leg swing conditions, and foot landing conditions to the inspection viewpoint compensation module and the navigation control module.

[0061] The inspection viewpoint compensation module is used to determine the viewpoint accessibility of candidate passing attitudes that meet the passage conditions. Based on the candidate passing attitudes, the module determines the sensor pose of the inspection sensor within the corresponding cross-sectional constraint unit. It then generates an observation ray based on the sensor pose and the orientation of the object to be inspected. The module determines whether the observation ray intersects with the sidewall boundary, top boundary, lateral protrusion boundary, or fuselage occupancy envelope boundary within the cross-sectional constraint unit, and calculates the observation distance between the inspection sensor and the object to be inspected. When the observation ray is obstructed or the observation distance is outside the set observation distance range, the module generates compensated sensor poses corresponding to pause observation actions, fuselage altitude adjustment actions, fuselage yaw adjustment actions, or retrospective observation actions. It then outputs the actions corresponding to the compensated sensor poses that meet the requirements of unobstructed observation rays and observation distance to the navigation control module.

[0062] The navigation control module generates an inspection navigation trajectory based on candidate walking postures, foot landing safety windows, and viewpoint compensation actions. The module outputs the robot's center trajectory, body posture sequence, gait sequence, foot landing sequence, viewpoint compensation action sequence, and sensor observation times, and controls the quadruped robot to perform inspections according to the navigation trajectory. During execution, the navigation control module continuously receives updated environmental perception data and body state data from the multi-source perception module, and triggers the joint planning module to recalculate candidate walking postures when the cross-sectional constraint units change.

[0063] The retreat control module generates a retreat path based on the already passed section constraint units, recorded safe stopping points, and retreatable postures when no candidate passage postures satisfying the robot's passage, leg swing, and foot landing conditions exist in the forward section constraint unit, or when the forward section constraint unit does not meet the turning conditions. Before entering the current narrow inspection section, the retreat control module records the safe stopping point, the candidate passage postures used when entering continuous section constraint units, and the retreatable postures corresponding to each passed section constraint unit. Each retreatable posture is saved in a one-to-one correspondence with each passed section constraint unit. When the foot contact status indicates foot slippage or the forward section is impassable, the retreat control module retrieves the recorded data, generates a retreat path in reverse order of the section constraint units, and outputs the retreat path to the navigation control module for execution.

[0064] Example 3 In one specific inspection process, a quadruped robot performs inspections within cable trenches or utility tunnels, with low-profile cable trays and lateral pipeline supports in front of the robot. The robot first generates cross-sectional constraint units along the inspection path, and records the top clearance formed by the low-profile cable trays, the position of lateral protrusions formed by the lateral pipeline supports, and the orientation of the object to be inspected within the corresponding cross-sectional constraint units.

[0065] Subsequently, the robot generates dynamic posture envelopes corresponding to the normal gait, low-profile gait, and narrow-stride gait. If the body occupancy envelope interferes with the top clearance in the normal gait, the candidate passage posture corresponding to the normal gait is eliminated. If the swing leg sweep envelope interferes with the lateral pipeline support in the low-profile gait, the candidate passage posture corresponding to the low-profile gait is eliminated. If the body occupancy envelope does not interfere with the top clearance in the narrow-stride gait, the swing leg sweep envelope does not interfere with the lateral protrusions, and the foot landing points are all within the foot landing safety window, then the candidate passage posture corresponding to the narrow-stride gait is retained.

[0066] Viewpoint accessibility is assessed for the retained candidate passage postures. If the robot passes through in a narrow stride and low posture, and the observation ray of the inspection sensor pointing to the object to be inspected is blocked by the body's occupancy envelope or lateral pipeline supports, a stopping observation action is generated in front of the corresponding cross-sectional constraint unit, or a looking-back observation action is generated after passing through the cross-sectional constraint unit. Only when the observation ray corresponding to the compensated sensor pose does not intersect with the side wall boundary, top boundary, lateral protrusion boundary, and body occupancy envelope boundary, and the observation distance is within the set observation distance range, is this action written into the inspection navigation trajectory.

[0067] If the robot continues forward and enters a narrow section where it cannot turn around, this embodiment records the safe stopping point and the reversible posture at the entrance side. When subsequent cross-sectional constraint units fail to provide a candidate passage posture that satisfies the conditions for body passage, leg swing, and foot landing due to temporary obstacles, the robot does not attempt to turn in place within the narrow section. Instead, it calls upon the recorded reversible posture and retreats to the safe stopping point in the reverse order of the cross-sectional constraint units it has already passed.

[0068] When the robot detects changes in the position of lateral pipeline supports, temporary obstacles under low-profile cable trays, or changes in the orientation of the object to be inspected due to updates in the robot's observation angle during inspection, the corresponding cross-sectional constraint units are updated, and candidate passage posture screening and viewpoint accessibility judgment are re-executed. If, after the update, there are still candidate passage postures that meet the conditions for body passage, leg swing, and foot landing, the inspection navigation trajectory is updated; otherwise, if there are no candidate passage postures that meet the conditions, the retreat control module is invoked to generate a retreat path.

[0069] The above process is not based solely on controlling the robot's forward movement according to the extension direction of the utility tunnel, nor solely on judging whether the terrain is passable based solely on the ground elevation map. Instead, under the cross-sectional constraints of a narrow space, the robot's occupancy envelope, the swing leg sweep envelope, the foot landing safety window, the inspection sensor observation rays, and the retreat path are all processed within the same inspection and navigation process.

[0070] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for inspection and navigation of a legged robot in confined spaces, characterized in that, Includes the following steps: S1: Acquire environmental perception data of the confined inspection space where the quadruped robot is located, as well as the body state data of the quadruped robot; S2: Determine the extension direction of the inspection channel based on the environmental perception data, and generate multiple continuously arranged cross-sectional constraint units along the extension direction of the inspection channel. S3: Based on the body state data, candidate fuselage attitude and candidate gait, generate dynamic attitude envelopes corresponding to each cross-section constraint unit. The dynamic attitude envelopes include fuselage occupancy envelopes and swing leg sweep envelopes. S4: Calculate the foot landing safety window within each cross-sectional constraint element; S5: Generate candidate passage postures based on cross-sectional constraint units, dynamic attitude envelopes and foot landing safety windows, and determine the target passage posture from the candidate passage postures that meets the fuselage passage conditions, leg swing conditions and foot landing conditions. S6: Determine the viewpoint reachability of the inspection sensor for the object to be inspected based on the target's passage posture, and generate a viewpoint compensation action when the viewpoint reachability does not meet the conditions. S7: Generates an inspection navigation trajectory based on the target's passage posture, foot landing safety window, and viewpoint compensation action, and controls the quadruped robot to perform inspections according to the inspection navigation trajectory; S8: If there are no candidate passing postures that meet the conditions for fuselage passage, leg swing, and foot landing in the front section constraint unit, or if the front section constraint unit does not meet the turning condition, generate a retreat path based on the section constraint units that have been passed, the recorded safe stopping points, and the retreatable postures.

2. The method for inspection and navigation of a legged robot in a confined space according to claim 1, characterized in that, Environmental perception data includes 3D point cloud data, depth image data, and location information of the object to be inspected; body state data includes fuselage attitude, joint state, and foot contact state. The cross-sectional constraint unit includes ground height, lateral clear width, top clear height, lateral protrusion position, and orientation of the object to be inspected, and is used to make spatial cross-judgments with the fuselage occupancy envelope, the swing leg sweep envelope, and the observation rays of the inspection sensors, respectively.

3. The method for inspection and navigation of a legged robot in a confined space according to claim 2, characterized in that, Multiple consecutively arranged cross-sectional constraint elements are generated along the extension direction of the inspection channel, including: The coordinates of the 3D point cloud data are unified and the point sets are divided to obtain the ground point set, side wall point set, top point set and protrusion point set; The extension direction of the inspection channel is determined based on the extended distribution of the side wall point set in the horizontal projection plane; Multiple cross-sectional planes are established along the extension direction of the inspection channel according to the set sampling interval. Each cross-sectional plane is perpendicular to the extension direction of the inspection channel. The sampling interval is set to be no greater than the single step length of the quadruped robot. Points within a preset thickness range before and after each cross-section plane are selected as point cloud data for the corresponding cross-section constraint unit. The preset thickness is determined by the point cloud sampling density and the set sampling interval. The ground height, lateral clearance, top clearance, and position of lateral protrusions are determined based on the point cloud data of the corresponding cross-sectional constraint unit. The object to be inspected, located between two adjacent cross-sectional planes, is projected onto the corresponding cross-sectional constraint unit to obtain the orientation of the object to be inspected.

4. The method for inspection and navigation of a legged robot in a confined space according to claim 3, characterized in that, The generation of dynamic pose envelopes includes: The basic envelope of the quadruped robot is established based on its body dimensions, the installation shape of the inspection sensors, and the boundary of its shape when its legs are folded up. Based on the candidate fuselage height, candidate roll angle, candidate pitch angle and candidate yaw angle, the basic fuselage envelope is transformed to obtain the fuselage occupancy envelope; The supporting leg and swinging leg are determined based on joint status and candidate gait, and the swing start point, swing end point and swing trajectory of each swinging leg are determined. The swing leg sweep envelope is generated based on the swing trajectory of each swing leg and the outer boundary of the corresponding leg link. The fuselage occupancy envelope and the swing leg sweep envelope are mapped to the coordinate system of the corresponding section constraint element to form the dynamic attitude envelope corresponding to that section constraint element.

5. The method for inspection and navigation of a legged robot in a confined space according to claim 4, characterized in that, The calculation for a foot-landing safety window includes: The candidate foot placement area is determined based on the foot contact state and candidate gait; Areas in the candidate landing area that overlap with side walls, lateral protrusions, and bottom obstacles are eliminated; Calculate the ground slope and elevation difference based on the point cloud height changes within the candidate landing area; The foot support area is determined by the overlap area between the projected area of ​​the foot contact surface and the ground contact area within the candidate foot landing area. The area where the ground slope does not exceed the set slope threshold, the height difference does not exceed the set height difference threshold, the foot support area is not less than the set support area threshold, the distance between the foot boundary and the side wall and lateral protrusions is not less than the set lateral gap threshold, and the distance between the swing leg sweep envelope and the side wall and lateral protrusions is not less than the set swing gap threshold is defined as the foot landing safety window. Among them, the set support area threshold is determined by the foot contact surface area and the foot contact stability condition; the set lateral clearance threshold is determined by the minimum avoidance distance from the foot boundary to the side wall and lateral protrusions; the set swing clearance threshold is determined by the minimum avoidance distance from the outer boundary of the swing leg sweep envelope to the side wall and lateral protrusions; and the set slope threshold and set height difference threshold are jointly determined by the center of gravity projection range, leg workspace and joint limits of the quadruped robot in static support state.

6. The method for inspection and navigation of a legged robot in a confined space according to claim 5, characterized in that, The target passing attitude is determined from the candidate passing attitudes, satisfying the fuselage pass conditions, leg swing conditions, and foot landing conditions, including: Multiple candidate flight attitudes are generated within the same cross-sectional constraint unit. Each candidate flight attitude includes candidate fuselage height, candidate roll angle, candidate pitch angle, candidate yaw angle, and candidate gait. Interference judgment is made between the fuselage occupancy envelope corresponding to each group of candidate passage attitudes and the lateral clear width, top clear height and lateral protrusion position of the cross-sectional constraint unit; Interference judgment is performed between the swing leg sweep envelope corresponding to each group of candidate passage postures and the side walls, top, lateral protrusions and bottom obstacles in the cross-sectional constraint unit; Match the foot landing point corresponding to each group of candidate passage postures with the foot landing safety window; When the fuselage occupancy envelope does not interfere, the swing leg sweep envelope does not interfere, and the landing point of each foot is within the corresponding foot landing safety window, the candidate passage posture is determined as the target passage posture.

7. The method for inspection and navigation of a legged robot in a confined space according to claim 6, characterized in that, The inspection sensor determines the viewpoint accessibility of the target object based on the target's travel attitude, and generates viewpoint compensation actions when the viewpoint accessibility conditions are not met, including: Determine the sensor pose of the inspection sensor within the corresponding cross-sectional constraint unit based on the target's passage posture; Based on the sensor pose and the orientation of the object to be inspected, an observation ray is generated pointing from the inspection sensor to the object to be inspected. Determine whether the observed ray intersects with the sidewall boundary, top boundary, lateral protrusion boundary, or fuselage occupancy envelope boundary in the corresponding cross-sectional constraint unit, and calculate the observation distance between the inspection sensor and the object to be inspected. When the observation ray intersects with the side wall boundary, top boundary, lateral protrusion boundary or fuselage occupancy envelope boundary, or when the observation distance is not within the set observation distance range, the corresponding compensation sensor poses for the pause observation action, fuselage height adjustment action, fuselage yaw adjustment action and retrospective observation action are generated respectively. When the observation ray corresponding to the pose of the compensation sensor does not intersect with the side wall boundary, top boundary, lateral protrusion boundary and fuselage occupancy envelope boundary, and the corresponding observation distance is within the set observation distance range, the action corresponding to the pose of the compensation sensor is determined as the viewpoint compensation action. The observation distance range is determined by the effective imaging distance of the inspection sensor, the focusing distance, the minimum identifiable size of the object to be inspected, the image resolution, and the sensor calibration parameters.

8. The method for inspection and navigation of a legged robot in a confined space according to claim 7, characterized in that, The pause observation action is to control the quadruped robot to stop moving on the front or rear side of the corresponding cross-sectional constraint unit and maintain the current body posture for observation; The aircraft height adjustment procedure involves adjusting the aircraft height while keeping the foot landing point within the foot landing safety window before conducting observation. The fuselage yaw adjustment action is to adjust the fuselage yaw angle and then observe it, provided that neither the fuselage occupancy envelope nor the swing leg sweep envelope interferes with the corresponding cross-sectional constraint element. The retrospective observation action involves controlling the quadruped robot to pass through the corresponding cross-sectional constraint unit, and then using the sensor pose located behind the corresponding cross-sectional constraint unit to observe the object to be inspected.

9. A method for inspection and navigation of a legged robot in a confined space according to claim 8, characterized in that, The generation of the inspection navigation trajectory includes: The target passage attitude sequence through each cross-section constraint unit is determined according to the arrangement order of the cross-section constraint units. The continuity judgment of the target's passage attitude between adjacent cross-section constraint units is performed. The continuity judgment includes the change in fuselage altitude, the change in roll angle, the change in pitch angle, the change in yaw angle, and the number of gait switching. Eliminate target passage posture sequences that do not meet the continuity judgment; Establish a connection between the target passage posture sequence that meets the continuity judgment and includes viewpoint compensation actions and the corresponding object to be inspected; The inspection navigation trajectory is generated based on the associated target passage posture sequence, foot landing safety window and viewpoint compensation action. The allowable ranges for changes in fuselage height, roll angle, pitch angle, yaw angle, and number of gait switching are determined by the quadruped robot's attitude control parameters, joint speed limits, joint angle limits, inspection speed, and gait library switching rules.

10. A legged robot inspection and navigation system for confined spaces, applicable to the legged robot inspection and navigation method for confined spaces as described in any one of claims 1-9, characterized in that, include: The multi-source sensing module is used to acquire environmental perception data of the confined inspection space where the quadruped robot is located, as well as the body state data of the quadruped robot. The cross-section constraint mapping module is used to determine the extension direction of the inspection channel based on environmental perception data, and generate multiple continuously arranged cross-section constraint units along the extension direction of the inspection channel. The dynamic attitude envelope generation module is used to generate dynamic attitude envelopes corresponding to each section constraint unit based on the body state data, candidate fuselage attitude and candidate gait. The dynamic attitude envelope includes fuselage occupancy envelope and swing leg sweep envelope. The foot safety window calculation module is used to calculate the foot safety window within each cross-sectional constraint unit; The joint planning module is used to generate candidate passage postures based on cross-sectional constraint units, dynamic attitude envelopes and foot landing safety windows, and to determine the target passage posture from the candidate passage postures that meets the fuselage passage conditions, leg swing conditions and foot landing conditions. The inspection viewpoint compensation module is used to determine the viewpoint accessibility of the inspection sensor to the object to be inspected based on the target's passage posture, and to generate viewpoint compensation actions when the viewpoint accessibility does not meet the conditions. The navigation control module is used to generate an inspection navigation trajectory based on the target's walking posture, foot landing safety window, and viewpoint compensation action, and to control the quadruped robot to perform inspections according to the inspection navigation trajectory. The retreat control module is used to generate a retreat path based on the passed section constraint units, the recorded safe stopping points, and the retreatable postures when there are no candidate passing postures that meet the fuselage passing conditions, leg swing conditions, and foot landing conditions in the forward section constraint unit, or when the forward section constraint unit does not meet the turning conditions.

Images