Long tunnel emergency evacuation robot and evacuation guiding method thereof

Abstract

The invention relates to the technical field of robot evacuation guiding, in particular to a long tunnel emergency evacuation robot and an evacuation guiding method thereof, and the method comprises the steps: fusing laser radar information and a depth image, generating a point cloud of an environment in front of the robot in real time, and collecting the environment temperature of each collection point in front of the robot in a visual field range; an obstacle is marked; matching the same obstacle at adjacent moments, identifying different types of obstacles, and constructing a differentiated two-dimensional bounding box for the obstacles in a horizontal plane; calculating an obstacle threat degree, a passing guide value, a temperature threat degree and a passing priority of each fan-shaped region; on the basis of the gravitational force of the artificial potential field algorithm, heuristic gravitational force is generated for each fan-shaped area, and the robot moves according to the direction of the resultant force of the gravitational force, the repulsive force and the heuristic gravitational force of the artificial potential field algorithm to guide people to evacuate safely. According to the method, the robot can be promoted to smoothly and efficiently move towards the target, and the evacuation efficiency in a complex environment is improved.

Description

Technical Field

[0001] This application relates to the field of robot evacuation guidance technology, specifically to an emergency evacuation robot for long tunnels and its evacuation guidance method. Background Technology

[0002] Emergency evacuation through long tunnels presents a significant challenge in the field of public safety. In the event of accidents such as fires or traffic crashes, the complex internal structure and low visibility of tunnels can easily lead to severe traffic congestion, making it difficult to guide trapped individuals to safety. Emergency evacuation robots, as crucial equipment for on-site rescue, must not only autonomously plan routes in the complex and dynamic environment of long tunnels, but also avoid obstacles and dangerous areas during the evacuation process, quickly guiding people to their destination.

[0003] Using artificial potential field methods for real-time path planning is the mainstream technical approach for emergency evacuation robots. However, in real long tunnel disaster scenarios, the internal environment is extremely complex and dynamically changing. For example, dense smoke from fires can reduce the robot's perception capabilities, causing it to ignore the safety threats posed by unevenly distributed high-temperature areas and lack the ability to track and predict dynamically moving obstacles. This makes the traditional artificial potential field method prone to oscillations and getting stuck in local optima near local obstacles, and may even plan paths that cross high-temperature danger zones, significantly increasing the risk of evacuation routes and making it impossible to complete efficient and safe evacuation guidance tasks. Summary of the Invention

[0004] To address the aforementioned technical problems, a long tunnel emergency evacuation robot and its evacuation guidance method are provided to solve the existing issues.

[0005] The solution to the technical problem presented in this application is to provide an emergency evacuation robot for long tunnels and its evacuation guidance method, including the following steps:

[0006] In a first aspect, embodiments of this application provide a method for guiding an emergency evacuation robot in a long tunnel, the method comprising the following steps:

[0007] By fusing LiDAR information and depth images, a point cloud of the environment in front of the robot is generated in real time, and the ambient temperature of each collection point within the field of view in front of the robot is collected, dividing the field of view into multiple fan-shaped areas.

[0008] Cluster the point cloud, analyze the distribution dispersion of the point cloud within each cluster, filter the clusters and mark the obstacles;

[0009] Based on the distance relationship between obstacles between adjacent time points, the difference in the number of points and the difference in their discrete distribution, the same obstacle at adjacent time points is matched, and different types of obstacles are identified according to the matching results and the movement of the same obstacle at adjacent time points, and a differentiated two-dimensional bounding box is constructed for it in the horizontal plane.

[0010] The obstacle threat level of each sector is calculated by using the proportion of the boundary of the two-dimensional bounding box in each sector and the distance between the boundary and the robot.

[0011] The deviation of the direction of each sector from the direction of the robot to the destination is evaluated, the passage guidance value of each sector is calculated, and the passage priority of each sector is obtained based on the average level of ambient temperature and obstacle threat level within the sector.

[0012] Based on passage priority, and on the basis of the gravitational force of the artificial potential field algorithm, heuristic gravity is generated for each sector area. The robot moves according to the direction of the resultant force of the gravitational force, repulsive force and heuristic gravity of the artificial potential field algorithm, guiding the safe evacuation of personnel.

[0013] Preferably, the process of filtering clusters and marking obstacles includes:

[0014] Calculate the degree of dispersion of the coordinate values ​​of all point clouds in each cluster on the X-axis, Y-axis, and Z-axis, respectively;

[0015] The dispersion of each cluster is the result of a positive fusion of the dispersion of all axes within each cluster;

[0016] Obtain the segmentation threshold of the distribution dispersion of all clusters, and mark all point clouds in clusters with a distribution dispersion greater than the segmentation threshold as an obstacle.

[0017] Preferably, the matching of the same obstacle at adjacent time points includes:

[0018] Project the point cloud corresponding to all obstacles on the local environment map at the current moment onto the horizontal plane to generate corresponding mapping points; obtain the centroid of all mapping points of each obstacle in the horizontal plane; count the number of all point clouds corresponding to each obstacle;

[0019] From all obstacles at the current moment and all obstacles at the previous moment, select one obstacle each as an obstacle pair; calculate the distance between the centroids of each obstacle pair, denoted as the movement distance; calculate the difference in the number of obstacles between each obstacle pair, denoted as the quantity difference; calculate the difference in the distribution dispersion between each obstacle pair, denoted as the distribution difference.

[0020] The correlation coefficients of each obstacle pair were negatively correlated with the movement distance, the difference in number, and the difference in distribution, respectively.

[0021] Based on the descending order of the correlation coefficient, obstacle pairs with a correlation coefficient higher than a preset threshold are iteratively identified as matching pairs, wherein there is a unique matching relationship between the obstacle pairs represented by the matching pairs.

[0022] Preferably, identifying different types of obstacles includes: based on the matching results of obstacle pairs, for any obstacle at the current moment, if it does not successfully match with an obstacle at the previous moment, then the obstacle is a newly added obstacle; if it successfully matches and the movement distance between the two successfully matched obstacles is equal to 0, then the obstacle is a static obstacle; if it successfully matches and the movement distance between the two successfully matched obstacles is greater than 0, then the obstacle is a dynamic obstacle.

[0023] Preferably, constructing differentiated two-dimensional bounding boxes in the horizontal plane includes:

[0024] For all mapping points of the newly added obstacle in the horizontal plane, a two-stage bounding box construction strategy is adopted to obtain the two-dimensional bounding box of the newly added obstacle;

[0025] For dynamic obstacles, the speed of the dynamic obstacle is calculated based on its distance between the current time and the previous time, the position of the dynamic obstacle in the next time is predicted, the point clouds of the dynamic obstacle at the two positions are merged, the merged point cloud is projected onto all mapping points in the horizontal plane, and a two-stage bounding box construction strategy is adopted to obtain the two-dimensional bounding box of the dynamic obstacle.

[0026] Construct a two-dimensional bounding box for all mapping points of a static obstacle in the horizontal plane.

[0027] Preferably, the calculation of the obstacle threat level for each sector region includes:

[0028] Sampling is performed on all edges of the 2D bounding box to obtain each boundary point; the distance between each boundary point in each sector and the robot's location is calculated and recorded as the relative distance; the proportion of the number of all boundary points in each sector to the total field of view is calculated.

[0029] The obstacle threat level is positively correlated with the proportion, but negatively correlated with the relative distance to all boundary points.

[0030] Preferably, the calculation of the passage guidance value for each sector area includes: taking the robot's location as the starting point, the direction from the angle bisector of each sector area to the arc is the passage direction for each sector area; taking the direction from the robot's location to the target point's location as the target direction; calculating the cosine of the angle between the passage direction and the target direction, and performing a positive mapping on it as the passage guidance value for each sector area.

[0031] Preferably, obtaining the passage priority of each sector region includes:

[0032] The temperature threat level of each sector is the result of positively fusing the ambient temperatures of all collection points within each sector.

[0033] The passage priority is negatively correlated with obstacle threat level and temperature threat level, but positively correlated with passage guidance value.

[0034] Preferably, generating heuristic gravity for each sector region includes: using an artificial potential field algorithm to obtain the gravitational force of the target point on the robot, using the product of the passage priority of each sector region and the gravity as the heuristic gravity value of each sector region, and obtaining the heuristic gravity with the heuristic gravity value of each sector region as the magnitude of the force and the passage direction as the direction of the force.

[0035] Secondly, embodiments of this application also provide a long tunnel emergency evacuation robot, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the evacuation guidance method for a long tunnel emergency evacuation robot described above.

[0036] This application has at least the following beneficial effects:

[0037] This application generates a point cloud of the robot's environment by fusing LiDAR information and depth images. Its advantages lie in significantly improving perception capabilities in harsh environments such as smoke and darkness through multi-source information complementarity. Clustering the point cloud and marking obstacles provides the advantage of accurately distinguishing independent obstacles from continuous tunnel structures by analyzing the distribution of point clouds within clusters, thus enabling the identification of real obstacles for subsequent tracking. Matching obstacles at adjacent time points identifies the same obstacle, and based on the matching results and the movement of the same obstacle, it identifies different types of obstacles, namely newly added obstacles, dynamic obstacles, and static obstacles. Based on the type of obstacle, differentiated two-dimensional bounding boxes are constructed. This has the advantage of defining larger danger zones for potentially high-risk new obstacles and dynamic obstacles, maximizing robot safety and improving the accuracy and safety of obstacle avoidance in emergency evacuation robots. Calculating the obstacle threat level of each sector region has the advantage of considering the distance between the obstacle boundary and the robot within the sector region, as well as the inclusion of the boundary, quantifying the impact of obstacles on the robot's passage within the sector region, and reflecting the difficulty and risk of the robot's passage in that direction. Calculating the passage guidance value for each sector region has the advantage of considering the local forward direction corresponding to the sector region and the global final destination. Consistency of target direction ensures that the robot does not lose its global orientation when performing local obstacle avoidance, overcoming local optimization problems such as walking along edges or wandering aimlessly, and guiding the robot to always efficiently approach the safety exit. Obtaining the passage priority of each sector area has the benefit of considering the impact of ambient temperature within the sector area on the robot's safety, enabling the robot to actively identify and avoid thermal hazard areas such as fire sources and high-temperature smoke. The superiority of the direction corresponding to the sector area is comprehensively evaluated from three dimensions: the impact of obstacles on passage, the impact of high-temperature hazard areas, and deviation from the passage direction. This ensures that the robot chooses a sector direction that both faces the target and avoids high-temperature and high-threat areas, significantly enhancing obstacle avoidance. The reliability and safety of evacuation paths are improved. Based on the priority of passage, directional heuristic gravity is generated for each sector area on the basis of the gravity of the artificial potential field algorithm. The robot moves according to the direction of the resultant force of the gravity, repulsion and heuristic gravity of the artificial potential field algorithm, guiding people to evacuate safely. Its beneficial effect is that it introduces directional heuristic gravity based on the priority of passage on the basis of the traditional artificial potential field method, which effectively makes up for the deficiencies of gravity and repulsion. This resultant force guidance avoids oscillation or stagnation caused by local minima, prompting the robot to move smoothly and efficiently toward the target, improving the evacuation efficiency in complex environments, significantly reducing the risk of collision, and ultimately achieving efficient and safe personnel evacuation guidance. Attached Figure Description

[0038] The following is a detailed description of an emergency evacuation robot guidance method for long tunnels, based on the accompanying drawings.

[0039] Figure 1 A flowchart illustrating the steps of an emergency evacuation robot guidance method for long tunnels, as provided in this application embodiment;

[0040] Figure 2 A flowchart illustrating the steps of the method for obtaining the passage priority of each sector area provided in the embodiments of this application. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description, in conjunction with the accompanying drawings and implementation examples, provides a further detailed explanation of the emergency evacuation robot for long tunnels and its evacuation guidance method proposed in this application. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit the scope of this application.

[0042] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0043] Please see Figure 1 The diagram illustrates a flowchart of an evacuation guidance method for an emergency evacuation robot in a long tunnel, according to an embodiment of this application. The method includes the following steps:

[0044] Step 1: By fusing LiDAR information and depth images, a point cloud of the environment in front of the robot is generated in real time, and the ambient temperature of each collection point within the field of view in front of the robot is collected.

[0045] In actual operation of highway tunnels, the narrow, enclosed space and low visibility make them prone to traffic accidents such as vehicle breakdowns, collisions, and hazardous material leaks, which can then trigger fires. When a tunnel fire occurs, it is sudden and extremely dangerous. After a disaster, traffic congestion makes it difficult for firefighters to reach the scene for rescue operations. Trapped occupants, unaware of the external fire situation, may experience an "island effect" and be unwilling to leave their vehicles, thus making evacuation difficult. Considering the difficulty of fire and ambulance vehicles reaching the disaster site and the urgency of evacuation, robots are used for rescue work. By assessing the current environmental conditions, the robot performs reasonable path planning for multi-target point proximity detection, thereby completing the evacuation of trapped personnel.

[0046] Based on the above analysis, the emergency evacuation robot integrates multiple source sensors, including LiDAR, inertial measurement unit (IMU), and depth camera. The LiDAR scans the environment in front of the robot in real time to generate point clouds.

[0047] In the event of a fire in the tunnel, the smoke generated by the fire will gradually spread over time. The denser smoke will absorb the laser emitted by the lidar transmitter. Therefore, the point cloud collected by the lidar needs to be defogging processed.

[0048] In this embodiment, in the event of a fire in the tunnel, the smoke generated by the fire will gradually accumulate above the tunnel structure over time. When the laser beam emitted by LiDAR penetrates the smoke, it will be absorbed and scattered, resulting in the loss of some laser energy. This significantly weakens the intensity of the laser signal that eventually returns to the receiver. Therefore, the intensity value of the points generated by the smoke will be much lower than that of the points reflected back from the surface of a solid object. By traversing the point cloud generated by the LiDAR and reading its intensity value, the point cloud with an intensity value less than a preset threshold is removed. The preset threshold is set to 10. As other implementation methods, the implementer can set it according to the actual situation.

[0049] Among them, the depth camera is equipped with an infrared emitter, two infrared cameras and a color camera. The infrared imaging device does not require lighting conditions and has the ability to penetrate smoke and dust environments to improve the perception ability of emergency rescue robots in smoke tunnels. The depth camera is used to collect depth images of the environment in front of the robot in real time. After filtering the depth images, the filtered depth images are converted into point clouds based on the extrinsic parameter matrix of the depth camera and spatial coordinate transformation.

[0050] In this embodiment, a bilateral filter is used to filter the depth image. The bilateral filter is a well-known technique and will not be described in detail here. Secondly, the process of converting the depth image into a point cloud is a well-known technique and will not be described in detail here.

[0051] The robot's pose is calculated using IMU integration to correct point cloud distortion caused by its own motion in long tunnel scenarios. IMU integration is a well-known technique and will not be elaborated here.

[0052] The point cloud generated by the lidar is fused with the point cloud generated by the depth image;

[0053] It should be noted that the point cloud fusion process is a well-known technology and will not be described in detail here.

[0054] The depth camera uses an infrared camera to collect real-time data on the ambient temperature in front of the robot within its field of view, and obtains infrared thermal images.

[0055] In this embodiment, the field of view is a sector with a radius of 5m centered on the robot's location and an azimuth angle of 120 degrees in front of the robot. In other implementation methods, the implementer can set the field of view according to the actual situation.

[0056] Since the depth value of each pixel in the depth image represents the distance of the point in space from the robot, and the temperature value of each pixel in the infrared thermal image represents the temperature at the point in space, the infrared thermal image and the depth image are matched pixel by pixel. Combined with the intrinsic and extrinsic parameter matrix of the infrared camera, the pixel coordinates and their corresponding temperature values ​​in the infrared thermal image are converted into three-dimensional spatial coordinates, so as to obtain the ambient temperature of each collection point within the field of view in real time.

[0057] At this point, we obtain the point cloud of the environment in front of the robot at the current moment, as well as the ambient temperature of each collection point within the field of view.

[0058] Step 2: Cluster the point cloud, analyze the discreteness of the point cloud distribution within each cluster, filter the clusters and mark the obstacles; based on the distance relationship between obstacles between adjacent time points, the difference in the number of points and the difference in their discrete distribution, match the same obstacle at adjacent time points, and based on the matching results and the movement of the same obstacle at adjacent time points, identify different types of obstacles and construct differentiated two-dimensional bounding boxes for them in the horizontal plane.

[0059] Within long tunnels, obstacles, such as vehicles and gravel, typically appear in point clouds as localized point cloud aggregates with high spatial density and clear boundaries. Their morphological features are prominent, distributed in different locations, and exist independently of each other. In contrast, environmental structures such as tunnel walls and road surfaces, although vast in scale, exhibit a broad and relatively uniform point cloud density distribution. Therefore, by clustering the point cloud, obstacles can be identified, specifically:

[0060] Cluster all point clouds in the robot's front environment at each time point to obtain multiple clusters;

[0061] In this embodiment, the DBSCAN clustering algorithm is used for clustering, where the neighborhood radius is 0.1m, the minimum number of neighborhood points is 10, and the metric distance is the Euclidean distance between two point clouds. The DBSCAN clustering algorithm is a well-known technology and will not be described in detail here.

[0062] Calculate the degree of dispersion of the coordinate values ​​of all point clouds in each cluster on the X-axis, Y-axis, and Z-axis, respectively;

[0063] In this embodiment, since a point cloud is a coordinate point in space, the degree of dispersion is measured by calculating the standard deviation of the coordinate values ​​of all point clouds in each cluster on the X-axis, the standard deviation of the coordinate values ​​on the Y-axis, and the standard deviation of the coordinate values ​​on the Z-axis. As other implementation methods, implementers may use other methods of the prior art, such as variance, etc. This embodiment does not impose any special restrictions on this.

[0064] The result of positively fusing the dispersion of all axes within each cluster is taken as the distribution dispersion of each cluster.

[0065] In this embodiment, the specific process of forward fusion is as follows: the mean of the dispersion of all axes within each cluster is taken as the distribution dispersion of each cluster; as another implementation, the implementer can calculate the sum of the dispersion of all axes within each cluster as the distribution dispersion of each cluster.

[0066] Obtain the segmentation threshold of the distribution dispersion of all clusters, and mark the clusters with distribution dispersion greater than the segmentation threshold as obstacles;

[0067] In this embodiment, the Otsu's method is used to obtain the segmentation threshold. Otsu's method is a well-known technique and will not be described in detail here. As other implementation methods, implementers may use other methods of the prior art, such as cross-validation. This embodiment does not impose any special restrictions on this.

[0068] It should be noted that the distribution dispersion measures the distribution dispersion relationship between point clouds within a cluster. The larger the value, the more three-dimensional the object is, and the more likely it is to be an obstacle that needs to be avoided.

[0069] Since emergency evacuation robots need to provide evacuation guidance services for personnel, in addition to ensuring the normal passage of robots, it is also necessary to ensure the normal evacuation passage of personnel. Since the robot only moves on the ground, the precise height of obstacles has little impact on whether it can pass. Therefore, in order to simplify the obstacle model in long tunnels, the point cloud corresponding to the obstacles is mapped onto the XOY plane for analysis.

[0070] Secondly, considering the enclosed space of long tunnels and the ever-changing nature of fire scenes, both dynamic and static obstacles exist when emergency evacuation robots enter the scene for rescue and evacuation guidance. Therefore, the emergency evacuation robots need to possess autonomous obstacle avoidance and emergency response capabilities to ensure their own safety and the safety of other rescue personnel. Thus, it is necessary to identify both static and dynamic obstacles, specifically:

[0071] Project the point cloud of all obstacles at the current moment onto the horizontal plane to generate the corresponding mapping points;

[0072] It should be noted that the horizontal plane refers to the XOY plane of the spatial coordinate system.

[0073] Obtain the centroid of all mapped points of each obstacle in the horizontal plane;

[0074] And count the number of all point clouds corresponding to each obstacle;

[0075] Choose one obstacle from all obstacles at the current moment and one obstacle from all obstacles at the previous moment, and form an obstacle pair;

[0076] Calculate the distance between the centers of mass of each obstacle pair and record it as the distance traveled;

[0077] In this embodiment, the distance is calculated by measuring the Euclidean distance between the centroids of each obstacle pair. Euclidean distance is a well-known technique and will not be described in detail here. As for other implementation methods, the implementer can set it according to the actual situation.

[0078] Calculate the difference in the number of each pair of obstacles and record it as the quantity difference;

[0079] Calculate the difference in the distribution dispersion between each pair of obstacles, and denote it as the distribution difference;

[0080] In this embodiment, the absolute value of the difference in the number of each obstacle pair is calculated and denoted as the quantity difference; the absolute value of the difference in the distribution dispersion of each obstacle pair is calculated and denoted as the distribution difference.

[0081] It should be noted that if there are M obstacles at time t and N obstacles at time t-1, then the number of obstacle pairs is M×N.

[0082] The correlation coefficients of each obstacle pair were negatively correlated with the movement distance, the difference in number, and the difference in distribution, respectively.

[0083] It should be noted that a negative correlation means that the dependent variable decreases as the independent variable increases, and increases as the independent variable decreases.

[0084] In this embodiment, the movement distance, quantity differences, and distribution differences of each obstacle pair are used to form an association vector. A comprehensive evaluation algorithm is used to comprehensively evaluate the association vectors of all obstacle pairs to obtain the evaluation score of each obstacle pair. The evaluation score is then normalized using the maximum-minimum normalization method and used as the association coefficient. Since all elements in the association vector are minimal indicators, the smaller the value, the more similar the obstacle at the current moment is to the obstacle at the previous moment, and the greater the probability that they are the same obstacle. Therefore, all elements in the association vector are positively oriented, transforming the minimal indicators into minimal indicators, and then the Topsis algorithm is used for comprehensive evaluation. The Topsis algorithm and the maximum-minimum normalization method are well-known technologies and will not be described in detail here. As other implementation methods, implementers can use other methods of existing technology, such as the analytic hierarchy process, etc. This embodiment does not impose any special restrictions on this.

[0085] It should be noted that the smaller the movement distance, the closer the two obstacles are in spatial position; the smaller the difference in quantity, the more stable the change in the number of obstacle point clouds observed in the two observations; the smaller the difference in distribution, the more similar the three-dimensional spatial morphology of the two obstacles; the larger the obtained correlation coefficient, the more similar the obstacle at the current moment is to the obstacle at the previous moment, and the greater the possibility that they belong to the same obstacle.

[0086] Based on the descending order of correlation coefficients, obstacle pairs with correlation coefficients higher than a preset threshold are iteratively identified as matching pairs. Each matching pair represents an obstacle pair with a unique matching relationship. The specific process is as follows:

[0087] All obstacle pairs are sorted in descending order of their correlation coefficients to form a discrimination sequence;

[0088] If the correlation coefficient of the first obstacle pair in the discrimination sequence is greater than the preset threshold, then a matching pair is marked, and all obstacle pairs containing one of the obstacles in the matching pair are removed from the discrimination sequence. The discrimination sequence is then updated. Based on the updated discrimination sequence, the operation is repeated until the correlation coefficient of the obstacle pairs in the discrimination sequence is less than or equal to the preset threshold. Finally, the obstacle pairs in the discrimination sequence are marked as non-matching pairs.

[0089] In this embodiment, the preset threshold value is 0.85. As for other implementation methods, the implementer can set it according to the actual situation.

[0090] It should be noted that a matching pair represents the correspondence of the same obstacle at two consecutive moments.

[0091] Due to the occurrence of fires, car accidents, disasters, etc. in long tunnels, the density of obstacles may be relatively high. Often, due to multiple obstacles being close together or obstructing each other, there will be the phenomenon of old obstacles being lost and new obstacles appearing. If the old obstacles are continuously tracked, a lot of computing resources will be required, and sudden changes in position are likely to occur, which is not conducive to the robot's evacuation guidance. Therefore, the old obstacles are not tracked.

[0092] For all matching pairs and all non-matching pairs, for any obstacle at the current moment, if it does not successfully match with an obstacle at the previous moment, then the obstacle is a newly added obstacle; if it successfully matches and the movement distance between the two successfully matched obstacles is equal to 0, then the obstacle is a static obstacle; if it successfully matches and the movement distance between the two successfully matched obstacles is greater than 0, then the obstacle is a dynamic obstacle.

[0093] Furthermore, due to the significant differences in uncertainty and risk levels among different types of obstacles, it is necessary to construct bounding boxes for obstacles to define their risk areas in order to achieve safe and efficient navigation and obstacle avoidance. Therefore, based on the type of obstacle at the current moment, its bounding box is constructed as follows:

[0094] For all mapping points of the newly added obstacle in the horizontal plane, a two-stage bounding box construction strategy is adopted. First, a directed bounding box is constructed, and then an axis-aligned bounding box is constructed based on the directed bounding box as the two-dimensional bounding box of the newly added obstacle.

[0095] In this embodiment, both the oriented bounding box (OBB) and the axis-aligned bounding box (AABB) are well-known technologies and will not be described in detail here.

[0096] For dynamic obstacles, the speed of the dynamic obstacle is calculated based on its distance between the current time and the previous time. The position of all point clouds on the dynamic obstacle in the next time is predicted to generate a predicted point cloud. The point cloud on the dynamic obstacle at the current time and the predicted point cloud are merged. The merged point cloud is projected onto all mapping points in the horizontal plane. A two-stage bounding box construction strategy is adopted. First, a directed bounding box is constructed, and then an axis-aligned bounding box is constructed based on the directed bounding box as the two-dimensional bounding box of the dynamic obstacle.

[0097] For all mapping points of a static obstacle in the horizontal plane, construct a directed bounding box as the two-dimensional bounding box of the static obstacle.

[0098] It should be noted that, due to the incomplete and highly uncertain information about newly added obstacles, robots need to consider higher fault tolerance when guiding evacuation, planning routes, and autonomously avoiding obstacles. Therefore, the bounding box range should be relatively large. For dynamic obstacles, it is necessary to predict their movement position and consider the position of dynamic obstacles in the next moment to prevent collisions between emergency evacuation robots or personnel and dynamic obstacles. Therefore, a large bounding box is constructed for the entire area where dynamic obstacles are located at the current moment and the next moment. For static obstacles, whose movement probability is low, a directed bounding box is used to fit the shape of the obstacle, thereby giving emergency evacuation robots more passage space.

[0099] It should be noted that since the emergency evacuation robot only moves on the ground, in order to ensure the efficiency of bounding box construction, bounding boxes can be constructed for the mapping points of obstacles in the horizontal plane, thereby constructing a two-dimensional bounding box. For other scenarios, when the height of obstacles needs to be considered, a three-dimensional bounding box can be constructed. In this embodiment, bounding boxes are constructed for all mapping points of obstacles in the horizontal plane to reduce computing resources.

[0100] Thus, we have obtained the 2D bounding boxes of different types of obstacles at the current moment.

[0101] Step 3: Calculate the obstacle threat level of each sector region by using the proportion of the boundary of the two-dimensional bounding box in each sector region and the distance between the boundary and the robot; evaluate the deviation of the direction of each sector region from the direction of the robot to the destination, calculate the passage guidance value of each sector region, and obtain the passage priority of each sector region based on the average level of the ambient temperature in the sector region and the obstacle threat level.

[0102] Furthermore, during the emergency evacuation of personnel in long tunnels by robots, the robots need to perform path planning and evacuation guidance in real time. The artificial potential field method, as an efficient, fast, and adaptable algorithm, is of great significance for improving planning efficiency, reducing computational complexity, and generating better paths.

[0103] The basic idea of ​​artificial potential field is to assume that the robot is a point mass, discretize the surrounding environment, establish the gravitational force field generated by the target position on the robot and the repulsive force field generated by the obstacle on the robot, and combine them linearly to form artificial potential field. In potential field, the robot is subjected to the repulsive force of the obstacle and the gravitational force of the target position, and moves towards the target position under its action until it reaches the target position.

[0104] Due to the complex internal structure of long tunnels, the uncertainty of local perception and the location of fire sources, robots are often highly affected by local obstacles when using the artificial potential field method for path planning. This causes the robot to move gradually near the obstacles, resulting in the phenomenon of the robot "walking along the edge" near the obstacles. Furthermore, because long tunnels are confined spaces, emergency evacuation robots are prone to stalling when facing concave obstacles.

[0105] Furthermore, the flowchart of the method for obtaining the passage priority of each sector area provided in this embodiment is as follows: Figure 2 As shown.

[0106] Based on the above analysis, the obstacle threat level is calculated by analyzing the distance between the boundary of the obstacle within the field of view and the robot, specifically as follows:

[0107] Sample all edges of the 2D bounding box to obtain the boundary points;

[0108] In this embodiment, each edge is sampled with a sampling accuracy of 0.01m, where the distance between two adjacent points is 0.01m. As for other implementation methods, the implementer can set the accuracy according to the actual situation.

[0109] Divide the robot's field of vision at the current moment into multiple sector-shaped areas;

[0110] In this embodiment, the azimuth angle of the robot's field of view is 120 degrees. Therefore, the field of view is divided into 10 sector areas at intervals of 12 degrees. As for other implementation methods, the implementer can set them according to the actual situation.

[0111] Calculate the distance between each boundary point within each sector and the robot's location, and record it as the relative distance;

[0112] In this embodiment, the distance is calculated by measuring the Euclidean norm between each boundary point within each sector and the robot's location. The calculation of the Euclidean norm is a well-known technique and will not be elaborated upon here.

[0113] Calculate the percentage of all boundary points within each sector region relative to the total field of view;

[0114] The obstacle threat level and proportion of each sector region are positively correlated, while they are negatively correlated with the relative distance to all boundary points;

[0115] It should be noted that a positive correlation means that the dependent variable increases as the independent variable increases and decreases as the independent variable decreases; a negative correlation means that the dependent variable decreases as the independent variable increases and increases as the independent variable decreases.

[0116] In this embodiment, the sum of the relative distances of all boundary points within each sector area is used as the ratio between the percentage and the sum, which is taken as the obstacle threat level. In other embodiments, the implementer may also calculate the reciprocal of the relative distance, and multiply the sum of the reciprocals of the relative distances of all boundary points within each sector area by the percentage, which is taken as the obstacle threat level.

[0117] It should be noted that the smaller the relative distance, the closer the obstacles in the fan-shaped area are to the robot, and the higher the threat to the robot's passage; the larger the proportion, the more complex and dense the outlines of the obstacles in the fan-shaped area, reflecting that the difficulty and risk of the emergency evacuation robot passing in this direction are relatively high; the greater the obstacle threat level, the more timely the robot needs to avoid obstacles in the fan-shaped area, and the extremely high risk of its passage.

[0118] Secondly, as the fire continues to spread, the smoke and high temperatures caused by the fire may affect the safety of personnel behind the robot. Therefore, by analyzing the changes in ambient temperature within the fan-shaped area, the temperature threat level is calculated, specifically:

[0119] The temperature threat level of each sector is the result of positively fusing the ambient temperatures of all collection points within each sector.

[0120] In this embodiment, the specific process of forward fusion is as follows: the average ambient temperature of all collection points in each sector area is used as the temperature threat level of each sector area; as another implementation method, the implementer can calculate the sum of the ambient temperatures of all collection points in each sector area as the temperature threat level of each sector area.

[0121] It should be noted that the greater the temperature threat level, the higher the temperature in the sector area, reflecting the greater the danger in the sector area, and the less likely the robot should guide personnel in that direction.

[0122] Furthermore, the consistency between the direction corresponding to each sector area and the direction of the target location is analyzed, and the traffic guidance value is calculated, specifically:

[0123] Starting from the robot's location, the direction from the angle bisector of each sector to the arc is the travel direction for each sector.

[0124] The direction from the robot's location to the target point is taken as the target direction;

[0125] It should be noted that the target point is the location where the robot needs to guide people to their final destination.

[0126] Calculate the cosine of the angle between the travel direction and the target direction, and perform a positive mapping on it to serve as the travel guidance value for each sector area;

[0127] In this embodiment, the specific process of positive mapping is as follows: the sum of the cosine value and the value 1 is used as the guiding value. In other implementation methods, the implementer can perform positive mapping using an exponential function. Let's assume the cosine value is denoted as... ,Will The result is used as the pass guide value, where, For an exponential function with the natural constant as the base, through a positive mapping process, the result of the pass guide value is greater than 0.

[0128] It should be noted that the larger the cosine value, the larger the resulting guidance value, indicating that the orientation of the sector area is more consistent with the direction of the final target point. By choosing to pass through this sector area, the robot can approach the final target with the highest efficiency.

[0129] The passage priority of each sector area is negatively correlated with the obstacle threat level and temperature threat level, but positively correlated with the passage guidance value.

[0130] In this embodiment, the obstacle threat level, temperature threat level, and passage guidance value of all sector regions are comprehensively evaluated, and the comprehensive score is normalized using the softmax function and used as the passage priority for each sector region. The specific process is as follows: the obstacle threat level, temperature threat level, and passage guidance value of all sector regions are normalized using the maximum-minimum value normalization method. Since the passage guidance value is a very large indicator, while the obstacle threat level and temperature threat level are very small indicators, the normalized obstacle threat level and temperature threat level are positively processed to transform the very small indicators into very large indicators. The Topsis algorithm is then used for comprehensive evaluation. The Topsis algorithm and the softmax function are well-known techniques and will not be described in detail here.

[0131] It should be noted that the higher the passage priority, the lower the passage risk for the emergency evacuation robot within the sector area, and the closer it gets to the target point by moving in the direction corresponding to the sector area, reflecting that the corresponding sector area is currently the safest and most efficient passage direction.

[0132] At this point, the passage priority of each sector area is obtained.

[0133] Step 4: Based on the priority of passage, heuristic gravity is generated for each sector area based on the gravity of the artificial potential field algorithm. The robot moves according to the direction of the resultant force of the gravity, repulsion and heuristic gravity of the artificial potential field algorithm to guide the safe evacuation of personnel.

[0134] The artificial potential field method reduces the algorithmic complexity of path planning, shortens planning time, improves planning efficiency, and smooths the motion trajectory by constructing a virtual repulsive potential field surrounding the obstacle and an attractive potential field about the target point. Specifically, the artificial potential field method uses a gravitational potential field function to describe the influence of the target point on the robot. The gravitational force of the potential field is proportional to the distance between the robot and the target point. It can be represented as: ; in, For the coefficient gain of the gravitational potential field, The location of the robot. The location of the target point; express Point and The Euclidean distance between them;

[0135] It should be noted that this gravitational potential function characterizes the magnitude of the gravitational force exerted by the target point on the robot; the closer to the target point, the smaller the gravitational force.

[0136] Based on the concept of potential energy field, the gravitational force in an artificial potential field is the negative gradient attracting the change in the potential field, representing the direction of the fastest change in the gravitational potential field function. The gravitational formula is: ; Repulsive field function It is expressed as follows: ; in, For the coefficient gain of the repulsive potential field, The distance between the obstacle and the robot. Let be the obstacle action threshold. According to the repulsive field formula, obstacles greater than the obstacle action threshold will not generate a repulsive force on the robot.

[0137] The repulsive force is calculated from the negative gradient of the repulsive potential field, and its direction is from the obstacle towards the robot. The formula for the repulsive force is: ; The above are the gravitational and repulsive field functions of the artificial potential field algorithm. The overall potential field can be obtained by superimposing the gravitational potential field of the target point and the repulsive potential field of the obstacle.

[0138] It should be noted that the artificial potential field algorithm is a well-known technology and will not be elaborated upon here.

[0139] The gravitational force of the target point on the robot is obtained by the artificial potential field algorithm; the product of the passage priority of each sector and the gravitational force is used as the heuristic gravitational force value of each sector.

[0140] Obtain the heuristic gravity with the magnitude of the force based on the heuristic gravity value of each sector region and the direction of the force based on the direction of passage;

[0141] Based on the heuristic gravity of all sector regions, the gravitational force of the target point on the robot, and the repulsive force of the obstacle on the robot, the resultant force is calculated. The robot moves according to the direction of the resultant force to generate an obstacle avoidance route to guide people in the long tunnel to evacuate safely.

[0142] Based on the same inventive concept as the above method, this application embodiment also provides a long tunnel emergency evacuation robot, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described methods for guiding evacuation of a long tunnel emergency evacuation robot.

[0143] It should be understood that, although Figure 1The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but may be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0144] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0145] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of this application, without departing from the content of the technical solution of this application, shall fall within the protection scope of the technical solution of this application.

Claims

1. A method for guiding the evacuation of robots in emergency evacuation in long tunnels, characterized in that, The method includes the following steps: By fusing LiDAR information and depth images, a point cloud of the environment in front of the robot is generated in real time, and the ambient temperature of each collection point within the field of view in front of the robot is collected, dividing the field of view into multiple fan-shaped areas. Cluster the point cloud, analyze the distribution dispersion of the point cloud within each cluster, filter the clusters and mark the obstacles; Based on the distance relationship between obstacles between adjacent time points, the difference in the number of points and the difference in their discrete distribution, the same obstacle at adjacent time points is matched, and different types of obstacles are identified according to the matching results and the movement of the same obstacle at adjacent time points, and a differentiated two-dimensional bounding box is constructed for it in the horizontal plane. The obstacle threat level of each sector is calculated by using the proportion of the boundary of the two-dimensional bounding box in each sector and the distance between the boundary and the robot. The deviation of the direction of each sector from the direction of the robot to the destination is evaluated, the passage guidance value of each sector is calculated, and the passage priority of each sector is obtained based on the average level of ambient temperature and obstacle threat level within the sector. Based on passage priority, and on the basis of the gravitational force of the artificial potential field algorithm, heuristic gravity is generated for each sector area. The robot moves according to the direction of the resultant force of the gravitational force, repulsive force and heuristic gravity of the artificial potential field algorithm, guiding the safe evacuation of personnel.

2. The evacuation guidance method for an emergency evacuation robot in a long tunnel as described in claim 1, characterized in that, The process of filtering clusters and marking obstacles includes: Calculate the degree of dispersion of the coordinate values ​​of all point clouds in each cluster on the X-axis, Y-axis, and Z-axis, respectively; The dispersion of each cluster is the result of a positive fusion of the dispersion of all axes within each cluster; Obtain the segmentation threshold of the distribution dispersion of all clusters, and mark all point clouds in clusters with a distribution dispersion greater than the segmentation threshold as an obstacle.

3. The evacuation guidance method for an emergency evacuation robot in a long tunnel as described in claim 2, characterized in that, The matching of the same obstacle at adjacent time points includes: Project the point cloud corresponding to all obstacles on the local environment map at the current moment onto the horizontal plane to generate corresponding mapping points; obtain the centroid of all mapping points of each obstacle in the horizontal plane; count the number of all point clouds corresponding to each obstacle; From all obstacles at the current moment and all obstacles at the previous moment, select one obstacle each as an obstacle pair; calculate the distance between the centroids of each obstacle pair, denoted as the movement distance; calculate the difference in the number of obstacles between each obstacle pair, denoted as the quantity difference; calculate the difference in the distribution dispersion between each obstacle pair, denoted as the distribution difference. The correlation coefficients of each obstacle pair were negatively correlated with the movement distance, the difference in number, and the difference in distribution, respectively. Based on the descending order of the correlation coefficient, obstacle pairs with a correlation coefficient higher than a preset threshold are iteratively identified as matching pairs, wherein there is a unique matching relationship between the obstacle pairs represented by the matching pairs.

4. The evacuation guidance method for an emergency evacuation robot in a long tunnel as described in claim 3, characterized in that, The identification of different types of obstacles includes: based on the matching results of obstacle pairs, for any obstacle at the current moment, if it does not successfully match with an obstacle at the previous moment, then the obstacle is a newly added obstacle; if it successfully matches and the movement distance between the two successfully matched obstacles is equal to 0, then the obstacle is a static obstacle; if it successfully matches and the movement distance between the two successfully matched obstacles is greater than 0, then the obstacle is a dynamic obstacle.

5. The evacuation guidance method for an emergency evacuation robot in a long tunnel as described in claim 4, characterized in that, The construction of differentiated two-dimensional bounding boxes in the horizontal plane includes: For all mapping points of the newly added obstacle in the horizontal plane, a two-stage bounding box construction strategy is adopted to obtain the two-dimensional bounding box of the newly added obstacle; For dynamic obstacles, the speed of the dynamic obstacle is calculated based on its distance between the current time and the previous time, the position of the dynamic obstacle in the next time is predicted, the point clouds of the dynamic obstacle at the two positions are merged, the merged point cloud is projected onto all mapping points in the horizontal plane, and a two-stage bounding box construction strategy is adopted to obtain the two-dimensional bounding box of the dynamic obstacle. Construct a two-dimensional bounding box for all mapping points of a static obstacle in the horizontal plane.

6. The evacuation guidance method for an emergency evacuation robot in a long tunnel as described in claim 1, characterized in that, The calculation of the obstacle threat level for each sector includes: Sampling is performed on all edges of the 2D bounding box to obtain each boundary point; the distance between each boundary point in each sector and the robot's location is calculated and recorded as the relative distance; the proportion of the number of all boundary points in each sector to the total field of view is calculated. The obstacle threat level is positively correlated with the proportion, but negatively correlated with the relative distance to all boundary points.

7. The evacuation guidance method for an emergency evacuation robot in a long tunnel as described in claim 1, characterized in that, The calculation of the passage guidance value for each sector area includes: taking the robot's location as the starting point, the direction from the angle bisector of each sector area to the arc is the passage direction for each sector area; taking the direction from the robot's location to the target point's location as the target direction; calculating the cosine of the angle between the passage direction and the target direction, and performing a positive mapping on it as the passage guidance value for each sector area.

8. The evacuation guidance method for an emergency evacuation robot in a long tunnel as described in claim 1, characterized in that, The process of obtaining the passage priority of each sector region includes: The temperature threat level of each sector is the result of positively fusing the ambient temperatures of all collection points within each sector. The passage priority is negatively correlated with obstacle threat level and temperature threat level, but positively correlated with passage guidance value.

9. The evacuation guidance method for an emergency evacuation robot in a long tunnel as described in claim 7, characterized in that, The process of generating heuristic gravity for each sector region includes: using an artificial potential field algorithm to obtain the gravitational force of the target point on the robot; using the product of the passage priority of each sector region and the gravity as the heuristic gravity value of each sector region; and obtaining the heuristic gravity with the heuristic gravity value of each sector region as the magnitude of the force and the passage direction as the direction of the force.

10. A long tunnel emergency evacuation robot, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the evacuation guidance method for an emergency evacuation robot in a long tunnel as described in any one of claims 1-9.

Images