A Tunnel Path Planning Method and System Based on Clearance-Driven Spatial Curvature Constraints

Abstract

This invention proposes a tunnel path planning method and system based on clearance-driven spatial variable curvature constraints, relating to the field of autonomous navigation technology for robots in tunnels and underground spaces. It includes: constructing a clearance distance field based on a target tunnel grid map and defining a clearance penalty term; constructing a basic turning radius scale field based on the clearance distance field, and incorporating local target velocity field, attitude risk field, environmental parameter field, and dynamic obstacle spatiotemporal risk field for correction, resulting in a composite spatial variable curvature constraint field; constructing a position-related basic cost field based on the clearance penalty term and environmental traversability parameters, establishing a direction-sensitive cost field based on the direction risk penalty coefficient, and fusing them to obtain a comprehensive cost field; inputting the composite spatial variable curvature constraint field, minimum turning radius constraint, and comprehensive cost field into a path planner to obtain the optimal path. This invention balances traversability flexibility, path smoothness, and terrain adaptability, significantly improving the safety and efficiency of robot operations in confined spaces.

Description

Technical Field

[0001] This invention relates to the field of autonomous navigation technology for robots in tunnels and underground spaces, and in particular to a tunnel path planning method and system based on clearance-driven spatial variable curvature constraints. Background Technology

[0002] During inspection, maintenance, and material transportation, underground tunnel robots often need to navigate autonomously in environments with limited clearance, continuous slope variations, and localized temporary obstacles. In such environments, the quality of the path is affected not only by the distribution of obstacles and the passage width, but also by local speed requirements, slope attitude risks, and changes in steering constraints caused by changes in environmental conditions.

[0003] Existing path planning methods typically set the minimum turning radius or curvature constraint as a fixed parameter, or set a uniform steering capability based solely on static obstacle distance. This results in a lack of adaptability in path steering strategies for different spatial locations and driving conditions: in low-headroom, low-speed obstacle avoidance areas, a fixed large turning radius can easily lead to insufficient maneuverability; in high-headroom, high-speed areas, if a smaller expected turning radius or a more lenient curvature limit is still used, it can easily lead to excessive path bends and insufficient driving stability; when reverse slopes, cross slopes, or local dynamic obstacles coexist, a single geometric constraint is difficult to balance passability, ride comfort, and attitude safety. Summary of the Invention

[0004] To address the challenge that existing fixed curvature constraints cannot simultaneously meet the maneuverability requirements of narrow clearance areas and the smoothness requirements of open areas, this invention proposes a tunnel path planning method and system based on variable curvature constraints in clearance-driven spaces. This method balances traffic flexibility, path smoothness, and terrain adaptability, significantly improving the safety and efficiency of robots operating in confined spaces.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a tunnel path planning method based on clearance-driven spatial variable curvature constraints, comprising: Obtain the target tunnel grid map, calculate the Euclidean distance from each grid point to the nearest obstacle or wall boundary to obtain the clearance distance field; and define the clearance penalty term based on the clearance distance field, the safety radius, and the penalty band width. A basic turning radius scale field is constructed based on the clearance distance field, and the basic turning radius scale field is coupled and corrected according to the local target velocity field, attitude risk field, environmental parameter field and dynamic obstacle spatiotemporal risk field to obtain a composite spatial variable curvature constraint field. A location-related basic cost field is constructed based on the aforementioned airspace penalty term and environmental accessibility parameters; a direction-sensitive cost field is established based on the direction risk penalty coefficient; the basic cost field and the direction-sensitive cost field are fused to obtain a comprehensive cost field. The optimal path is obtained by inputting the composite spatial variable curvature constraint field, the robot's inherent minimum turning radius constraint, and the comprehensive cost field into the path planner.

[0006] Secondly, the present invention provides a tunnel path planning system based on clearance-driven spatial variable curvature constraints, comprising: The environment modeling module is configured to acquire a target tunnel grid map, calculate the Euclidean distance from each grid point to the nearest obstacle or wall boundary to obtain a clearance distance field, and define a clearance penalty term based on the clearance distance field, safety radius, and penalty band width. The composite spatial variable curvature constraint field construction module is configured to construct a basic turning radius scale field based on the clearance distance field, and to couple and correct the basic turning radius scale field according to the local target velocity field, attitude risk field, environmental parameter field and dynamic obstacle spatiotemporal risk field to obtain the composite spatial variable curvature constraint field. The cost field construction module is configured to construct a location-related basic cost field based on the airspace penalty term and environmental accessibility parameters; establish a direction-sensitive cost field based on the direction risk penalty coefficient; and fuse the basic cost field and the direction-sensitive cost field to obtain a comprehensive cost field. The path planning module is configured to input the composite spatial variable curvature constraint field, the robot's inherent minimum turning radius constraint, and the comprehensive cost field into the path planner to obtain the optimal path.

[0007] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the tunnel path planning method based on clearance-driven spatial variable curvature constraints described in the first aspect.

[0008] Fourthly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the tunnel path planning method based on clearance-driven spatial variable curvature constraints described in the first aspect.

[0009] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention centers on the global construction of a composite spatial variable curvature constraint field. It generates a continuous clearance representation based on a clearance distance field in the global space, and combines local target velocity, pitch / roll attitude risk, tunnel environmental parameters, and the spatiotemporal risk of optional dynamic obstacles to couple and correct the basic turning radius scale field, thus constructing a composite spatial variable curvature constraint field. Specifically, a position-dependent basic cost field characterizes the passage difficulty of candidate positions, a direction-sensitive cost field characterizes the attitude safety risk of the same position under different headings, and the composite spatial variable curvature constraint field determines the local expected turning radius and curvature regularization intensity of the same candidate state. These three elements collaboratively determine whether a candidate state is preferentially expanded and the allowed local turning amplitude during the expansion and cost accumulation process of the same position-heading state, thereby achieving path planning where turning capability adaptively changes with spatial position, driving state, and environmental risk.

[0010] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0011] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute a limitation thereof.

[0012] Figure 1 The main flowchart of a tunnel path planning method based on clearance-driven spatial variable curvature constraint provided in an embodiment of the present invention; Figure 2 A coupling diagram of the position-dependent basic cost field, the orientation-sensitive cost field, and the composite spatial variable curvature constraint field provided for embodiments of the present invention; Figure 3 A schematic diagram comparing path planning results in a simulated tunnel environment provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a composite spatial variable curvature constraint field provided in an embodiment of the present invention. Detailed Implementation

[0013] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0014] As mentioned in the background section, tunnels and underground spaces are typical complex and confined working environments, characterized by strong boundary constraints, narrow passageways, dense obstacle distribution, and significant terrain undulations. Among these, tunnel clearance refers to the minimum safe distance from each location within the robot's passage area to the nearest wall or obstacle; it is a core geometric parameter determining the robot's turning space and passage safety. In such scenarios, robot path planning must not only meet basic obstacle avoidance and accessibility requirements but also consider turning smoothness, minimum turning radius constraints, and pitch and roll safety when traveling along slopes.

[0015] Existing path planning methods generally employ fixed curvature constraints or fixed minimum turning radii for planning, using a uniform description of turning capability for open and narrow sections, which is difficult to adapt to the dynamic changes in tunnel clearance. For example, limiting turning behavior solely by preset fixed parameters can easily lead to traffic obstruction in narrow clearance areas due to excessively large turning radii, while in open clearance areas, the constraints are too tight, resulting in excessive path bends and insufficient smoothness. At the same time, the lack of coupling between clearance and terrain posture and turning capability makes it impossible to achieve robot posture optimization in complex terrain, failing to meet the engineering requirement of "able to turn in narrow places and smooth in wide places."

[0016] To address the aforementioned issues, this invention proposes a tunnel path planning method, system, medium, and device based on clearance-driven spatial variable curvature constraints. It focuses on the spatial adaptive construction mechanism of the curvature constraint parameters themselves, enabling the path to have different turning preferences in different clearance areas, thus simultaneously considering the passability in narrow areas and the smoothness in open areas. This invention maps the clearance information formed by obstacles and boundaries into a basic turning radius scale field that varies with spatial position. Velocity correction, attitude risk correction, joint environmental parameter correction, and / or dynamic obstacle risk correction are then superimposed on this basic turning radius scale field to construct a composite spatial variable curvature constraint field. This composite spatial variable curvature constraint field is then coupled with a position-related basic cost field and a direction-sensitive cost field in a unified solution model. This solves the problem that existing fixed curvature constraints cannot simultaneously adapt to both narrow and open areas, achieving a path generation effect of "turning in narrow areas, smoother in wide areas, long-distance obstacle coverage, and more stable attitude."

[0017] Example 1 like Figure 1 As shown in the figure, this embodiment discloses a tunnel path planning method based on the variable curvature constraint of the clearance-driven space, including the following steps: S1: Obtain the target tunnel grid map, calculate the Euclidean distance from each grid point to the nearest obstacle or wall boundary to obtain the clearance distance field; and define the clearance penalty term based on the clearance distance field, safety radius, and penalty band width. S2: Construct a basic turning radius scale field based on the clearance distance field, and couple and correct the basic turning radius scale field according to the local target velocity field, attitude risk field, environmental parameter field and dynamic obstacle spatiotemporal risk field to obtain a composite spatial variable curvature constraint field. S3: Construct a location-related basic cost field based on the aforementioned airspace penalty term and environmental accessibility parameters; establish a direction-sensitive cost field based on the direction risk penalty coefficient; merge the basic cost field and the direction-sensitive cost field to obtain a comprehensive cost field; S4: Input the composite spatial variable curvature constraint field, the robot's inherent minimum turning radius constraint, and the comprehensive cost field into the path planner to obtain the optimal path.

[0018] Next, combined Figure 2 This embodiment provides a detailed description of a tunnel path planning method based on clearance-driven spatial variable curvature constraints.

[0019] (I) Constructing a composite spatial variable curvature constraint field driven by clearance Based on the environmental occupancy map formed by obstacles and boundaries, the Euclidean distance from any point to the nearest obstacle or boundary is calculated to obtain the clearance distance field. This clearance distance field is then normalized and mapped onto the basic turning radius scale field. A larger turning radius scale corresponds to a larger local expected turning radius, resulting in a smoother path in that area; a smaller turning radius scale corresponds to a smaller local expected turning radius, allowing for more flexible turning in that area. The local expected turning radius refers to the ideal turning radius reference value that path planning should follow under the local environmental constraints of the current location.

[0020] First, construct a tunnel environment map. This can be achieved through measured point clouds, elevation maps, raster maps, or simulated terrain fields to obtain raster information about the tunnel environment. Both obstacle and boundary areas in the environment are considered as impassable zones. Obstacle zones include local obstacles such as construction machinery, maintenance equipment, protruding components, and accumulated debris, while boundary areas include wall boundaries. Binary obstacle masks are then created for both obstacle and boundary areas.

[0021] Then, based on the binary obstacle mask, the Euclidean distance from each grid point to the nearest obstacle or boundary is calculated to obtain the clearance distance field: ; in, Represents any grid point in the environment. Represents a set of obstacles and boundaries. This represents the Euclidean distance operator.

[0022] The physical meaning of the clearance distance field formula is as follows: the Euclidean distance from each grid point to the nearest obstacle or wall boundary is used as the passable safety margin at that location. The larger the distance, the greater the lateral obstacle avoidance margin and turning space of the robot at that location; the smaller the distance, the closer the location is to the wall or obstacle, and the higher the risk of collision, scratch or attitude correction failure.

[0023] The mathematical basis of this formula is the Euclidean distance transformation principle, which treats the set of obstacles and wall boundaries as an impassable set. For any point in free space, the minimum distance to the impassable set is calculated, thus obtaining the safe distance field in continuous or discrete space. Since both wall boundaries and obstacles in a tunnel environment restrict the robot's passage, it is reasonable to include both in the obstacle and boundary set.

[0024] This formula ensures the feasibility of motion in real-world scenarios: the path planner uses this clearance distance as a basis in the subsequent cost construction and curvature constraint construction processes, giving positions closer to obstacles or walls a higher penalty or more restricted motion preference, thus preventing the planned path from crossing obstacles or getting too close to the boundary from the source.

[0025] To enhance the differentiation of near-obstacle areas, a safety radius or penalty zone width parameter is introduced to convert clearance distance into a clearance penalty term or standardized clearance index. The smaller the clearance, the closer the location is to the obstacle and the higher the passage risk.

[0026] Specifically, a clearance penalty term is defined to characterize the intensity of the risk of wall contact or the penalty for proximity to obstacles: Let the safety radius be... The width of the penalty band is The net clearance penalty term is defined as follows: ; ; Among them, the clearance penalty when approaching obstacles or boundaries. When the value is large, and the object is far from the obstacle or boundary, The value is relatively small.

[0027] The physical meaning of the clearance penalty term is to convert the clearance distance into the near-obstacle risk intensity in the planning cost. When the current position is less than the safe radius from the obstacle or wall, it indicates that the robot's outline or safety envelope is already in a high-risk area, and the penalty term takes a larger value. As the current position gradually moves away from the obstacle and exceeds the range corresponding to the width of the penalty zone, the near-obstacle risk gradually decreases, and the penalty term decreases accordingly.

[0028] The mathematical basis of this formula is the idea of ​​piecewise monotonic mapping, which divides the clearance distance into high-risk, transition, and low-risk zones by using the safety radius and penalty band width. Within the transition zone, the penalty value decreases monotonically as the clearance increases, thus avoiding abrupt changes in the cost field; outside the safety zone, the penalty tends to be smaller, thus avoiding excessive penalties even in areas far from the obstacle.

[0029] The rationale for this setting is that the actual movement of a tunnel robot does not only require "no collisions," but also needs to retain a certain margin for positioning error, control error, and attitude disturbance. Therefore, constructing a continuous penalty using a safety radius and penalty band width can ensure that the path avoids obstacles without causing path jitter or planning instability due to hard threshold switching.

[0030] Furthermore, the clearance distance field is mapped to the basic turning radius scale field. Let the reference clearance be... First, construct the net normalized quantity: ; The physical meaning of the normalized clearance value lies in converting clearance distances under different map scales or different tunnel cross-sections into a unified dimensionless index, enabling clearance information to participate in subsequent calculations along with other normalized values ​​such as velocity, attitude risk, and environmental parameters. A larger normalized value indicates a more open location; a smaller value indicates a narrower location.

[0031] The mathematical basis of this formula is the scale normalization method, which scales the actual clearance distance by using a reference clearance and usually incorporates an upper limit truncation to ensure that clearance values ​​in different environments fall within a comparable range. The reference clearance can be selected based on the robot's dimensions, safety envelope, or tunnel design clearance.

[0032] This approach is reasonable: if the original distance values ​​are used directly, it is difficult to uniformly tune the parameters under different map resolutions or different tunnel sizes; by normalizing the clearance, the applicability of the algorithm in different tunnel scenarios and the transferability of parameters can be improved.

[0033] Subsequently, the net air normalization quantity was analyzed. Smoothing is performed to obtain a continuous headroom representation. Based on this, construct the basic turning radius scale field: ; in, and These are the lower and upper bounds of the turning radius scale, respectively.

[0034] The physical meaning of the basic turning radius scale field is that it directly transforms the spatial clearance into the desired turning scale at that location. When the clearance is small, the robot has limited turning space and needs to allow for a smaller turning radius to complete obstacle avoidance and passage; when the clearance is large, the robot has more space to move around and can use a larger turning radius to reduce the curvature of the path and improve driving smoothness.

[0035] The mathematical basis of this formula is a monotonic interpolation mapping, which uses a continuous headroom representation as the interpolation variable to generate a position-dependent turning radius scale between the lower and upper bounds of the turning radius scale. This mapping ensures that the larger the headroom, the larger the basic turning radius scale; and the smaller the headroom, the closer the basic turning radius scale is to the lower bound.

[0036] The rationale behind this setting is that if the robot is still limited by a fixed large turning radius in narrow areas, it may be unable to complete the necessary turns; conversely, if a small turning radius is still used in open areas, unnecessary bends and high-speed turns may occur. Therefore, by using the clearance-driven basic turning radius scale field, the path can simultaneously meet the practical motion requirements of "being able to turn in narrow places" and "being smoother in wide places." Specifically, the lower bound of the turning radius scale should not be less than the minimum effective turning scale allowed by the robot's chassis structure and control system; the upper bound of the turning radius scale is used to avoid excessive detours caused by excessive curvature constraints in open areas. Through these upper and lower bound constraints, it can be ensured that the generated scale field is both adaptive and does not exceed the robot's actual executable range.

[0037] In this embodiment, by setting upper and lower bounds for the curvature scale, a continuous, stable, and controllable basic turning radius scale field is formed.

[0038] To further reflect the impact of the robot's actual driving state and dynamic environmental changes on its steering ability, a velocity-curvature coupling correction factor, an attitude risk suppression factor, a joint environmental parameter correction factor, and a dynamic obstacle time decay correction factor are introduced into the basic turning radius scale field to obtain a composite spatial variable curvature constraint field: ; in, A normalized representation of the velocity or velocity level of a local target; This represents the attitude risk characterization consisting of pitch risk and / or roll risk; This represents the risk characterization of environmental parameters, used to characterize the intensity of risks in the environmental parameter field that affect passage stability, attitude safety, or execution cost. Represents the spatiotemporal risk characterization of dynamic obstacles; , , , These are the velocity correction factor, attitude risk correction factor, environmental parameter joint correction factor, and dynamic obstacle risk correction factor. This indicates the upper and lower bound truncation operator.

[0039] The physical significance of the composite spatial variable curvature constraint field lies in the fact that, based on the fundamental turning radius determined by the clearance, robot driving state and environmental risk factors are further introduced, enabling the steering constraint under the same clearance conditions to be adaptively corrected according to speed, attitude risk, ground access conditions, and dynamic obstacle risk. A larger field magnitude indicates that the location is more suitable for a larger turning radius and stronger curvature smoothing constraint; a smaller field magnitude indicates that the location allows for more flexible local steering.

[0040] In this formula, the velocity correction term reflects the velocity-curvature coupling relationship. Based on the kinematic characteristics of vehicles or mobile robots, performing small-radius turns at higher speeds will result in greater lateral acceleration and tracking errors. Therefore, when the local target speed is high, the desired turning radius should be increased to make the path smoother; when the local target speed is low, the robot has stronger low-speed maneuverability, and the desired turning radius can be appropriately reduced.

[0041] The basis for attitude risk correction is that slopes, cross slopes, or terrain undulations can affect the robot's pitch and roll stability. When the attitude risk is high, if the path still makes frequent sharp turns, it is easy to cause the attitude disturbances to accumulate. Therefore, it is necessary to correct the curvature scale to make the path more gentle or conservative in the high attitude risk area.

[0042] The joint correction term for environmental parameters is used to reflect the impact of non-geometric factors such as roughness, slipperiness, adhesion coefficient, water accumulation, and local energy consumption on motion stability. Although these factors may not directly constitute obstacles, they affect wheel-ground contact, trajectory tracking error, and energy consumption. Therefore, incorporating them into curvature constraint correction is engineeringly reasonable.

[0043] The dynamic obstacle risk correction item is used to handle risk sources that change over time, such as construction workers, mobile equipment, and temporary work vehicles. When the dynamic obstacle risk increases, path planning not only needs to increase the passage cost in that area, but also needs to simultaneously adjust the local turning scale and speed so that the robot avoids high-speed sharp turns or passing close to dynamic obstacles during obstacle avoidance.

[0044] The formula adopts a mathematical form of multiplicative coupling and upper and lower bound truncation. Its rationality lies in the fact that multiplicative coupling can reflect the combined amplification or suppression effect of multiple risk factors on the basic turning radius scale; upper and lower bound truncation ensures that the final constraint field is always within the preset controllable range, avoiding the turning radius scale being too large or too small due to abnormal local risk parameters, thereby ensuring that the planning result meets the actual motion capability of the robot.

[0045] In one implementation, the dynamic obstacle risk characterization The trajectory can be constructed from the predicted dynamic obstacle trajectory, relative velocity, estimated arrival time, distance decay factor, and spatiotemporal occupancy probability, and then normalized to ensure... .

[0046] For example, the construction steps of dynamic obstacle risk characterization are as follows: by calculating the predicted trajectory of dynamic obstacles in real time, the spatial location range of the obstacles in the future is determined; by combining the relative speed of the obstacle with its own speed, the rate at which the two approach or move away is quantified; by calculating the expected time to reach the same spatiotemporal point through relative position and relative speed, the collision time window is determined; the risk weight of distant obstacles is weakened and the influence of nearby obstacles is strengthened by using a distance attenuation factor; finally, the spatiotemporal occupancy probability (i.e., the probability of an obstacle appearing at a certain spatiotemporal point) is fused, and the above parameters are integrated into a unified dynamic risk value through weighted fusion and other methods to achieve accurate quantification of dynamic risk.

[0047] When a dynamic obstacle is predicted to enter the current grid or its neighboring grid during the prediction period, the composite spatial curvature constraint field, local target velocity, and the direction-related comprehensive cost field at the corresponding location are synchronously and adaptively corrected according to the collaborative rule of "increased risk - adaptive adjustment of local expected turning radius - increased direction-related comprehensive cost - decreased local target velocity" in order to improve the safety and stability of the obstacle avoidance process.

[0048] When the clearance is small Approaching the lower limit improves steering agility in tight areas; when the clearance is large and the target speed is high, Increase the size appropriately to improve path smoothness and reduce the risk of high-speed turns.

[0049] In this embodiment, a composite spatial variable curvature constraint field is constructed by integrating multiple factors such as clearance-based mapping, velocity-curvature coupling correction, attitude risk suppression correction, environmental parameter joint correction, dynamic obstacle spatiotemporal risk correction, and upper and lower bound truncation. This makes the curvature constraint no longer dependent on a fixed constant, but can change in conjunction with spatial clearance, local traffic speed requirements, attitude risk, and dynamic obstacle risk.

[0050] Compared to traditional solutions that control path shape solely with a fixed minimum turning radius, this embodiment emphasizes the spatial generation mechanism of curvature constraint parameters and their dynamic adjustment capability for path turning preferences. It extends curvature constraints from fixed parameters to a composite spatial field that can change with the environment and motion state, improving the adaptive turning capability under different clearance areas. This allows the path turning characteristics to adapt to changes in the local clearance environment, balancing smooth traffic flow with turning flexibility in confined spaces.

[0051] (II) Joint Construction of Location-Related Basic Cost Field and Direction-Sensitive Cost Field 1. Location-dependent basic cost field After constructing a composite spatial variable curvature constraint field, the clearance penalty term is further coupled with environmental accessibility parameters to obtain a location-dependent basic cost field, which characterizes the differences in access costs at different spatial locations. The clearance penalty increases when approaching walls or obstacles; as the access risk characterized by environmental parameters increases, the corresponding location cost increases accordingly. This ensures that the path optimization process considers not only "whether it is passable" but also "whether the cost of passage is reasonable."

[0052] Specifically, a clearance penalty term and at least one environmental accessibility parameter are selected to quantify the local accessibility difficulty at different spatial locations, forming a location-related basic cost field. The environmental accessibility parameter is used to reflect the impact of surface undulation, slope changes, surface irregularities, adhesion status, slipperiness, local energy consumption, or other environmental factors on the robot's accessibility stability, posture safety, or execution cost.

[0053] In one implementation, the environmental traversability parameter includes normalized roughness. and normalized slope strength In other embodiments, it may also include one or more of the following: adhesion coefficient, slipperiness, vibration intensity, wall roughness, and local energy dissipation factor. The headroom penalty is... The location-related fundamental cost field is a combined representation of the impact of airspace risk and environmental accessibility, and its specific form can be achieved by a weighted linear combination or other monotonic mapping forms: .

[0054] in, The weighting coefficient for the net clearance penalty item. This represents the i-th normalized environment traversability parameter. For the corresponding weighting coefficients, The number of selected environmental accessibility parameters. In this way, the basic cost field can reflect the differences in access risk at different locations, providing a location cost basis for subsequent joint solutions with the composite spatial variable curvature constraint field and the orientation-sensitive cost field.

[0055] The physical meaning of the location-related basic cost field is to evaluate the basic passage difficulty of a robot passing through a certain grid point from a spatial location perspective. The clearance penalty term reflects the safety distance risk between the location and the wall or obstacle, while the environmental passability parameter reflects the impact of factors such as slope, roughness, slipperiness, and adhesion conditions on passage stability and energy consumption.

[0056] The mathematical basis for using a weighted linear combination in this formula is that each normalized risk factor can be regarded as a monotonic contribution to the local passage cost, and the weight coefficients are used to express the importance of different risk factors in a specific robot platform and a specific tunnel environment. Since all parameters have been normalized, the linear combination can achieve multi-factor risk fusion while maintaining computational simplicity and interpretability.

[0057] The rationale behind this setting is that the actual tunnel passage risk is not solely determined by the distance to obstacles. For example, even in an area with ample clearance, a robot may still slip, vibrate, or become unstable if the slope is steep, the ground is rough, or it is slippery. Therefore, constructing a basic cost field by combining clearance penalties with environmental accessibility parameters allows the planned path to simultaneously move away from obstacles and avoid locations with poor accessibility.

[0058] To ensure the feasibility of movement, the environmental accessibility parameters can be normalized or thresholded based on the robot's maximum climbing ability, maximum allowable side tilt angle, wheel-ground adhesion conditions, and chassis passability. When a certain position exceeds the robot's allowable access limit, the cost of that position can be set to a maximum value or an impassable state.

[0059] 2. Direction-sensitive cost field Furthermore, the steepest downhill direction is calculated based on the terrain gradient, and a direction-sensitive cost is constructed in conjunction with the heading direction to penalize uphill, excessively steep downhill, and unfavorable lateral tilt directions. This direction-sensitive cost, together with the composite spatial variable curvature constraint field, acts on the path generation process, ensuring that the path not only meets geometric accessibility and steering capability requirements, but also attitude safety and motion comfort requirements.

[0060] To improve the ability of directional cost to represent attitude risk and overcome the shortcomings of existing directional-independent cost fields or single angle penalties that are difficult to distinguish the differences in attitude risk between "different headings at the same position", this embodiment constructs pitch risk along the direction of travel and roll risk perpendicular to the direction of travel, and introduces a net clearance-related amplification factor and an optional velocity-related amplification factor to form an adaptive directional-sensitive cost field.

[0061] Specifically, let the current heading angle be... The corresponding unit direction vector is Its normal vector is Then the risk item of pitch and tilt risk items They are defined as follows: ; ; in, This represents the gradient vector of the terrain height field.

[0062] The physical meaning of the pitch risk term is as follows: the component of the terrain slope in the robot's current heading direction is taken as the slope risk along the direction of travel. This value reflects the degree of uphill or downhill slope the robot faces when moving along the current heading, and can be used to determine whether there is a risk of difficulty in climbing, difficulty in braking downhill, or longitudinal attitude instability.

[0063] The physical meaning of the aforementioned roll risk term is as follows: the component of the terrain slope perpendicular to the robot's heading direction is taken as the lateral slope risk. This value reflects the degree of lateral tilt of the robot as it travels along the current heading. The greater the roll risk, the more likely the robot is to experience lateral load transfer, decreased tire adhesion, or rollover.

[0064] The mathematical basis of this formula is the principle of vector projection. The terrain height field gradient represents the direction of the fastest change in altitude. By projecting this gradient onto the heading unit vector and the heading normal vector respectively, the slope at the same location can be decomposed into a pitch component along the direction of travel and a roll component perpendicular to the direction of travel. This decomposition can distinguish the differences in attitude risk at the same location under different headings.

[0065] This setting is reasonable: on the same slope, the attitude risks of a robot are not the same when traveling uphill, uphill, and acrosshill. Using only direction-independent slope costs cannot accurately express this difference, but by using heading-related projections to obtain pitch and roll risks, the planner can prioritize the more stable passing direction.

[0066] Further construct a direction-sensitive cost field: ; in, This represents the normalized net void representation. This is the uphill penalty coefficient. This is the speed-related amplification factor, used to characterize the amplification effect of higher travel speeds on the risk of unfavorable course conditions. The penalty coefficient for steep downhill slopes. This is the roll penalty factor. This is the risk amplification factor in the near-obstacle area. , , These are the corresponding thresholds. Their physical meaning is that, under the same gradient conditions, the lateral correction margin in the near-obstacle area is smaller, the recoverable space after attitude instability is more limited, and higher travel speeds amplify the passage risks caused by attitude disturbances. Therefore, the same pitch or roll risk should correspond to higher directional costs under near-obstacle or high-speed conditions. By introducing clearance-related amplification factors and optional speed-related amplification factors, the ability of planning results to suppress high-risk headings can be enhanced, and the "position safe but heading unsafe" path that occurs when relying solely on position costs can be reduced.

[0067] The physical meaning of the orientation-sensitive cost field is that, at the same spatial location, different costs are assigned to the potential pitch risk, steep downhill risk, and roll risk depending on the robot's chosen heading. The higher the cost, the higher the attitude risk when the robot passes through that location with that heading, and the less likely the planner is to choose that direction.

[0068] In this formula, the uphill penalty term is used to suppress reverse slopes or steep uphill directions exceeding a set threshold; the excessively steep downhill penalty term is used to suppress routes with difficult braking or high risk of glide; and the roll penalty term is used to suppress routes with high risk of cross slope. Each penalty term adopts a linear growth form after the threshold, indicating that the risk does not increase significantly within the safe threshold, and gradually increases with the degree of exceedance after exceeding the threshold.

[0069] The mathematical basis of this formula is the piecewise linear penalty model and the heading-related risk projection model. The piecewise linear penalty can express engineering thresholds, such as the maximum comfortable slope and the maximum allowable roll angle, while avoiding the problem of hard constraints directly cutting off the feasible region and causing no path to be planned; the heading-related projection ensures that the same position has different costs under different headings.

[0070] The rationale for the clearance-related amplification factor lies in the fact that in the near-obstacle zone, the correction space after a robot tilts, yaws, or slips laterally is smaller; therefore, the same pitch or roll risk should be met with a higher cost in the near-obstacle zone. The rationale for the speed-related amplification factor lies in the fact that higher speeds amplify the effects of lateral acceleration, braking distance, and attitude disturbances; therefore, unfavorable headings at high speeds should be more strongly suppressed.

[0071] With the above settings, the orientation-sensitive cost field does not arbitrarily increase the path cost, but rather filters the heading based on the robot's posture stability, slope clearance capability, and near-obstacle correction margin, thereby ensuring that the selected path is not only geometrically reachable, but also easier to execute at the robot's posture and control levels.

[0072] When the robot travels on an uphill, steep downhill, or crosshill direction, the cost of directional sensitivity increases; when the robot is in a near-obstacle area, the cost of unfavorable directions is further amplified, thus causing the path planner to prioritize a more stable and less risky course.

[0073] In this embodiment, reverse slope, cross slope, near obstacle, and local speed requirements are all incorporated into the path constraints, transforming path planning from a single geometric obstacle avoidance to a joint environment-state optimization.

[0074] 3. Comprehensive Cost Field The location-related basic cost field and the orientation-sensitive cost field are fused to obtain the orientation-related comprehensive cost field: .

[0075] in, Represents grid points At heading angle The overall cost of passage, The direction-sensitive cost term is represented by the heading-related term; the comprehensive passage cost corresponding to the obstacle area is set to a preset maximum value to indicate impassability. The direction-related comprehensive cost field is used together with the composite spatial variable curvature constraint field to characterize the local passage cost and turning constraint requirements in the path solving process.

[0076] The physical meaning of the integrated cost field is to simultaneously evaluate whether it is appropriate for the robot to pass through a certain location and whether it is safe to pass through that location in a certain heading. Among them, the location-related basic cost reflects the difficulty of passing through the location itself, while the direction-sensitive cost reflects the posture safety and motion comfort of the robot when passing through the location in a specific heading.

[0077] The mathematical basis of this formula is the idea of ​​state-space cost fusion, which expands the two-dimensional position cost into a position-heading state cost. Traditional position cost can only determine whether a grid is suitable to pass through, but cannot distinguish the risks of different headings under the same grid; the integrated cost field, by introducing a heading variable, enables path search or trajectory optimization to compare the advantages and disadvantages of position and attitude orientation simultaneously when the state is expanded.

[0078] The rationale for setting the obstacle region's cost to a preset maximum value is that obstacles or wall boundaries belong to areas that the robot cannot traverse. By maximizing the cost, these obstacles can be excluded from feasible paths during the optimization solution. This approach is compatible with continuous cost optimization and also ensures that the path will not enter actually impassable areas.

[0079] This comprehensive cost field ensures motion feasibility: when selecting candidate states, the planner not only avoids positions with high near-obstacle risk and high environmental risk, but also avoids unfavorable headings such as uphill, steep downhill, and side tilt, thus outputting a path that is more suitable for the robot chassis to drive stably and control tracking.

[0080] (III) Joint solution mechanism for unified objective of composite space variable curvature constraint field and comprehensive cost field This embodiment incorporates a location-dependent fundamental cost field, a direction-sensitive cost field, and a composite spatial curvature constraint term into the same objective function or search evaluation function. The location-dependent fundamental cost field determines the passage difficulty of a candidate location, the direction-sensitive cost field determines the attitude safety differences of that location under different headings, and the composite spatial curvature constraint term determines the local expected turning radius and curvature penalty intensity for the same candidate state. These three factors work together during the expansion and ranking of the same candidate location-heading state, forming a synergistic joint optimization. The location-dependent fundamental cost field characterizes the passage cost at different spatial locations, the direction-sensitive cost field characterizes the attitude safety and passage risk under given location and heading conditions, and the composite spatial curvature constraint field characterizes the adjustment intensity of the path curvature at that location.

[0081] In a specific implementation, the path solving objective can be represented as a weighted combination of location-dependent basic cost, orientation-sensitive cost, and curvature constraint term: ; in, Indicates the location-related basic cost. Indicates position Location and heading Related directional sensitive costs, Indicates path curvature. This represents the value of the composite spatial variable curvature constraint field corresponding to the current position of the path. For direction-sensitive cost trade-off coefficients, is the curvature constraint term tradeoff coefficient.

[0082] The physical meaning of the objective function is to simultaneously accumulate positional travel costs, heading and attitude risk costs, and curvature constraint costs along the entire path, so that the final path achieves a balance between obstacle avoidance safety, attitude stability, and steering smoothness. Position-related basic costs enable the path to avoid high-risk areas such as near obstacles, rough terrain, and steep slopes; orientation-sensitive costs enable the path to avoid headings with unfavorable attitudes; and curvature constraint terms enable the path to avoid unnecessary sharp turns or abrupt curvature changes.

[0083] The mathematical basis of this objective function is the concept of weighted multi-objective optimization. Tunnel robot path planning needs to satisfy multiple objectives simultaneously, and these objectives may conflict. For example, moving away from obstacles may increase path length, while reducing curvature may cause detours. Therefore, different objectives are unified into the same evaluation function through weighting coefficients, enabling the path planner to search for a compromise solution with lower total cost within the feasible region.

[0084] The rationale for the curvature constraint term lies in the fact that path curvature and robot turning radius are inversely related; the greater the curvature, the smaller the corresponding turning radius, requiring the robot to perform more drastic turning maneuvers. Incorporating both path curvature and the composite spatial variable curvature constraint field into the objective function allows the curvature penalty intensity at different locations to vary with clearance, speed, and risk, thus avoiding the problem that fixed curvature weights cannot simultaneously adapt to both narrow and open areas.

[0085] The objective function ensures the feasibility of actual motion: on the one hand, the robot's inherent minimum turning radius serves as a hard constraint or lower limit constraint to prevent the generation of mechanically unexecutable small-radius turns; on the other hand, the curvature regularization term serves as a soft constraint to suppress unnecessary sharp turns and improve trajectory continuity; combined with the comprehensive cost field, the output path can simultaneously meet the requirements of obstacle avoidance, attitude safety, and turning capability.

[0086] Therefore, the position-dependent basic cost field determines which regions the path should preferentially pass through, the direction-sensitive cost field determines which headings should be preferentially adopted within a given region, and the composite spatial variable curvature constraint field determines the allowable turning amplitude and curvature continuity requirements of the path within these regions. The three participate in optimization together in a unified objective function and jointly determine the cumulative cost and local turning capability of the candidate position-heading state.

[0087] Subsequently, based on the solution objective, the composite spatial variable curvature constraint field is mapped to the local desired turning radius parameter, and the weight of the local curvature regularization term is determined by the local desired turning radius parameter; then, the local desired turning radius parameter, the weight of the local curvature regularization term, the robot's inherent minimum turning radius constraint, and the direction-related comprehensive cost field are input into the path planner for optimization and solution.

[0088] The path planner is a solver capable of performing path search or continuous path optimization in the position-heading state space. During the solution process, it can receive local curvature regularization term weights and / or local desired turning radius parameters, while simultaneously satisfying the robot's inherent minimum turning radius constraint and curvature continuity requirement. Preferably, the path planner is a solver capable of handling curvature regularization terms, local desired turning radius parameters, and curvature continuity constraints.

[0089] The final path obtained from the solution is output to the robot control system as input for subsequent chassis control, speed planning, or trajectory tracking modules. Because the path is simultaneously affected by clearance, environmental parameters, directional safety, and spatial variable curvature constraints throughout the solution process, the output path has higher passability, greater safety margin, and better smoothness compared to traditional fixed curvature constraint paths.

[0090] In this embodiment, in areas with steep gradients, the probability of choosing unfavorable headings is reduced by increasing the directional cost; in areas with limited clearance and frequent turning, necessary maneuverability is maintained by using a smaller curvature scale. This achieves joint optimization of "direction safety constraints" and "curvature adaptive constraints".

[0091] To verify the effectiveness of this embodiment in a confined tunnel environment, a two-dimensional tunnel grid simulation example was constructed. The simulation map includes wall boundaries, local static obstacles, narrow passages, open passage areas, and dynamic obstacle risk areas. As shown in Table 1, the fixed curvature constraint path planning method is compared with the composite spatial variable curvature constraint path planning method of this embodiment. Both methods use the same start point, end point, map boundaries, obstacle distribution, and robot safety radius.

[0092] Table 1. Performance comparison between existing technology and this embodiment;

[0093] As shown in Table 1, compared with the fixed curvature constraint method, the minimum clearance of the method in this embodiment is increased from 0.39366 m to 0.41324 m, indicating that a greater safety margin is maintained between the path and the wall or obstacle; the maximum curvature is reduced from 8.142 pm to 1.8884 pm, and the rate of change of curvature is reduced from 0.17514 to 0.037891, indicating that this embodiment can suppress local sharp turns and improve the continuity of path curvature through the composite spatial variable curvature constraint field; the cumulative cost of dynamic risk is reduced from 3.9705 to 2.7484, indicating that this embodiment can actively avoid dynamic obstacle risk areas during path planning.

[0094] Meanwhile, the path length of the method in this embodiment increased from 12.231 m to 12.576 m, indicating that in this example, the method in this embodiment sacrificed a smaller path length to obtain a larger headroom safety margin, a lower curvature abrupt change, and a lower dynamic obstacle risk; this result reflects the comprehensive optimization characteristics of this embodiment in terms of safety, smoothness, and dynamic obstacle avoidance capability.

[0095] Therefore, the above simulation examples can verify that this embodiment, through the joint solution of the clearance-driven composite spatial variable curvature constraint field, the position-related basic cost field, and the direction-sensitive cost field, can improve the path clearance safety margin, reduce local sharp turns and curvature abrupt changes, and enhance the ability to avoid dynamic obstacle risk areas in a confined tunnel environment.

[0096] At the same time, such as Figure 3The diagram shows a comparison of paths using the fixed curvature constraint method and the method of this embodiment on the same two-dimensional tunnel simulation map. Black areas represent walls or obstacles, gray areas represent safety radius expansion areas, dashed areas represent dynamic obstacle risk areas, red dashed lines represent fixed curvature constraint paths, and blue solid lines represent the path of this application. Figure 3 As can be seen, the path in this embodiment actively avoids dynamic obstacle risk areas and maintains a smoother trajectory when passing through open areas.

[0097] In addition, such as Figure 4 As shown, the distribution of the composite spatial variable curvature constraint field in this embodiment is illustrated. A larger color value indicates a larger local desired turning radius at that location, meaning the path tends to remain smooth in that region; a smaller color value indicates that more flexible local turning is allowed at that location. Figure 4 It is evident that open areas correspond to larger turning radii, while narrow or near-obstacle areas correspond to smaller turning radii, which meets the path planning requirement of "turning in narrow places and smoother in wide places".

[0098] This specific embodiment constructs and modifies a composite spatial variable curvature constraint field, enabling the path to have differentiated local steering preferences under different clearance, speed requirements, and attitude risks. Then, it combines the position-related basic cost and the direction-sensitive cost for unified solution, thereby achieving path planning that balances passability, ride comfort, and attitude safety in a restricted tunnel environment.

[0099] Example 2 This embodiment provides a tunnel path planning system based on clearance-driven spatial curvature constraints, including: The environment modeling module is configured to acquire a target tunnel grid map, calculate the Euclidean distance from each grid point to the nearest obstacle or wall boundary to obtain a clearance distance field, and define a clearance penalty term based on the clearance distance field, safety radius, and penalty band width. The composite spatial variable curvature constraint field construction module is configured to construct a basic turning radius scale field based on the clearance distance field, and to couple and correct the basic turning radius scale field according to the local target velocity field, attitude risk field, environmental parameter field and dynamic obstacle spatiotemporal risk field to obtain the composite spatial variable curvature constraint field. The cost field construction module is configured to construct a location-related basic cost field based on the airspace penalty term and environmental accessibility parameters; establish a direction-sensitive cost field based on the direction risk penalty coefficient; and fuse the basic cost field and the direction-sensitive cost field to obtain a comprehensive cost field. The path planning module is configured to input the composite spatial variable curvature constraint field, the robot's inherent minimum turning radius constraint, and the comprehensive cost field into the path planner to obtain the optimal path.

[0100] As one implementation, the environmental modeling module is used to acquire a target tunnel grid map and construct a clearance distance field, an environmental parameter field characterizing trafficability and attitude safety, and a local target velocity field. The environmental parameter field includes at least one environmental element that affects traffic stability, attitude safety, or execution cost. In one implementation, the environmental element includes at least slope information and may optionally include one or more of roughness, adhesion coefficient, slipperiness, water accumulation, wall roughness, or local energy consumption factor. When the environmental element includes slope information, a pitch risk field and a roll risk field are constructed based on the slope information, and an attitude risk field can be further constructed based on the pitch risk field and the roll risk field.

[0101] As one implementation method, the composite spatial variable curvature constraint field construction module is used to construct a basic turning radius scale field based on the clearance distance field, and to couple and correct the basic turning radius scale field according to the local target velocity field, attitude risk field, environmental parameter field and / or dynamic obstacle spatiotemporal risk field to obtain the composite spatial variable curvature constraint field; wherein, the coupling correction includes at least a velocity-curvature coupling term, an attitude risk suppression term, an environmental parameter joint correction term and a dynamic obstacle time decay term.

[0102] As one implementation method, the cost field construction module includes a basic cost construction module, a direction-sensitive cost construction module, and a comprehensive cost fusion module.

[0103] Among them, the basic cost construction module is used to construct a location-related basic cost field based on the airspace penalty term and environmental accessibility parameters; The orientation-sensitive cost construction module is used to construct an orientation-sensitive cost field based on pitch risk, roll risk, and clearance-related amplification factor. The attitude risk field is used to perform orientation-independent global conservative correction on the basic turning radius scale field. The pitch risk and roll risk are used to construct orientation-sensitive costs under a given heading condition, and the two have different levels of action. The integrated cost fusion module is used to fuse the location-related basic cost field and the direction-sensitive cost field to obtain the direction-related integrated cost field.

[0104] As one implementation, the path planning module is used to map the composite spatial variable curvature constraint field into a local desired turning radius parameter, and further determine the weight of the local curvature regularization term based on the local desired turning radius parameter; the local desired turning radius parameter, the weight of the local curvature regularization term, the robot's inherent minimum turning radius constraint, and the direction-related comprehensive cost field are input together into the path planner to obtain the optimal path.

[0105] The path generated in this embodiment has better maneuverability in narrow, low-speed maneuvering areas and higher curvature continuity and stability in high-clearance passage areas, making it suitable for scenarios such as tunnel inspection, underground transportation, and navigation of mobile robots in confined spaces.

[0106] It should be noted that each module in this embodiment corresponds to each step in Embodiment 1. The specific implementation process can be found in the relevant description in Embodiment 1, and will not be repeated here.

[0107] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the tunnel path planning method based on clearance-driven spatial variable curvature constraints as described in Embodiment 1 above.

[0108] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the tunnel path planning method based on clearance-driven spatial variable curvature constraints as described in Embodiment 1 above.

[0109] The steps or modules involved in Embodiments 2 to 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0110] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A tunnel path planning method based on clearance-driven spatial variable curvature constraints, characterized in that, include: Obtain the target tunnel grid map, calculate the Euclidean distance from each grid point to the nearest obstacle or wall boundary, and obtain the clearance distance field; Based on the clearance distance field, safety radius, and penalty band width, a clearance penalty term is defined; A basic turning radius scale field is constructed based on the clearance distance field, and the basic turning radius scale field is coupled and corrected according to the local target velocity field, attitude risk field, environmental parameter field and dynamic obstacle spatiotemporal risk field to obtain a composite spatial variable curvature constraint field. A location-related basic cost field is constructed based on the aforementioned airspace penalty term and environmental accessibility parameters; a direction-sensitive cost field is established based on the direction risk penalty coefficient; the basic cost field and the direction-sensitive cost field are fused to obtain a comprehensive cost field. The optimal path is obtained by inputting the composite spatial variable curvature constraint field, the robot's inherent minimum turning radius constraint, and the comprehensive cost field into the path planner.

2. The tunnel path planning method based on clearance-driven spatial variable curvature constraints as described in claim 1, characterized in that, The construction of the basic turning radius scale field based on the clearance distance field specifically includes: The clearance distance field is normalized and smoothed to obtain a continuous clearance characterization. Based on the lower and upper bounds of the curvature scale and the continuous headroom representation, a basic turning radius scale field is constructed.

3. The tunnel path planning method based on clearance-driven spatial variable curvature constraints as described in claim 1, characterized in that, The local target velocity field refers to the normalized representation of the local target velocity or velocity level; The attitude risk field refers to the attitude risk characterization consisting of pitch risk and / or roll risk; The environmental parameter field refers to the intensity of risk that affects traffic stability, attitude safety, or execution cost. The dynamic obstacle spatiotemporal risk field is constructed by the predicted trajectory of the dynamic obstacle, relative velocity, estimated arrival time, distance decay factor, and spatiotemporal occupancy probability.

4. The tunnel path planning method based on clearance-driven spatial variable curvature constraints as described in claim 1, characterized in that, The environmental traversability parameters include environmental roughness and slope intensity.

5. The tunnel path planning method based on clearance-driven spatial variable curvature constraints as described in claim 1, characterized in that, The directional risk penalty coefficient includes an uphill penalty coefficient, a speed-related amplification coefficient, an excessively steep downhill penalty coefficient, a roll penalty coefficient, and a near-obstacle area directional risk amplification coefficient; the speed-related amplification coefficient is used to characterize the amplification effect of high driving speed on unfavorable directional risks.

6. The tunnel path planning method based on clearance-driven spatial variable curvature constraints as described in claim 1, characterized in that, After obtaining the comprehensive cost field, before inputting the composite spatial variable curvature constraint field, the robot's inherent minimum turning radius constraint, and the comprehensive cost field into the path planner, it also includes constructing a path solution objective based on the basic cost field, the direction-sensitive cost field, and the composite spatial variable curvature constraint field.

7. The tunnel path planning method based on clearance-driven spatial variable curvature constraints as described in claim 1, characterized in that, The process of inputting the composite spatial variable curvature constraint field, the robot's inherent minimum turning radius constraint, and the comprehensive cost field into the path planner specifically involves: The composite spatial variable curvature constraint field is mapped to a local desired turning radius parameter, and the weight of the local curvature regularization term is determined based on the local desired turning radius parameter. The local expected turning radius parameter, the weight of the local curvature regularization term, the robot's inherent minimum turning radius constraint, and the comprehensive cost field are all input into the path planner.

8. A tunnel path planning system based on clearance-driven spatial variable curvature constraints, characterized in that, include: The environment modeling module is configured to acquire a target tunnel grid map, calculate the Euclidean distance from each grid point to the nearest obstacle or wall boundary to obtain a clearance distance field, and define a clearance penalty term based on the clearance distance field, safety radius, and penalty band width. The composite spatial variable curvature constraint field construction module is configured to construct a basic turning radius scale field based on the clearance distance field, and to couple and correct the basic turning radius scale field according to the local target velocity field, attitude risk field, environmental parameter field and dynamic obstacle spatiotemporal risk field to obtain the composite spatial variable curvature constraint field. The cost field construction module is configured to construct a location-related basic cost field based on the airspace penalty term and environmental accessibility parameters; establish a direction-sensitive cost field based on the direction risk penalty coefficient; and fuse the basic cost field and the direction-sensitive cost field to obtain a comprehensive cost field. The path planning module is configured to input the composite spatial variable curvature constraint field, the robot's inherent minimum turning radius constraint, and the comprehensive cost field into the path planner to obtain the optimal path.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the tunnel path planning method based on the variable curvature constraint of the clearance-driven space as described in any one of claims 1-7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the tunnel path planning method based on the variable curvature constraint of the clearance-driven space as described in any one of claims 1-7.

Images