Task completion degree evaluation method for amphibious unmanned vehicle cross-domain login algorithm

Abstract

The invention relates to a task completion degree evaluation method for a cross-domain login algorithm of an amphibious unmanned vehicle, belongs to the technical field of task evaluation, and solves the problem of lack of an evaluation method for the task completion degree of the cross-domain login algorithm in the prior art. The method comprises the following steps: constructing a plurality of evaluation data packets, and randomly issuing the evaluation data packets to a to-be-evaluated algorithm; the evaluation data packet comprises environment data and task data; the to-be-evaluated algorithm is an amphibious unmanned vehicle cross-domain login algorithm; calculating environment complexity and task complexity according to the evaluation data packet of the to-be-evaluated algorithm; obtaining attitude data, positioning data and identified shoreline data returned when the to-be-evaluated algorithm executes the login task, and calculating the basic completion degree of the to-be-evaluated algorithm according to the attitude data, the positioning data and the identified shoreline data; and obtaining a comprehensive task completion degree of the to-be-evaluated algorithm based on the environment complexity, the task complexity and the basic completion degree. The objective and comprehensive evaluation of the task completion degree of the amphibious unmanned vehicle cross-domain login algorithm is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of task evaluation, and in particular to a method for evaluating the degree of task completion of a cross-domain landing algorithm for an amphibious unmanned vehicle. Background Art

[0002] Amphibious vehicles are special vehicles that have both water and land driving functions and can complete underwater detection, offshore landing and other functions. They play an irreplaceable role in unique geographical environments. With the development of technologies such as machine learning, deep learning, and reinforcement learning, amphibious vehicles are moving towards lightweight and unmanned. Amphibious unmanned vehicles are equipped with control algorithms that can perform visual detection, scene recognition, path planning and control the operation of amphibious unmanned vehicles through sensors. The cross-domain landing mission of amphibious unmanned vehicles is one of the important tasks of unmanned amphibious vehicles, and it is also one of the working conditions that unmanned amphibious vehicles must achieve. The cross-domain landing of unmanned amphibious vehicles generally includes two major working conditions: "surface navigation" and "landing". The "surface navigation" working condition generally has the following requirements: one is normal navigation on the water surface, that is, controlling the vehicle to move at a certain speed and direction, and at the same time, it is necessary to be able to control the state of the vehicle in real time for natural conditions (waves, wind, etc.); the second is obstacle avoidance, that is, the ability to identify and avoid other participants on the water surface and obstacles such as reefs. The unmanned algorithm installed in the unmanned amphibious vehicle controls the operation of the vehicle under two working conditions and the switching of working conditions. The landing process of an amphibious vehicle generally includes the following steps: first, shoreline identification and path planning, that is, identifying the landing area and planning the landing path; then judging the appropriate time to switch working conditions, switching the vehicle working condition, that is, switching from "water working condition" to "land and water working condition". This step will make corresponding adjustments to the vehicle's suspension, clutch, sluice gate, etc., so that it changes from a complete water driving state to a landing state; the last step is landing, that is, controlling the vehicle to complete the landing in a certain direction and reach the designated location. Amphibious unmanned vehicles also need to complete some special tasks during the cross-domain landing process, which in practice include but are not limited to survey and patrol, information collection, disaster relief, personnel and material transportation, and fire support.

[0003] Currently, there are unmanned algorithms for completing corresponding login tasks, but there is no evaluation method for the task completion of unmanned algorithms in cross-domain login scenarios, making it difficult to quantitatively evaluate the performance of different algorithms. Summary of the invention

[0004] In view of the above analysis, an embodiment of the present invention aims to provide a method for evaluating the task completion of an amphibious unmanned vehicle cross-domain landing algorithm, so as to solve the problem of the existing lack of an evaluation method for the task completion of an amphibious unmanned vehicle cross-domain landing algorithm.

[0005] On the one hand, an embodiment of the present invention provides a method for evaluating the task completion of an amphibious unmanned vehicle cross-domain landing algorithm, comprising the following steps:

[0006] Construct multiple evaluation data packets and randomly distribute the evaluation data packets to the algorithm to be evaluated; the evaluation data packets include environmental data and task data; the algorithm to be evaluated is an amphibious unmanned vehicle cross-domain landing algorithm;

[0007] Calculate the environment complexity and task complexity based on the evaluation data package of the algorithm to be evaluated;

[0008] Obtaining the posture data, positioning data and identified shoreline data returned by the algorithm to be evaluated when performing the landing task according to the task data, and calculating the basic completion degree of the algorithm to be evaluated according to the posture data, positioning data and identified shoreline data;

[0009] The comprehensive task completion of the algorithm to be evaluated is obtained based on the environment complexity, task complexity and basic completion.

[0010] Based on the further improvement of the above method, the basic completion degree of the algorithm to be evaluated is calculated according to the posture data, positioning data and recognized shoreline data, including:

[0011] The number of mission points passed by the amphibious unmanned vehicle is calculated based on the real-time positioning data returned by the algorithm to be evaluated to obtain the mission point completion degree;

[0012] According to the real-time positioning data returned by the algorithm to be evaluated, it is determined whether the amphibious unmanned vehicle has a collision risk and the traffic safety degree is obtained;

[0013] The shoreline identification integrity and shoreline identification continuity are calculated according to the shoreline data identified by the algorithm to be evaluated in each inspection frame, and the shoreline identification completion degree is obtained;

[0014] According to the heading angle and positioning data of the amphibious vehicle-free person returned by the algorithm to be evaluated at the landing point, the landing direction deviation and landing position deviation are calculated;

[0015] According to the pitch angle and roll angle returned by the algorithm to be evaluated in each inspection frame, the safety degree of the navigation attitude of each inspection frame is calculated to obtain the safety degree of the navigation attitude of the algorithm to be evaluated;

[0016] Obtaining the traffic efficiency of the algorithm to be evaluated according to the test completion time of the algorithm to be evaluated;

[0017] The basic completion degree of the algorithm to be evaluated is obtained based on the completion degree of the mission points, traffic safety, shoreline identification completion degree, landing direction deviation degree, landing position deviation degree, navigation posture safety degree and traffic efficiency.

[0018] Based on the further improvement of the above method, the following formula is used to calculate the shoreline identification integrity and shoreline identification continuity according to the shoreline data returned by the algorithm to be evaluated in each inspection frame to obtain the shoreline identification completion degree:

[0019] M3=M 3-1 +M 3-2

[0020]

[0021] Among them, M3 represents the shoreline recognition completion of the algorithm to be evaluated, N IF represents the number of inspection frames, M 3-1 Indicates the completeness of shoreline identification, M 3-2 Indicates the continuity of shoreline identification, X i,max represents the maximum horizontal coordinate of the real coastline of the i-th survey frame, X i,min represents the minimum horizontal coordinate of the real coastline of the i-th survey frame, x i,max represents the maximum horizontal coordinate of the coastline of the i-th inspection frame identified by the algorithm to be evaluated, x i,min represents the minimum horizontal coordinate of the coastline of the i-th survey frame identified by the algorithm to be evaluated, μ standard Represents the standard coastline point cloud spacing variance, μ i Represents the point cloud spacing variance of the coastline identified by the algorithm to be evaluated.

[0022] Based on the further improvement of the above method, the following formula is used to calculate the navigation attitude safety of each investigation frame according to the pitch angle and roll angle returned by the algorithm to be evaluated in each investigation frame, and the navigation attitude safety of the algorithm to be evaluated is obtained:

[0023]

[0024] Among them, M6 represents the navigation attitude safety of the algorithm to be evaluated, N IF represents the number of inspection frames, σ i1 represents the pitch angle of the i-th survey frame, σ1 lim represents the limit value of the pitch angle, σ i2 represents the roll angle of the i-th inspection frame, σ2 lim represents the roll angle limit, ω1 and ω2 represent weights, α i1 and α i2 is the intermediate parameter.

[0025] Based on the further improvement of the above method, the following formula is used to calculate the landing direction deviation and landing position deviation according to the heading angle and positioning data of the amphibious vehicle-free person returned by the algorithm to be evaluated at the landing point:

[0026]

[0027] Among them, M4 represents the landing direction deviation, M5 represents the landing position deviation, (X0, Y0) represents the target landing point coordinates, (X self ,Y self) represents the coordinate position of the amphibious unmanned vehicle when it completes landing, δ represents the heading angle of the amphibious unmanned vehicle when it completes landing, θ represents the expected landing direction in the global coordinate system, and R d Represents the error radius.

[0028] Based on the further improvement of the above method, the following method is used to determine whether the amphibious unmanned vehicle has a collision risk according to the real-time positioning data returned by the algorithm to be evaluated, and the traffic safety degree is obtained:

[0029] According to the distance between the positioning data returned by the algorithm to be evaluated in each inspection frame and the center point of the obstacle, it is judged whether the amphibious unmanned vehicle collides with the obstacle in the inspection frame;

[0030] The traffic safety is obtained according to the following formula:

[0031]

[0032] Among them, M2 represents the traffic safety, m 2i represents the communication security of the i-th inspection frame, N RBi represents the total number of risk obstacles in the i-th inspection frame, n RBi N represents the number of times the amphibious unmanned vehicle collides with risk obstacles in the i-th inspection frame. IF Indicates the number of inspection frames.

[0033] Based on the further improvement of the above method, the task complexity is calculated according to the evaluation data package of the algorithm to be evaluated, including:

[0034] Calculate the average proportion of semantic label types in each scene frame, and obtain the label type complexity based on the average proportion of semantic label types in each scene frame;

[0035] The complexity of the number of inspection frames is obtained according to the number of inspection frames in the task data;

[0036] Obtaining the guidance point complexity according to the guidance point information in the task data;

[0037] The complexity of the landable area is obtained according to the area ratio of the landable area and the area ratio of the passable area in the task data;

[0038] The complexity of the slope of the landing point is obtained according to the slope of the landing point;

[0039] The task complexity is obtained based on the complexity of label type, complexity of inspection frame number, complexity of guidance point, complexity of landing area and complexity of landing point slope.

[0040] Based on the further improvement of the above method, the complexity of the guidance point is obtained according to the guidance point information in the task data, including:

[0041] The dimension complexity of the guiding point is obtained according to the coordinate dimension of the guiding point;

[0042] According to the number of guiding points, the complexity of the number of guiding points is obtained;

[0043] Calculating the curvature of each guiding point, and obtaining the curvature complexity of the guiding point according to the ratio of the number of guiding points whose curvature exceeds a first threshold to the total number of guiding points;

[0044] The guidance point complexity is obtained based on the guidance point dimension complexity, the guidance point dimensional complexity and the guidance point curvature complexity.

[0045] Based on the further improvement of the above method, the curvature of each guide point is calculated in the following way:

[0046] For each guide point, four nearest guide points are found, and data fitting is performed on the guide point and its four nearest guide points to obtain a fitting function;

[0047] According to the formula Calculate the curvature of the guide point, where K represents the curvature of the guide point and y" represents the curvature of the fitting function at x. h The second derivative value of the point, y' represents the fitting function at x h The first derivative value of the point, x h Indicates the X-axis coordinate value of the guide point.

[0048] Based on the further improvement of the above method, the following formula is used to calculate the average semantic label type ratio of each scene frame:

[0049]

[0050] Among them, N F Indicates the number of scene frames, N tagi Indicates the number of label types for the i-th scene frame.

[0051] Compared with the prior art, the present invention constructs multiple evaluation data packets and randomly sends evaluation data packets to the algorithm to be evaluated, thereby evaluating multiple algorithms to be evaluated at the same time, and calculates the basic completion degree of the algorithm to be evaluated according to the posture data, positioning data and identified shoreline data returned by the algorithm to be evaluated, and calculates the environmental complexity and task complexity of the landing task performed by the algorithm to be evaluated according to the evaluation data packet of the algorithm to be evaluated, and obtains the comprehensive task completion degree of the algorithm to be evaluated according to the environmental complexity, task complexity and basic completion degree, thereby not only examining the performance of the algorithm to be evaluated in completing the landing task, but also considering the environment and task complexity of its landing task, thereby comprehensively and objectively quantitatively evaluating the performance of the algorithm to be evaluated.

[0052] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings are only used for the purpose of illustrating specific embodiments and are not to be considered as limiting the present invention. In the entire drawings, the same reference symbols represent the same components;

[0054] Figure 1 This is a flow chart of a method for evaluating the task completion of an amphibious unmanned vehicle cross-domain landing algorithm according to an embodiment of the present invention;

[0055] Figure 2 This is a composition diagram of the cross-domain login task completion evaluation indicators in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0057] Environmental factors such as wind, waves, and currents can affect the heading, attitude, and power control of amphibious unmanned vehicles. The complexity of the scene in the landing mission affects the scene understanding and path planning of the amphibious unmanned vehicle algorithm, which in turn affects the completion of its landing mission. Therefore, the inspection of the completion of the landing mission not only examines whether the algorithm can control the amphibious unmanned vehicle to safely reach the designated landing point, but also examines the environmental factors in the process of completing the mission and the complexity of the mission itself.

[0058] A specific embodiment of the present invention discloses a method for evaluating the task completion of an amphibious unmanned vehicle cross-domain landing algorithm. Figure 1 As shown, the following steps are included:

[0059] S1. Construct multiple evaluation data packages and randomly distribute the evaluation data packages to the algorithm to be evaluated; the evaluation data packages include environmental data and task data; the algorithm to be evaluated is an amphibious unmanned vehicle cross-domain landing algorithm;

[0060] S2. Calculate the environment complexity and task complexity according to the evaluation data package of the algorithm to be evaluated;

[0061] S3, obtaining the posture data, positioning data and identified shoreline data returned by the algorithm to be evaluated when performing the landing task according to the task data, and calculating the basic completion degree of the algorithm to be evaluated according to the posture data, positioning data and identified shoreline data;

[0062] S4. Based on the environment complexity, task complexity and basic completion, the comprehensive task completion of the algorithm to be evaluated is obtained.

[0063] It should be noted that the algorithm to be evaluated completes the evaluation task based on the received evaluation data packet, and returns posture data, positioning data and identified shoreline data in accordance with the evaluation task requirements during the task completion process. The evaluation end evaluates the task completion degree of the algorithm to be evaluated based on the data returned by the algorithm to be evaluated.

[0064] Compared with the prior art, the task completion evaluation method for the cross-domain landing algorithm of an amphibious unmanned vehicle provided in this embodiment constructs multiple evaluation data packets and randomly sends evaluation data packets to the algorithm to be evaluated, thereby simultaneously evaluating multiple algorithms to be evaluated, and calculating the basic completion of the algorithm to be evaluated based on the posture data, positioning data and identified shoreline data returned by the algorithm to be evaluated, and calculating the environmental complexity and task complexity of the landing task performed by the algorithm to be evaluated based on the evaluation data packet of the algorithm to be evaluated, and obtaining the comprehensive task completion of the algorithm to be evaluated based on the environmental complexity, task complexity and basic completion, thereby not only examining the performance of the algorithm to be evaluated in completing the landing task, but also considering the environment and task complexity of its landing task, thereby comprehensively and objectively quantitatively evaluating the performance of the algorithm to be evaluated.

[0065] During implementation, the evaluation data package includes environmental data and login task data. The environmental data and login task data of different evaluation data packages may be different. The evaluation algorithm can be evaluated through an individual evaluation data package, or the algorithm to be evaluated can be made to perform the evaluation tasks of multiple evaluation data packages. The evaluation results of multiple evaluation data packages can be combined to obtain the final evaluation result of the algorithm to be evaluated.

[0066] During implementation, the environmental data in the assessment data package include water surface conditions, water surface environmental characteristic levels, shoreline change levels, shoreline terrain categories, weather types and lighting condition levels.

[0067] When an unmanned amphibious vehicle is sailing on the sea, the state of the water surface, including the height of the waves and the water flow rate, will have a great impact on the heading, attitude and power control of the unmanned vehicle, and may even cause the vehicle to capsize. Therefore, the water surface state is an aspect that reflects the complexity of the environment. The complexity corresponding to the sea state level is shown in Table 1:

[0068] Table 1 Water surface state complexity standard

[0069]

[0070] The water surface conditions corresponding to each sea state level are as follows:

[0071] Level 0: Wave height 0 meters. The sea is calm, the water surface is as smooth as a mirror, or there are only swells, and the boat is stationary;

[0072] Level 1: Wave height 0-0.1m. Ripples or swells and ripples exist at the same time, small waves are like fish scales, and there are no sprays.

[0073] Level 2: Wave height 0.1-0.5 meters. The waves are small, the crests begin to break, but the waves are not white but glassy.

[0074] Level 3: Wave height 0.5-1.25 meters. The waves are not big, but very eye-catching, with broken crests, shiny foam, and sometimes whitecaps.

[0075] Level 4: Wave height 1.25-2.5 meters. The waves have a distinct shape and whitecaps are formed everywhere.

[0076] Level 5: Wave height 2.5-4 meters. There are tall wave crests, and the waves occupy a large area on the crests. The wind begins to cut off the waves on the crests. The sea surface is violently turbulent and navigation is difficult.

[0077] Level 6 and above: Wave height is greater than 4 meters. The sea is rough, the air is filled with white spray, water droplets and foam, visibility is reduced, and navigation is very dangerous.

[0078] The water surface environment features include natural environment features and artificial features. Natural features include natural objects such as reefs and floating ice on the sea surface and hydrological information on the sea surface. Artificial features include ships and buoys. These features will affect the navigation of amphibious unmanned vehicles. For example, reefs are static obstacles, and ships are dynamic obstacles that need to be avoided. Therefore, the water surface environment features are classified and given the complexity, as shown in Table 2:

[0079] Table 2 Water surface feature complexity standards

[0080]

[0081] The corresponding status of each category is as follows:

[0082] Artificial features are obvious: various buoys and large and small ships are clearly visible on the sea surface;

[0083] Obvious natural features: The natural features on the sea surface, such as reefs, pumice, and floating ice, are clear, and the ocean hydrological information is obvious;

[0084] No features: None of the above information is obvious.

[0085] Affected by the tidal effect, the position of the shoreline will change over time, especially during high tide and low tide, when the shoreline position changes greatly. This dynamic change of the shoreline will affect the shoreline recognition of the amphibious unmanned vehicle, thereby affecting the moment of working condition conversion. Therefore, the complexity of the shoreline boundary is analyzed and calculated. The coast is classified by the degree of position change and the complexity is given in Table 3:

[0086] Table 3 Coastline boundary complexity standards

[0087]

[0088] The quantitative basis for classifying shorelines is as follows:

[0089] No-change shoreline boundary: The change in the horizontal position of the shoreline within 24 hours is less than 10m;

[0090] Slightly changing shoreline boundary: The horizontal position of the shoreline changes between 10m and 100m within 24 hours;

[0091] Significantly changing shoreline boundary: The horizontal position of the shoreline changes by more than 100m within 24 hours.

[0092] The impact of shoreline topography on landing missions mainly exists in the following two aspects: First, the slope of the coastline determines whether the amphibious vehicle can land smoothly. A steeper slope may cause the vehicle to get stuck. Second, the solidity of the ground also affects the landing process. Softer mud and sand may cause the vehicle to get stuck, while artificially paved slopes are easier to land. Therefore, the coastline is divided into the following typical categories according to the terrain, and the complexity is given according to the difficulty of landing, as shown in the following table:

[0093] Table 4 Coastline terrain complexity standards

[0094]

[0095] The classification characteristics of various types of coastlines are as follows:

[0096] Artificially paved slope: mostly made of concrete, with a straight road surface and a gentle slope;

[0097] Gravel bank: mainly composed of gravel and sand, forming landforms such as sandbanks and sand dams. The particles of this type of bank material are coarse, and after being washed and transported by sea water, it forms a beach that slopes gently toward the sea;

[0098] Muddy coast: mainly composed of silt and fine sand, with a wide coastal zone, a small slope and a straight coastline;

[0099] Soil bank: mainly composed of mud, often covered with vegetation, often found in rivers, lakes and other water bodies, the angle between the land and the water surface is not large, and the height difference between the land and the water surface is small;

[0100] Bedrock coast: A hard coast with more than 75% of the bottom matrix being rock, including rocky coastal islands and sea cliffs. This type of coast has a steep terrain with a large angle between the land and the sea.

[0101] Weather conditions have an impact on the perception and control of amphibious unmanned vehicles. For example, rain and snow will affect the images collected by the camera to a certain extent, solid particles will greatly affect the recognition accuracy of the lidar, and the wind speed will directly or indirectly affect the vehicle's heading and speed. Here, the weather conditions are analyzed, and the weather factors that affect the navigation of the unmanned vehicle are classified and given the complexity, as shown in Table 5:

[0102] Table 5 Weather state complexity standards

[0103]

[0104] The classification of each weather element is based on the following:

[0105] No wind: wind speed is less than 3m / s;

[0106] Windy: wind speed is between 3m / s and 8m / s;

[0107] Storm: wind speed greater than 8m / s;

[0108] Light rain: 24-hour rainfall is less than 10 mm;

[0109] Moderate rain: 24-hour rainfall is between 10 mm and 50 mm;

[0110] Heavy rain: 24-hour rainfall is between 50 mm and 100 mm;

[0111] Heavy rain: rainfall greater than 100 mm in 24 hours;

[0112] Light snow: 24-hour precipitation is less than 2.4 mm;

[0113] Moderate snow: 24-hour precipitation is between 2.4 mm and 4.9 mm;

[0114] Heavy snow: 24-hour precipitation is between 4.9 mm and 9.9 mm;

[0115] Blizzard: 24-hour precipitation is greater than 9.9 mm;

[0116] Light fog: horizontal visibility is greater than or equal to 400m;

[0117] Dense fog: horizontal visibility is less than 400m.

[0118] Lighting conditions will affect the data collection of the image sensor of the unmanned vehicle. Poor lighting conditions will greatly affect the environmental perception ability of the unmanned vehicle, thus hindering planning and control and even causing misjudgment. The lighting conditions are classified according to their complexity, as shown in Table 6:

[0119] Table 6 Complexity standards of lighting conditions

[0120]

[0121] During implementation, the complexity results of each environmental data are weighted and summed to obtain the final environmental complexity.

[0122] Task data includes:

[0123] (1) A set of scene frame sequences for testing, the environment semantic label of each scene frame, and the radar point cloud data corresponding to each scene frame. Each scene frame is identified by a unique timestamp.

[0124] (2) Guidance point information: Guidance points are points that the amphibious unmanned vehicle needs to pass through during the planning process. Their function is to provide path information and guide the algorithm to be evaluated to perform path planning.

[0125] (3) Mission point information, including the positioning information of each mission point. The last mission point is the landing point. The landing point information includes positioning information and slope information. The amphibious unmanned vehicle should pass through every mission point as much as possible.

[0126] (4) Examination frame timestamp: The evaluation task uses certain scene frames as examination frames. When the algorithm to be evaluated executes the scene frame with the corresponding timestamp, it needs to return the information required by the evaluation task, such as posture data and recognized shoreline data.

[0127] The algorithm to be evaluated understands the scene for each scene frame in timestamp order, identifies shoreline information based on radar point cloud data, plans paths based on environmental semantic labels, guide point information, and mission point information in the scene frames, and controls the amphibious unmanned vehicle to perform cross-domain landing tasks.

[0128] It should be noted that the algorithm to be evaluated returns positioning data in real time when performing the login task.

[0129] The complexity of the task will also affect the evaluation results of the evaluation algorithm, so it is necessary to calculate the task complexity. Specifically, the task complexity is calculated based on the evaluation data package of the algorithm to be evaluated, including:

[0130] S21, calculating the average proportion of semantic label types in each scene frame, and obtaining label type complexity according to the average proportion of semantic label types in each scene frame;

[0131] The proportion of the number of semantic labels contained in each scene frame to the total semantic labels of the shoreline expresses the scene richness of the target shoreline of the landing mission scene. The more semantic labels there are, the more complex the shoreline scene is and the higher the difficulty of scene understanding is. The total semantic labels of the shoreline landing mission include the following eight categories: water area, shoreline, easy landing, difficult landing, impossible landing, risk obstacle, inaccessible obstacle, pedestrian, etc.

[0132] The following formula is used to calculate the average semantic label type ratio of each scene frame:

[0133]

[0134] Among them, N F Indicates the number of scene frames, N tagi Indicates the number of label types for the i-th scene frame.

[0135] According to the average proportion of semantic label types in each scene frame, that is, the proportion of the number of semantic labels in a single shoreline frame to the total number of labels, the label type complexity results are obtained according to Table 7.

[0136] Table 7. Complexity of semantic labels for a single frame of the coastline

[0137]

[0138] S22, obtaining the complexity of the number of inspection frames according to the number of inspection frames in the task data;

[0139] The more frames are examined, the more scenes the landing mission examines, and the higher the difficulty of scene understanding. The complexity corresponding to the number of examination frames is shown in Table 8:

[0140] Table 8 Coastline survey frame complexity standard

[0141]

[0142] S23, obtaining the guidance point complexity according to the guidance point information in the task data;

[0143] The guidance point is the point that the amphibious unmanned vehicle needs to pass through during the planning process. Its function is to provide path information and guide the amphibious unmanned vehicle to plan the path. The information it provides will have a great impact on the difficulty of the unmanned vehicle's path planning.

[0144] In addition, when sailing on water, the amphibious vehicle relies on the adjustment of the water gate and rudder to control the heading, and in principle can achieve on-the-spot turning. However, considering the influence of water flow, wind speed, etc., the actual completion of large-angle turning is very challenging for the control accuracy of the unmanned vehicle, so the curvature of the guide point is used as an important indicator affecting the control complexity.

[0145] Therefore, the guidance point complexity is obtained according to the guidance point information in the task data, including:

[0146] The dimension complexity of the guiding point is obtained according to the coordinate dimension of the guiding point;

[0147] According to the number of guiding points, the complexity of the number of guiding points is obtained;

[0148] Calculating the curvature of each guiding point, and obtaining the curvature complexity of the guiding point according to the ratio of the number of guiding points whose curvature exceeds a first threshold to the total number of guiding points;

[0149] The guidance point complexity is obtained based on the guidance point dimension complexity, the guidance point dimensional complexity and the guidance point curvature complexity.

[0150] The dimension of the guide point includes the position coordinates of the guide point in the three directions of x, y, and z in the global coordinate system, as well as the vehicle heading information at the guide point. The dimensional complexity of the guide point indicates the comprehensiveness of the guide point information provided by the evaluation task. The more guide point dimensions provided in the evaluation data package, that is, the more comprehensive the guide point information, the lower the path planning difficulty. The specific complexity can be shown in Table 9:

[0151] Table 9 Guidance point dimensional complexity standard

[0152]

[0153]

[0154] The number of guide points reflects the global path planning difficulty provided by the assessment. The more guide points there are, the lower the path planning difficulty. The number of guide points can be represented by the average interval of guide points in the task data, calculated according to the following formula:

[0155]

[0156] Among them, N GP is the total number of guiding points, L i is the distance between every two guiding points (the i-th guiding point and the i+1-th guiding point). The specific complexity of the number of guiding points is shown in Table 10:

[0157] Table 10: Complexity table of number of guiding points

[0158]

[0159] Specifically, the curvature of each guide point is calculated in the following way:

[0160] For each guide point, four nearest guide points are found, and data fitting is performed on the guide point and its four nearest guide points to obtain a fitting function;

[0161] According to the formula Calculate the curvature of the guide point, where K represents the curvature of the guide point and y" represents the curvature of the fitting function at x. h The second derivative value of the point, y' represents the fitting function at x h The first derivative value of the point, x h Indicates the X-axis coordinate value of the guide point.

[0162] During implementation, for the guide point (x h ,y h ) and the four nearest guide points of the guide point are fitted by the least square method to obtain a fitting function. During implementation, the fitting function is determined according to the distribution of the five points, for example, it can be a logarithmic function.

[0163] Find the fitting function at x = x h The first-order derivative value y' and the second-order derivative value y" at this time can be used to obtain the curvature of the guiding point according to the above curvature calculation formula.

[0164] During implementation, the guidance points whose curvature exceeds the first threshold are difficulty points. The first threshold can be set to 0.05. According to the ratio of the number of difficulty points to the total number of guidance points, the curvature complexity of the guidance points is obtained by referring to Table 13.

[0165] Table 11 Guidance point curvature complexity standard

[0166]

[0167] The guiding point complexity is obtained by adding the guiding point dimensional complexity, the guiding point dimension complexity and the guiding point curvature complexity.

[0168] S24, obtaining the complexity of the landable area according to the area ratio of the landable area and the area ratio of the passable area in the task data;

[0169] For most coastline environments, only a part of the coast can be used for landing missions. Therefore, for path planning, the complexity of the landing area will directly affect the difficulty of the task. Among them, the complexity of the landing area is composed of two aspects: the proportion of the landing area and the continuity of the landing area.

[0170] The proportion of the landing area represents the ratio of the landing area to the entire target beach area, which directly determines the area where path planning can be performed and the richness of path planning. It can be quantitatively represented by the proportion R1 of the number of grids labeled "shoreline", "easy to land" and "difficult to land" in all scene frames in the mission data to the total number of grids.

[0171] The proportion of the landable area directly affects the difficulty of path planning in the landing task. The larger the proportion of the landable area, the lower the complexity and the simpler the path planning. The complexity is calculated linearly according to the proportion, and its specific standards are given in Table 12.

[0172] Table 12 Login area ratio complexity standard

[0173]

[0174] The continuity of the landing area refers to whether the shoreline of the target beach is a continuous and unified area, that is, whether it is blocked by obstacles or inaccessible areas. The higher the shoreline continuity, the lower the difficulty of path planning during landing. The continuity complexity can be evaluated by the ratio R2 of the total number of grids labeled "risk obstacles" and "inaccessible obstacles" within the shoreline range to the number of grids labeled "shoreline". The specific indicators and evaluation criteria are shown in Table 13.

[0175] Table 13 Continuity standards for landing areas

[0176]

[0177] S25, obtaining the complexity of the slope of the landing point according to the slope of the landing point;

[0178] When landing on the beach, the slope of the landing point tests the control of the throttle, clutch, etc. The slope value of the landing point is quantified, and the specific complexity is shown in Table 14.

[0179] Table 14 Landing point slope complexity standards

[0180]

[0181] S26. Obtain the task complexity according to the complexity of the label type, the complexity of the number of inspection frames, the complexity of the guidance point, the complexity of the landing area and the complexity of the landing point slope.

[0182] The task complexity is obtained by adding the complexity of label type, complexity of inspection frame number, complexity of guidance point, complexity of landing area and complexity of landing point slope.

[0183] The basic completion degree of the algorithm to be evaluated is calculated based on the posture data, positioning data and recognized shoreline data returned by the algorithm to be evaluated, including:

[0184] S31, calculating the number of mission points passed by the amphibious unmanned vehicle according to the positioning data returned in real time by the algorithm to be evaluated to obtain the completion degree of the mission points;

[0185] During the landing process, the amphibious unmanned vehicle needs to complete the designated mission at the designated mission point, and the amphibious unmanned vehicle should be able to pass through every mission point. Therefore, the positioning data returned by the algorithm to be evaluated in real time is used to determine whether the amphibious unmanned vehicle has passed through the mission point, and the mission point completion degree of the algorithm to be evaluated is obtained based on the ratio of the number of mission points passed to the total number of mission points. N finishis the number of completed mission points, N mission is the total number of mission points.

[0186] During implementation, based on the positioning data returned by the algorithm to be evaluated, if the distance difference between the positioning returned by the algorithm to be evaluated and the positioning of a certain task point is less than a preset threshold, such as 2m, then it is considered that the algorithm to be evaluated passes through the task point.

[0187] S32, judging whether the amphibious unmanned vehicle has a collision risk according to the positioning data returned in real time by the algorithm to be evaluated, and obtaining a traffic safety degree;

[0188] The vehicle safety index is used to evaluate the safety of the vehicle and the environment participants in the process of completing the mission, which is of great significance in the actual landing mission. During implementation, the obstacles that affect the driving of the amphibious unmanned vehicle in each scene frame mainly include: risk obstacles and impassable obstacles.

[0189] The following method is used to determine whether the amphibious unmanned vehicle has a collision risk based on the positioning data returned in real time by the algorithm to be evaluated, and the traffic safety degree is obtained:

[0190] According to the distance between the positioning data returned by the algorithm to be evaluated in each inspection frame and the center point of the obstacle, it is judged whether the amphibious unmanned vehicle collides with the obstacle in the inspection frame;

[0191] The traffic safety is obtained according to the following formula:

[0192]

[0193] Among them, m 2i represents the communication security of the i-th inspection frame, N RBi represents the total number of risk obstacles in the i-th inspection frame, n RBi N represents the number of times the amphibious unmanned vehicle collides with risk obstacles in the i-th inspection frame. IF represents the number of inspection frames, and M2 represents the degree of traffic safety.

[0194] During implementation, if the distance difference between the location returned by the algorithm to be evaluated and the center point of an obstacle is less than a preset threshold, it is considered that the amphibious unmanned vehicle has collided with the obstacle. The thresholds corresponding to different types of obstacles may be different.

[0195] S33, calculating the shoreline identification integrity and shoreline identification continuity according to the identified shoreline data returned by the algorithm to be evaluated in each inspection frame, and obtaining the shoreline identification completion degree;

[0196] The completeness of shoreline recognition plays an important role in landing path planning and vehicle control during the landing of amphibious unmanned vehicles. This test mainly examines the integrity and continuity of shoreline recognition. The integrity mainly examines whether the identified shoreline boundary is correct, and the continuity mainly examines whether there are breakpoints in the identification of the shoreline.

[0197] Specifically, the shoreline identification integrity and shoreline identification continuity are calculated according to the shoreline data returned by the algorithm to be evaluated in each inspection frame to obtain the shoreline identification completion degree:

[0198] M3=M 3-1 +M 3-2

[0199]

[0200] Among them, M3 represents the shoreline recognition completion of the algorithm to be evaluated, N IF represents the number of inspection frames, M 3-1 Indicates the completeness of shoreline identification, M 3-2 Indicates the continuity of shoreline identification, X i,max represents the maximum horizontal coordinate of the real coastline of the i-th survey frame, X i,min represents the minimum horizontal coordinate of the real coastline of the i-th survey frame, x i,max represents the maximum horizontal coordinate of the coastline of the i-th inspection frame identified by the algorithm to be evaluated, x i,min represents the minimum horizontal coordinate of the coastline of the i-th survey frame identified by the algorithm to be evaluated, μ standard Represents the standard coastline point cloud spacing variance, μ i Represents the point cloud spacing variance of the coastline identified by the algorithm to be evaluated.

[0201] S34, calculating the landing direction deviation and the landing position deviation according to the heading angle and positioning data of the amphibious vehicle-free person returned by the algorithm to be evaluated at the landing point;

[0202] In a landing mission, in order to facilitate the next step planning after landing, an expected landing direction is usually set. The greater the deviation from the expected landing direction, the lower the completion rate.

[0203] The last mission point of the landing mission is the final target point of the landing, that is, the landing point. The amphibious unmanned vehicle must stop at the target point after landing. The greater the position deviation, the lower the completion rate.

[0204] Specifically, the landing direction deviation and landing position deviation are calculated using the following formula based on the heading angle and positioning data of the amphibious vehicle-free person returned by the algorithm to be evaluated at the landing point:

[0205]

[0206] Among them, M4 represents the landing direction deviation, M5 represents the landing position deviation, (X0, Y0) represents the target landing point coordinates, (X self ,Y self ) represents the coordinate position of the amphibious unmanned vehicle when it completes landing, δ represents the heading angle of the amphibious unmanned vehicle when it completes landing, θ represents the expected landing direction in the global coordinate system, and R d Represents the error radius.

[0207] S35, calculating the navigation attitude safety of each investigation frame according to the pitch angle and roll angle returned by the algorithm to be evaluated in each investigation frame, and obtaining the navigation attitude safety of the algorithm to be evaluated;

[0208] When driving on the water, the navigation attitude of the amphibious vehicle has an important impact on the safety of the vehicle. Therefore, the control and evaluation of the navigation attitude has a far-reaching impact on the safety of the landing cross-domain mission. The navigation attitude safety mainly examines the longitudinal and heel angles.

[0209] Specifically, the following formula is used to calculate the navigation attitude safety of each inspection frame according to the pitch angle and roll angle returned by the algorithm to be evaluated in each inspection frame, and the navigation attitude safety of the algorithm to be evaluated is obtained:

[0210]

[0211] Among them, M6 represents the navigation attitude safety of the algorithm to be evaluated, N IF represents the number of inspection frames, σ i1 represents the pitch angle of the i-th survey frame, σ1 lim represents the limit value of the pitch angle, σ i2 represents the roll angle of the i-th inspection frame, σ2 lim represents the roll angle limit, ω1 and ω2 represent weights, α i1 and α i2 is the intermediate parameter.

[0212] During implementation, for example, the pitch angle limit value is set to 20°, and the roll angle limit value is set to 15°.

[0213] S36, obtaining the traffic efficiency of the algorithm to be evaluated according to the test completion time of the algorithm to be evaluated;

[0214] The speed of an amphibious unmanned vehicle has a significant impact on the difficulty of planning and control, because the faster the speed, the shorter the reaction time of the unmanned vehicle to the scene, and the higher the requirements for the sensitivity of the sensor, the efficiency of information transmission and the response of the actuator. At the same time, environmental factors will also affect the speed of the amphibious unmanned vehicle. In order to more comprehensively evaluate the completion effect of the landing mission and avoid sacrificing speed to improve other indicators, the traffic efficiency of the unmanned vehicle is examined.

[0215] The traffic efficiency M7 is calculated using the following formula:

[0216]

[0217] Where T is the manual driving time, and t is the test completion time of the algorithm to be evaluated.

[0218] S37. Based on the task point completion, traffic safety, shoreline identification completion, landing direction deviation, landing position deviation, navigation posture safety and traffic efficiency, the basic completion of the algorithm to be evaluated is obtained.

[0219] The basic completion degree is obtained by adding up the mission point completion degree, passage safety degree, shoreline identification completion degree, landing direction deviation degree, landing position deviation degree, navigation attitude safety degree and passage efficiency.

[0220] For each algorithm to be evaluated, its environmental complexity and task complexity are used as weights, and multiplied by the basic completion to obtain the comprehensive task completion of the algorithm to be evaluated. Thus, considering the environmental complexity and task complexity, the landing task completion of the amphibious unmanned vehicle is comprehensively and objectively evaluated to obtain the evaluation results.

[0221] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0222] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for evaluating the task completion of an amphibious unmanned vehicle cross-domain landing algorithm, characterized in that: The following steps are involved: Construct multiple evaluation data packages and randomly distribute the evaluation data packages to the algorithm to be evaluated; the evaluation data packages include environmental data and task data; the algorithm to be evaluated is an amphibious unmanned vehicle cross-domain landing algorithm; Calculate the environment complexity and task complexity based on the evaluation data package of the algorithm to be evaluated; Obtaining the posture data, positioning data and identified shoreline data returned by the algorithm to be evaluated when performing the landing task according to the task data, and calculating the basic completion degree of the algorithm to be evaluated according to the posture data, positioning data and identified shoreline data; The comprehensive task completion of the algorithm to be evaluated is obtained based on the environment complexity, task complexity and basic completion.

2. The method for evaluating the task completion of the cross-domain landing algorithm for an amphibious unmanned vehicle according to claim 1 is characterized in that: The basic completion degree of the algorithm to be evaluated is calculated based on the posture data, positioning data and identified shoreline data, including: The number of mission points passed by the amphibious unmanned vehicle is calculated based on the real-time positioning data returned by the algorithm to be evaluated to obtain the mission point completion degree; Based on the real-time positioning data returned by the algorithm to be evaluated, it is determined whether the amphibious unmanned vehicle has a collision risk and the traffic safety degree is obtained; The shoreline identification integrity and shoreline identification continuity are calculated according to the shoreline data identified by the algorithm to be evaluated in each inspection frame, and the shoreline identification completion degree is obtained; According to the heading angle and positioning data of the amphibious vehicle-free person returned by the algorithm to be evaluated at the landing point, the landing direction deviation and landing position deviation are calculated; According to the pitch angle and roll angle returned by the algorithm to be evaluated in each inspection frame, the safety degree of the navigation attitude of each inspection frame is calculated to obtain the safety degree of the navigation attitude of the algorithm to be evaluated; Obtaining the traffic efficiency of the algorithm to be evaluated according to the test completion time of the algorithm to be evaluated; The basic completion of the algorithm to be evaluated is obtained based on the completion of the mission points, traffic safety, shoreline identification completion, landing direction deviation, landing position deviation, navigation posture safety and traffic efficiency.

3. The method for evaluating the task completion of the cross-domain landing algorithm for an amphibious unmanned vehicle according to claim 2 is characterized in that: The following formula is used to calculate the shoreline identification integrity and shoreline identification continuity based on the shoreline data returned by the algorithm to be evaluated in each inspection frame to obtain the shoreline identification completion degree: M3=M 3-1 +M 3-2 Among them, M3 represents the shoreline recognition completion of the algorithm to be evaluated, N IF represents the number of inspection frames, M 3-1 Indicates the completeness of shoreline identification, M 3-2 Indicates the continuity of shoreline identification, X i,max represents the maximum horizontal coordinate of the real coastline of the i-th survey frame, X i,min represents the minimum horizontal coordinate of the real coastline of the i-th survey frame, x i,max represents the maximum horizontal coordinate of the coastline of the i-th inspection frame identified by the algorithm to be evaluated, x i,min represents the minimum horizontal coordinate of the coastline of the i-th survey frame identified by the algorithm to be evaluated, μ standard Represents the standard coastline point cloud spacing variance, μ i Represents the point cloud spacing variance of the coastline identified by the algorithm to be evaluated.

4. The method for evaluating the task completion of the cross-domain landing algorithm for an amphibious unmanned vehicle according to claim 2 is characterized in that: The following formula is used to calculate the navigation attitude safety of each investigation frame according to the pitch angle and roll angle returned by the algorithm to be evaluated in each investigation frame, and the navigation attitude safety of the algorithm to be evaluated is obtained: Among them, M6 represents the navigation attitude safety of the algorithm to be evaluated, N IF represents the number of inspection frames, σ i1 represents the pitch angle of the i-th survey frame, σ1 lim represents the limit value of the pitch angle, σ i2 represents the roll angle of the i-th inspection frame, σ2 lim represents the roll angle limit, ω1 and ω2 represent weights, α i1 and α i2 is the intermediate parameter.

5. The method for evaluating the task completion of the cross-domain landing algorithm for an amphibious unmanned vehicle according to claim 2 is characterized in that: The following formula is used to calculate the landing direction deviation and landing position deviation based on the heading angle and positioning data of the amphibious vehicle-free person returned by the algorithm to be evaluated at the landing point: Among them, M4 represents the landing direction deviation, M5 represents the landing position deviation, (X0, Y0) represents the target landing point coordinates, (X self ,Y self ) represents the coordinate position of the amphibious unmanned vehicle when it completes landing, δ represents the heading angle of the amphibious unmanned vehicle when it completes landing, θ represents the expected landing direction in the global coordinate system, and R d Represents the error radius.

6. The method for evaluating the task completion of the cross-domain landing algorithm for an amphibious unmanned vehicle according to claim 2 is characterized in that: The following method is used to determine whether the amphibious unmanned vehicle has a collision risk based on the positioning data returned in real time by the algorithm to be evaluated, and the traffic safety degree is obtained: According to the distance between the positioning data returned by the algorithm to be evaluated in each inspection frame and the center point of the obstacle, it is judged whether the amphibious unmanned vehicle collides with the obstacle in the inspection frame; The traffic safety is obtained according to the following formula: Among them, M2 represents the traffic safety, m 2i Indicates the communication security of the i-th inspection frame, N RBi represents the total number of risk obstacles in the i-th inspection frame, n RBi N represents the number of times the amphibious unmanned vehicle collides with risk obstacles in the i-th inspection frame. IF Indicates the number of inspection frames.

7. The method for evaluating the task completion of the cross-domain landing algorithm for an amphibious unmanned vehicle according to claim 1 is characterized in that: Calculate the task complexity based on the evaluation data package of the algorithm to be evaluated, including: Calculate the average proportion of semantic label types in each scene frame, and obtain the label type complexity based on the average proportion of semantic label types in each scene frame; The complexity of the number of inspection frames is obtained according to the number of inspection frames in the task data; Obtaining the guidance point complexity according to the guidance point information in the task data; The complexity of the landable area is obtained according to the area ratio of the landable area and the area ratio of the passable area in the task data; The complexity of the slope of the landing point is obtained according to the slope of the landing point; The task complexity is obtained based on the complexity of label type, complexity of inspection frame number, complexity of guidance point, complexity of landing area and complexity of landing point slope.

8. The method for evaluating the task completion of the cross-domain landing algorithm for an amphibious unmanned vehicle according to claim 1 is characterized in that: The guidance point complexity is obtained according to the guidance point information in the task data, including: The dimension complexity of the guiding point is obtained according to the coordinate dimension of the guiding point; According to the number of guiding points, the complexity of the number of guiding points is obtained; Calculating the curvature of each guiding point, and obtaining the curvature complexity of the guiding point according to the ratio of the number of guiding points whose curvature exceeds a first threshold to the total number of guiding points; The guidance point complexity is obtained based on the guidance point dimension complexity, the guidance point dimensional complexity and the guidance point curvature complexity.

9. The method for evaluating the task completion of the cross-domain landing algorithm for an amphibious unmanned vehicle according to claim 8 is characterized in that: The curvature of each guide point is calculated as follows: For each guide point, find four nearest guide points, and perform data fitting on the guide point and its four nearest guide points to obtain a fitting function; According to the formula Calculate the curvature of the guide point, where K represents the curvature of the guide point, y ,, Indicates that the fitting function is in x h The second derivative value of the point, y , Indicates that the fitting function is in x h The first derivative value of the point, x h Indicates the X-axis coordinate value of the guide point.

10. The method for evaluating the task completion of the cross-domain landing algorithm for an amphibious unmanned vehicle according to claim 7 is characterized in that: The following formula is used to calculate the average semantic label type ratio of each scene frame: Among them, N F Indicates the number of scene frames, N tagi Indicates the number of label types for the i-th scene frame.

Images