A method, system, computer device and medium for planning distributed flyable paths of multiple intelligent agents under communication constraints

Abstract

The present invention belongs to the technical field of unmanned aerial vehicle path planning, and relates to a multi-agent distributed flyable path planning method under communication constraints, comprising the following steps: each agent obtains the surrounding local obstacle parameters and surrounding agent parameter information respectively through its own sensor and its own communication device; each agent selects the optimal local target point by using an artificial potential field model; each agent performs local Dubins path planning according to the optimal local target point; each agent moves for a specified time according to its own Dubins path planning, local obstacle information, and information of other agents within the communication range, and uses an improved RVO obstacle avoidance algorithm, while performing obstacle avoidance between unmanned agents and between unmanned agents and obstacles; each agent performs target arrival detection; if the target is not reached, the current position of the agent is updated. The method solves the problem of unmanned cluster path planning in sudden unknown scenes, where targets and obstacles are highly dynamic and meet dynamic constraints.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicle path planning, and specifically relates to a multi-agent distributed flyable path planning method, system, computer equipment and medium under communication constraints. Background Art

[0002] With the development of modern information communication and artificial intelligence, the application of autonomous unmanned intelligent agents is becoming more and more widespread. Unmanned swarms solve the problem of insufficient ability of single drones / cars to handle complex tasks. They have high robustness and anti-destruction capabilities and are widely used in military and civilian fields. When unmanned swarms perform tasks, they need to plan their own motion paths. Therefore, the problem of collaborative path planning for multiple unmanned intelligent agents has become a current hot issue. Traditional centralized path planning methods are heavily dependent on prior map information, have low computational efficiency, rarely consider the dynamic constraints of unmanned devices, have poor response capabilities to dynamic targets and dynamic obstacles, and have difficulty in dealing with sudden scenarios and unknown environments. In recent years, distributed collaborative path planning methods have become a research hotspot for path planning problems.

[0003] Existing distributed multi-UAV agent path planning methods can be divided into two categories: one is based on deep reinforcement learning, through which each agent continuously conducts trial and error iterations directly with the environment to obtain feedback information, thereby optimizing its own strategy, and the final generated strategy is used to make decisions on its own movement. However, this type of method does not consider the dynamic constraints of the agent, and there are problems such as complicated calculation process, long training time, unstable and difficult to converge training strategy, poor scalability, and difficulty in practical application. The other is based on artificial potential field, by defining the workspace of the agent as a potential field, defining the target and obstacle as low potential energy points and high potential energy points respectively, and using the superimposed potential field force to determine the next movement of the agent, and finally completing the path planning of the agent. This type of method still does not consider the dynamic constraints of the agent, and the obtained trajectory is highly correlated with the potential energy function, and the parameter debugging is complex. Therefore, it is difficult for existing methods to solve the problem of unmanned cluster path planning in real emergency scenarios, where the target and obstacles are highly dynamic and can fly by satisfying dynamic constraints. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-agent distributed flyable path planning method, system, computer equipment and medium under communication constraints to solve the problem of unmanned swarm path planning under sudden scenarios with high dynamics of targets and obstacles and flyable paths that meet dynamic constraints.

[0005] The present invention is achieved through the following technical solutions:

[0006] A multi-agent distributed flyable path planning method under communication constraints comprises the following steps:

[0007] Step 1: Each agent obtains the local environment information around it through its own sensor and obtains the parameters of the local obstacles around it;

[0008] Each agent communicates with other agents within the communication range through its own communication device to obtain parameter information of surrounding agents;

[0009] Step 2: Each agent uses the improved artificial potential field model and its own target information to select the optimal local target point from the surrounding local environment;

[0010] Step 3: According to the optimal local target point in step 2, each agent combines its own dynamic constraints, the position of the current point and the local target point, performs scene detection decomposition, and completes the local Dubins path planning;

[0011] Step 4: Each agent moves at a specified time based on its own Dubins path planning, local obstacle information, and parameter information of other agents within the communication range, using the improved RVO obstacle avoidance algorithm, while avoiding obstacles between unmanned agents and between unmanned agents and obstacles;

[0012] Step 5: Each agent performs target arrival detection; if it has not reached the target, it updates its current position and returns to step 1 to perform the next step of rolling time domain path planning; if it has reached the target, it ends.

[0013] Furthermore, in step 1, each agent obtains the local environment information around it through its own sensor and obtains the parameters of the local obstacles around it, which are specifically:

[0014] Step 101: Acquisition of surrounding environment information:

[0015] Each drone agent a i ,i∈{1,…,N u}, the radius is r i , the pose is p i =(x i ,y i ,θ i ), the detection radius of the sensor carried is R D i θ i is the direction angle of the agent at the current moment; N u is the number of UAV intelligent bodies;

[0016] Then for any obstacle o k =(x k ,y k ,R k ,v k ),o k ∈{1,…,N o},No is the number of obstacles; R k represents the radius of the obstacle, v k Indicates the speed of the obstacle;

[0017] If and only if the following equation is satisfied, it means that the current obstacle o k Can be used by agent a i perceived;

[0018]

[0019] So agent a i The set of perceived obstacles is expressed as:

[0020]

[0021] Furthermore, in step 1, each agent communicates with other agents within the communication range through its own communication device to obtain parameter information of surrounding agents, specifically:

[0022] Each drone agent a i , whose position is p i =(x i ,y i ,θ i ), the communication radius of the communication device carried is R C i , then for any other UAV agent a j , if and only if the following equation is satisfied, it means that agent a i Can obtain agent a j Information j ;

[0023]

[0024] m j ={p j ,v j},v j is the current agent a j speed;

[0025] Agent a i The information set of other agents that can be obtained is expressed as

[0026] Further, step 2 specifically includes the following steps:

[0027] Step 201: Each agent a i Get the current position of the target p T i =(x Ti ,y T i ,), construct the potential field model using the following function:

[0028] The attractive potential of the point to be selected by the agent is defined as U attract (x), the repulsive potential of these obstacles and the adjacent agents to a certain point to be selected by the agent is defined as U repel (x), the potential field of the point x to be selected around each agent can be expressed as:

[0029] U(x)=U attract (x)+U repel (x);

[0030]

[0031] Among them, k attract represents the gravitational potential energy coefficient, x represents the position of a point to be selected by the current agent, d(x,p T i ) represents the Euclidean distance from the current point to be selected to the target point;

[0032]

[0033] K repel represents the repulsive potential energy coefficient, d(x,x′) represents the Euclidean distance from the current point to be selected to the obstacle or other intelligent agent, and d min is the sum of the radius of the current agent and the radius of the corresponding obstacle or other agents;

[0034] When d(x,x′)≤d min When , it means that if the agent moves to x, it will collide with surrounding obstacles or other agents, so the repulsive potential energy is defined as positive infinity;

[0035] Step 202: Generate a set of local optimal points to be selected:

[0036] Each agent obtains the current set running speed v set i , the pose is p i =(x i ,y i ,θ i ), according to the rolling time domain, each path planning needs to plan at least T l Time length path; combined with dynamic constraints, the direction change angle θ of the agent's sampling point can be obtained △ ;

[0037] The sampling method designed is as follows: take the current agent position as the center of the circle, and set i×T l The length is the radius, with θ i ±θ △ The direction angle is limited, and the total number of sampling points is N evenly on the corresponding arc. select The sampling of the selected points is expressed as Ω i ;

[0038] Step 203: Select the optimal local target point:

[0039] Each agent has a candidate point set Ω obtained in step 202. i For each point in the graph, calculate its corresponding potential energy value. i , and its optimal local target point position is

[0040] If the current agent is less than the detection radius R D i When , it is considered that the target can be observed, then the optimal local target point position is the final target point, p local i =p T i .

[0041] Further, step three specifically includes the following steps:

[0042] Step 301, scene detection decomposition:

[0043] Each agent sets the direction of the local optimal target point to p i Point to p local i direction, current pose p i and the local optimal target point p local i The line segment formed by connecting the lines is

[0044] Each agent has an obstacle set O i The position of each obstacle in the circle is the center of the circle, with obstacle R k +Agent i Make a circle with radius, line segment Scenarios that intersect these circles fall into three categories:

[0045] 1. If they do not intersect, use step 302 to perform ordinary Dubins planning;

[0046] 2. There is a single circle secant, and the obstacle avoidance Dubins planning is performed using step 303, and the obstacle to be avoided is O3;

[0047] 3. If there are two or more circles that intersect each other, insert an auxiliary point P between the two intersecting points. A1 ,PA2 , the auxiliary point direction points to the optimal local target point, respectively for p i To P A1 , P A1 To P A2 , P A2 to p local i Perform multiple obstacle avoidance Dubins planning in step 303, and the resulting curve combination is the desired path;

[0048] Step 302: Common Dubins path planning:

[0049] Each agent obtains the minimum curvature radius r according to its current speed and its own dynamic constraints. min , in order to perform ordinary Dubins path planning;

[0050] When the distance between the planned path and the obstacle is less than r i When , the S straight line portion in the Dubins path returns to step 301, the straight line detection starting point is the starting point of the straight line portion of the Dubins path, and the new path obtained replaces the original straight line portion;

[0051] Step 303: Obstacle avoidance Dubins path planning:

[0052] Using the shortest obstacle avoidance Dubins path planning method, each agent obtains the minimum curvature radius r according to its current speed and its own dynamic constraints. min According to the scene detection decomposition result in step 301, the starting point P of the obstacle avoidance Dubins path planning is obtained s and the planned end point P e ;

[0053] Determine the circumnavigation direction based on the azimuth of the line connecting the starting point Ps and the planned end point Pe;

[0054] Then calculate the starting circle and the ending circle position, whose radius is r min , the formula used is as follows:

[0055]

[0056] Among them, the starting point posture is P s (x s ,y s ,θ s ), the end point position is P e (x e ,y e ,θ e ), the center of the starting circle is P cs (x cs ,y cs ), the center of the termination circle isP cf (x cf ,y cf ), when it is flying to the left, the symbol in the equation is "+", and when it is flying to the right, the symbol in the equation is "-"

[0057] After that, calculate the starting circle cutting point P l , obstacle circle entry point P n1 , obstacle circle cut-out point P n2 , End circle entry point P m ;

[0058]

[0059] Among them, the obstacle circle O1(x o1 ,y o1 ,r o1 ), augmented obstacle circle O1′(x o1 ,y o1 ,r o1 ′), the obstacle circle entry point P n1 (x n1 ,y n1 ), the starting circle cuts out point P l (x l ,y l ), augmented obstacle circle tangent point P n1 ′(x n1 ′,y n1 ′); radius of obstacle circle O1 r o1 is the radius of the obstacle R1 and the radius of the agent r i The sum of the radius r of the augmented obstacle circle O1′ o1 ′=R1+r i +r min ;

[0060] Similarly, calculate the obstacle circle cut-out point P n2 and the end circle entry point P m ;

[0061]

[0062] Among them, the obstacle circle O1(x o1 ,y o1 ,r o1 ), augmented obstacle circle O1′(x o1 ,y o1 ,r o1 ′), the obstacle circle cuts out point P n2 (x n2 ,y n2 ), the end circle cuts into the point P m (x m ,y m), augmented obstacle circle tangent point P n2 ′(x n2 ′,y n2 ′); radius of obstacle circle O1 r o1 is the radius of the obstacle R1 and the radius of the agent r i The sum of the radius r of the augmented obstacle circle O1′ o1 ′=R1+r i +r min ;

[0063] The shortest obstacle-avoiding Dubins path is obtained by arc segment Straight line segment Arc Straight line segment Arc composition;

[0064] The arc segment and arc segments The corresponding radius is r min , arc segment The corresponding radius is r o1 , the path points are uniformly sampled, and the pose of a certain point points to the pose of the next path point;

[0065] Step 304: Connect the obtained paths end to end to obtain the current position p of each agent i To the optimal local target point p local i A flyable Dubins path that satisfies its own dynamic constraints.

[0066] Further, step four specifically includes the following steps:

[0067] Step 401: Each agent obtains the optimal speed based on the Dubins path and the current position.

[0068] Each agent is based on its current position p now i , find the path point p closest to the current position from the path obtained in step 4 du1 i , whose next path point is p du2 i , let p now i Point to p du2 i The direction angle is θ path i , then the corresponding preferred velocity direction angle Determined by the following formula:

[0069] θ pref i =λθdu1 i +(1-λ)θ path i ;

[0070] Among them, θ du1 i is the nearest path point p du1 i The corresponding direction angle, λ is the adjustment coefficient;

[0071] Set the speed v according to the current agent set i With direction angle The optimal speed of the current agent can be calculated

[0072] Step 402: Each agent establishes its own speed obstacle space:

[0073] First, each agent obtains local obstacle information O i , use the following formula to calculate the speed obstacle space of the obstacle part:

[0074]

[0075] where p A is the agent position, v A is the current agent speed, v B is the obstacle speed; It is a cone-shaped area;

[0076] Then, each agent uses the parameter information M of other agents within the communication range to i , use the following formula to calculate the speed obstacle space of its agent:

[0077]

[0078] in is the corresponding VO speed obstacle space, v A is the current agent speed, v C is the velocity of the neighboring agent, v′ A is a certain speed in the reciprocal speed obstacle space RVO;

[0079] Each agent will superimpose all the speed obstacle spaces obtained to obtain the complete speed obstacle space Φ corresponding to each agent. i ;

[0080] Step 403: Each agent calculates the final speed v final i :

[0081] Each agent is based on the current posture p nowi , movement speed v now i and its own dynamic constraints, at its own maximum and minimum speed [v min ,v max ] range, combined with the current azimuth angle θ now i and the maximum steering angle θ △now i , perform equally spaced velocity sampling in the enclosed fan-shaped area;

[0082] When the sampled speed set has at least one speed that is not in the speed obstacle space Φ i When the speed is within The speed closest to the Euclidean distance is taken as the current final speed v final i ;

[0083] Step 404: Each agent executes the final speed within the minimum execution interval τ;

[0084] If it has been T since the last Dubins path planning e time, then go to step 5; otherwise, return to step 401.

[0085] Furthermore, if there is no speed obstacle space Φ i When the speed is within, assuming there are 4 candidate speeds {v cand 1 , v cand 2 , v cand 3 , v cand 4}, three speed obstacle spaces Φ i ={φ1,φ2,φ3}, φ1~φ3 are VO or RVO, calculated by the corresponding obstacles / agents respectively;

[0086] Selected speed v cand 1 , v cand 2 , v cand 3 , v cand 4 They are located in the speed obstacle spaces (φ1, φ2), (φ2, φ3), (φ1, φ2), and (φ2) respectively;

[0087] For a candidate speed of an intelligent agent located in a speed obstacle space, the importance description function of the speed obstacle space to the current intelligent agent is defined as follows:

[0088]

[0089] In the formula, p m is the current speed obstacle space φ m The corresponding obstacle / agent position, ‖p m -p i ‖ is the Euclidean distance between the current agent and the corresponding obstacle / agent, R m is the radius of the obstacle / agent, r i is the radius of the current agent; n is the unit direction vector of the current agent pointing to the corresponding obstacle / agent position;

[0090] A candidate speed The motion characteristics preservation function is described as follows:

[0091]

[0092] For an agent a i , its candidate speed The joint calculation function is as follows:

[0093]

[0094] In the formula, K mot and K imp They are motion characteristic weight and importance weight respectively;

[0095] For an intelligent agent a i , joint calculation function The selected speed corresponding to the minimum value is the desired final velocity,

[0096] The present invention also discloses a multi-agent distributed flyable path planning system under communication constraints, comprising a plurality of UAV agents, each of which is equipped with a sensor, a communication module, a planner and an actuator;

[0097] The sensor and communication module are connected to the planner, and the planner is connected to the actuator;

[0098] The planner includes an obstacle parameter acquisition module, an agent parameter acquisition module, an optimal local target point calculation module, a path planning module, an obstacle avoidance module and a detection module;

[0099] The obstacle parameter acquisition module is used for each agent to obtain the surrounding local environment information through its own sensor and obtain the surrounding local obstacle parameters;

[0100] The agent parameter acquisition module is used for each agent to communicate with other agents within the communication range through its own communication device to obtain parameter information of surrounding agents;

[0101] The optimal local target point calculation module is used for each agent to select the optimal local target point from the surrounding local environment by using the improved artificial potential field model and combining its own target information;

[0102] The path planning module is used to perform scene detection decomposition based on the optimal local target point, and each agent combines its own dynamic constraints, the current point and the position of the local target point to complete the local Dubins path planning;

[0103] The obstacle avoidance module is used for each agent to move in time based on its own Dubins path planning, local obstacle information, and parameter information of other agents within the communication range, using the improved RVO obstacle avoidance algorithm, while avoiding obstacles between unmanned agents and between unmanned agents and obstacles;

[0104] The detection module is used by each agent to detect when it has reached the target. If it has not reached the target, it updates its current position and performs the next step of rolling time domain path planning. If it has reached the target, it ends.

[0105] The present invention also discloses a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method when executing the computer program.

[0106] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method are implemented.

[0107] Compared with the prior art, the present invention has the following beneficial technical effects:

[0108] The present invention discloses a multi-agent distributed flyable path planning method under communication constraints. First, each agent only relies on its own local perception and communication data at the local planning moment, without relying on prior map information, and is suitable for unknown environments and emergency scenarios; second, when calculating the path, only the part from the current point to the optimal local target point is calculated, and the amount of calculation is significantly reduced; third, when calculating the path, the self-dynamic constraints are considered to plan the Dubins path that meets the flyable conditions; then, when calculating the local optimal target point, the current target position is obtained in real time, and the dynamic target is effectively dealt with in the next path planning; finally, the improved RVO obstacle avoidance algorithm is used to achieve obstacle avoidance of dynamic obstacles and neighboring agents, thereby improving the robustness of the algorithm. Compared with the existing distributed path planning algorithm, the method proposed by the present invention considers its own dynamic constraints when planning the Dubins path, and meets the flyable conditions. At the same time, the method does not need to be trained for a long time for different scenarios, and can be used after simple debugging, and the scale of the drone agent can be adaptively expanded at any time. In addition, the selection of the optimal local target point significantly reduces the computational complexity of path planning. Finally, the introduction of rolling time domain and RVO obstacle avoidance significantly improves the robustness of this method, thus having good engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0109] Figure 1 It is a framework diagram of the method of the present invention;

[0110] Figure 2 is a flow chart of the method of the present invention;

[0111] Figure 3 This is a schematic diagram of the potential field of the intelligent body in step 201 of the present invention;

[0112] FIG. 4 is a schematic diagram of the decomposition of scene detection in step 301 of the present invention; FIG. 4a shows a line segment does not intersect these circles; Figure 4b shows the line segment There is a single circle tangent to these circles; Figure 4c shows the line segment There are two or more circles tangent to these circles;

[0113] Figure 5 This is a schematic diagram of the obstacle avoidance Dubins planning in step 303 of the present invention;

[0114] Figure 6 This is a schematic diagram of the combined final speed selection method in step 403 of the present invention. DETAILED DESCRIPTION

[0115] The method of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments of the present invention.

[0116] The present invention uses N u UAV intelligent energy body N oTake the distributed flyable path planning under communication constraints in an obstacle scenario as an example. Figure 1 The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0117] like Figure 2 As shown, the present invention comprises the following steps:

[0118] Step 1: Each agent obtains local environmental information through its own sensors and calculates the parameters of local obstacles:

[0119] Each drone agent a i ,i∈{1,…,N u}, the radius is r i , the pose is p i =(x i ,y i ,θ i ), the detection radius of the sensor carried is R D i θ i is the direction angle of the agent at the current moment;

[0120] Then for any obstacle o k =(x k ,y k ,R k ,v k ),o k ∈{1,…,N o}. If and only if the following equation is satisfied, it means that the current obstacle o k Can be used by agent a i Perceived.

[0121]

[0122] So agent a i The set of perceived obstacles is expressed as:

[0123]

[0124] Each agent communicates with other agents within the communication range through its own communication device to obtain parameter information of surrounding agents. Surrounding agent information acquisition:

[0125] Each drone agent a i , whose position is p i =(x i ,y i ,θ i ), the communication radius of the communication device carried is Then for any UAV agent a iand a j , if and only if the following equation is satisfied,

[0126]

[0127] Represents agent a i Can obtain agent a j Information j . m j ={p j ,v j},v j is the current moment a j The speed of agent a i The information set of other agents that can be obtained is expressed as

[0128] Step 2: Each agent uses the improved artificial potential field model and its own target information to select the optimal local target point p from the surrounding local environment. local i ;

[0129] Step 201: Each agent a i Get the current position of the target p T i =(x T i ,y T i ,), construct the potential field model using the following function:

[0130] According to the artificial potential field theory, the target point in the potential field generates gravitational force on the current agent, and obstacles and other agents in the environment generate repulsive force. The combined force of gravitational force and repulsive force affects the next movement of the moving object. Therefore, in the present invention, the position of a certain point to be selected by the current agent is represented by x, and its potential field can be represented by U(x). The target position of the agent is p T i , the attractive potential of the point to be selected by the agent is U attract (x). The obstacles that the agent can perceive and the agent information obtained are O i and M i The repulsive potential of these obstacles and adjacent agents on a certain point to be selected by the agent is defined as U repel (x).

[0131] Therefore, in the present invention, the potential field of the point x to be selected around each agent can be expressed as

[0132] U(x)=U attract (x)+U repel (x)

[0133] The potential energy generated by the target point to the selected point is related to the distance between the two. The greater the distance, the greater the potential energy, and vice versa. When the distance between the two is 0, it means that the target point has been reached. Therefore, the gravitational potential energy is proportional to the distance, which is defined in the present invention as

[0134]

[0135] k attract Represents the gravitational potential energy coefficient, which is set according to the specific scenario. d(x,p T i ) represents the Euclidean distance from the current point to be selected to the target point.

[0136] The surrounding obstacles and surrounding intelligent agents generate a repulsive field on the point to be selected. The magnitude of the repulsive field potential energy is related to the distance between the current point to be selected and the surrounding obstacles and surrounding intelligent agents. The closer the distance, the greater the repulsive potential energy, and vice versa. In the present invention, it is defined as:

[0137]

[0138] K repel Represents the repulsive potential energy coefficient, which is set according to the specific scenario. d(x,x′) represents the Euclidean distance from the current point to be selected to the obstacle or other intelligent body. min is the sum of the radius of the current agent and the radius of the corresponding obstacle or other agents. min , it means that if the agent moves to x, it will collide with surrounding obstacles or other agents, so the repulsive potential energy is defined as positive infinity.

[0139] According to the above formula in the present invention, each agent can complete the definition of the potential field in the surrounding perception area, and the obtained potential field is as follows: Figure 3 shown.

[0140] Step 202: Generate candidate local optimal points: In the present invention, in order to reduce the computational overhead of each planning, a sampling method is used to select candidate local target points. Each agent obtains the current set running speed size v set i , the pose is p i =(x i ,y i ,θ i ), according to the rolling time domain, each path planning needs to plan at least T l Combined with the dynamic constraints, the direction change angle θ of the agent's sampling point can be obtained △ The designed sampling method is specifically described as: taking the current agent position as the center of the circle and v set i ×T l The length is the radius, with θi ±θ △ The direction angle is limited, and the total number of sampling points is N evenly on the corresponding arc. select The resulting set of points to be selected is represented by Ω i .

[0141] Step 203: Select the optimal local target point: Each agent selects the point set Ω obtained in step 302. i For each point in the graph, calculate its corresponding potential energy value. i , and its optimal local target point position is If the current agent is less than the detection radius R D i When , it is considered that the target can be observed, then the optimal local target point is the final target point, that is, p local i =p T i .

[0142] Step 3: According to the optimal local target point in step 203, each agent combines its own dynamic constraints, the position and posture of the current point and the local target point to perform scene detection decomposition and complete local Dubins path planning;

[0143] Step 301, scene detection decomposition: each agent sets the direction of the local optimal target point to p i Point to p local i direction, current pose p i and the local optimal target point p local i The line segment formed by connecting the lines is The present invention uses the scene detection decomposition method shown in FIG4 to fully consider the possible scenes and complete the local path planning.

[0144] In Figure 4, each agent has an obstacle set O i The position of each obstacle in the circle is the center of the circle, with obstacle R k +Agent i Make a circle with radius . Scenarios that intersect these circles fall into three categories:

[0145] 1. No intersection, as shown in FIG4(a), using step 302 to perform ordinary Dubins planning;

[0146] 2. There is a single circle secant, as shown in FIG4( b ), and the obstacle avoidance Dubins planning is performed using step 303, and the obstacle to be avoided is O3;

[0147] 3. There are two or more circles that intersect each other, as shown in Figure 4(c). Insert an auxiliary point P between the two intersecting points. A1 ,P A2 , the auxiliary point direction points to the optimal local target point, respectively for p i To P A1 ,P A1 To P A2 , P A2 to p local i Step 303 is used to perform obstacle avoidance Dubins planning, and the resulting curve combination is the desired path.

[0148] Step 302: Ordinary Dubins path planning: Each agent obtains the minimum curvature radius r according to its current speed and its own dynamic constraints. min , and perform ordinary Dubins path planning. Each agent generates four types of Dubins paths, namely LSL, RSR, LSR, and RSL, according to the starting point and end point positions (R is a right-turn arc, L is a left-turn arc, and S is a straight line). The shortest path among the four Dubins paths is selected as the planned path. When the distance between the planned path and the obstacle is less than r i When , the S straight line part in the Dubins path returns to step 301, the straight line detection starting point is the starting point of the straight line part of the Dubins path, and the new path replaces the original straight line part. The path points are uniformly sampled, and the pose of a certain point is the pose pointing to the next path point.

[0149] Step 303, obstacle avoidance Dubins path planning: In order to obtain the shortest path that meets the obstacle avoidance requirement, the present invention adopts the following method: Figure 5 The shortest obstacle avoidance Dubins path planning method shown in Figure 1. Each agent obtains the minimum curvature radius r according to its current speed and its own dynamic constraints. min According to the scene detection decomposition result in step 301, the starting point P of this obstacle avoidance Dubins path planning is obtained. s and the planned end point P e .

[0150] First determine the direction of the flight. Figure 5 Medium s and P e The azimuth of the connecting line is P s and the obstacle center P o1 The right side of the connecting line azimuth, so the agent needs to fly to the right, otherwise it needs to fly to the left. Then calculate the position of the starting circle and the ending circle, and its radius is r min , the formula used is as follows:

[0151]

[0152] Among them, the starting point posture is Ps (x s ,y s ,θ s ), the end point position is P e (x e ,y e ,θ e ), the center of the starting circle is P cs (x cs ,y cs ), the center of the termination circle is P cf (x cf ,y cf ), when it is flying to the left, the sign in the equation is "+", and when it is flying to the right, the sign in the equation is "-".

[0153] After that, calculate the starting circle cutting point P l , obstacle circle entry point P n1 , obstacle circle cut-out point P n2 , End circle entry point P m The radius of the obstacle circle O1 is r o1 is the radius of the obstacle R1 and the radius of the agent r i The radius r of the augmented obstacle circle O1′ o1 ′=R1+r i +r min . Passing through the center of the starting circle P cs Get a straight line tangent to the augmented obstacle circle, with the tangent point being P n1 '. Line P o1 P n1 ′ intersects with the obstacle circle O1 at P n1 Point. Another point through the center of the starting circle P cs Get with P n1 P n1 ′ Parallel straight lines intersecting the starting circle at P l Point. Due to geometric relationships, we know that quadrilateral P l P n1 P n1 ′P cs is a rectangle, so P l That is the starting point of the circle, P n1 This is the entry point of the obstacle circle.

[0154]

[0155] Among them, the obstacle circle O1(x o1 ,y o1 ,r o1 ), augmented obstacle circle O1′(x o1 ,y o1 ,r o1 ′), the obstacle circle entry point P n1 (xn1 ,y n1 ), the starting circle cuts out point P l (x l ,y l ), tangent point P n1 ′(x n1 ′,y n1 ′).

[0156] Similarly, the obstacle circle cut-out point P can be calculated n2 and the end circle entry point P m .

[0157]

[0158] Among them, the obstacle circle O1(x o1 ,y o1 ,r o1 ), augmented obstacle circle O1′(x o1 ,y o1 ,r o1 ′), the obstacle circle cuts out point P n2 (x n2 ,y n2 ), the end circle cuts into the point P m (x m ,y m ), augmented obstacle circle tangent point P n2 ′(x n2 ′,y n2 ′); radius of obstacle circle O1 r o1 is the radius of the obstacle R1 and the radius of the agent r i The sum of the radius r of the augmented obstacle circle O1′ o1 ′=R1+r i +r min .

[0159] Therefore, the shortest obstacle-avoiding Dubins path is composed of five parts: the arc segment Straight line segment Arc Straight line segment Arc The arc segment and arc segments The corresponding radius is r min , arc segment The corresponding radius is r o1 The path points are uniformly sampled, and the pose of a certain point points to the pose of the next path point.

[0160] Step 304: Connect the obtained paths end to end, and each agent can obtain the current position p i To the optimal local target point p local iA flyable Dubins path that satisfies its own dynamic constraints.

[0161] Step 4: Each agent calculates its own Dubins path and local obstacle information O i , parameter information M of other agents within the communication range i , using the improved RVO obstacle avoidance algorithm, T e Time movement, while avoiding obstacles between unmanned agents and between unmanned agents and obstacles;

[0162] Step 401: Each agent obtains the optimal speed based on the Dubins path and the current position. Each agent is based on its current position p now i , find the path point p closest to the current position from the path obtained in step 4 du1 i , whose next path point is p du2 i Let p now i Point to p du2 i The direction angle is θ path i , then the corresponding preferred velocity direction angle It can be determined by the following formula:

[0163] θ pref i =λθ du1 i +(1-λ)θ path i

[0164] Among them, θ du1 i is the nearest path point p du1 i The corresponding direction angle, λ is the adjustment coefficient, which can generally be taken as 0.95.

[0165] Set the speed v according to the current agent set i With direction angle The preferred speed of the current agent can be calculated.

[0166] Step 402: Each agent establishes its own speed obstacle space: First, each agent obtains local obstacle information O iCalculate the speed obstacle space of the obstacle. According to the speed obstacle space VO concept, the agent will collide with the obstacle if and only if the relative speed of the two is in a cone area with the agent's position as the vertex and the two sides tangent to a circle with the obstacle as the center and the radius of the obstacle and the agent as the radius. For the collision avoidance scenario between the agent and the obstacle, the agent needs to bear all the collision avoidance responsibilities, so only VO needs to be calculated. The agent traverses O i For each obstacle in the obstacle space, use the following formula to calculate the speed of the obstacle part:

[0167]

[0168] where p A is the agent position, v A is the agent speed, v B is the obstacle speed. is the conical area described.

[0169] Then, each agent uses the parameter information M of other agents within the communication range to i Calculate the speed obstacle space of the agent. Since agents can avoid obstacles with each other, RVO calculation is used. i For each surrounding agent, use the following formula to calculate the speed obstacle space of its agent:

[0170]

[0171] in is the corresponding VO speed obstacle space, v A is the current agent speed, v C is the velocity of the neighboring agent, v′ A is a certain speed in the reciprocal speed obstacle space RVO;

[0172] Each agent will superimpose all the speed obstacle spaces obtained to obtain the complete speed obstacle space Φ corresponding to each agent. i .

[0173] Step 403: Each agent calculates the final speed v final i :Each agent is based on the current posture p now i , movement speed v now i and its own dynamic constraints, at its own maximum and minimum speed [v min ,v max ] range, combined with the current azimuth angle θ now i and the maximum steering angle θ △now i, and perform equally spaced velocity sampling in the enclosed fan-shaped area.

[0174] When the sampled speed set has at least one speed that is not in the speed obstacle space Φ i When the speed is within the range, select the speed with v pref i The speed closest to the Euclidean distance is taken as the current final speed v final i .

[0175] If there is no object that is not in the speed obstacle space Φ i When the speed is within the range, the present invention designs a combined final speed selection method that comprehensively considers the motion characteristics and importance. Figure 6 As shown, the current agent prefers speed Maximum and minimum speed limits of the device min ,v max ], current azimuth angle θ now i and the maximum steering angle θ △now i All are marked on the figure.

[0176] In this example, there are 4 speeds to be selected. cand 1 , v cand 2 , v cand 3 , v cand 4}, three speed obstacle spaces Φ i ={φ1,φ2,φ3}, φ1~φ3 are the speed obstacle space VO or reciprocal speed obstacle space RVO described above, which are calculated by the corresponding obstacles / agents. It can be seen from the figure that the selected speed v cand 1 , v cand 2 , v cand 3 , v cand 4 They are located in the speed obstacle space (φ1, φ2), (φ2, φ3), (φ1, φ2), (φ2) respectively. Therefore, for a certain candidate speed of the intelligent agent located in a certain speed obstacle space, the present invention defines the importance description function of the speed obstacle space to the current intelligent agent as follows:

[0177]

[0178] In the formula, p m is the current speed obstacle space φ m The corresponding obstacle / agent position, ‖p m -pi ‖ is the Euclidean distance between the current agent and the corresponding obstacle / agent, R m is the radius of the obstacle / agent, r i is the radius of the current agent. n is the unit direction vector of the current agent pointing to the corresponding obstacle / agent position.

[0179] A candidate speed The motion characteristics preservation function is described as follows:

[0180]

[0181] Therefore, for an agent a i , its candidate speed The joint calculation function is as follows:

[0182]

[0183] In the formula, K mot and K imp They are motion characteristic weight and importance weight respectively. The default value is 1 and can be modified according to actual conditions.

[0184] Therefore, for an agent a i , joint calculation function The selected speed corresponding to the minimum value is the desired final velocity,

[0185] Step 404: Each agent executes the final speed within the minimum execution interval τ. If T has passed since the last Dubins path planning, e time, then go to step 5; otherwise, return to step 401.

[0186] Step 5: Each agent performs target arrival detection. If it has not reached the target, it updates its current position and returns to step 1 to perform the next step of rolling time domain path planning; if it has reached the target, the algorithm ends.

[0187] The present invention also discloses a multi-agent distributed flyable path planning system under communication constraints, such as Figure 1 As shown, it includes multiple intelligent agents, each of which is equipped with a sensor, a communication module, a planner and an actuator;

[0188] The sensor and communication module are connected to the planner, and the planner is connected to the actuator;

[0189] The planner includes an obstacle parameter acquisition module, an agent parameter acquisition module, an optimal local target point calculation module, a path planning module, an obstacle avoidance module and a detection module;

[0190] The obstacle parameter acquisition module is used for each agent to obtain the surrounding local environment information through its own sensor and obtain the surrounding local obstacle parameters;

[0191] The agent parameter acquisition module is used for each agent to communicate with other agents within the communication range through its own communication device to obtain parameter information of surrounding agents;

[0192] The optimal local target point calculation module is used for each agent to select the optimal local target point from the surrounding local environment by using the improved artificial potential field model and combining its own target information;

[0193] The path planning module is used to perform scene detection decomposition based on the optimal local target point, and each agent combines its own dynamic constraints, the current point and the position of the local target point to complete the local Dubins path planning;

[0194] The obstacle avoidance module is used for each agent to move in time based on its own Dubins path planning, local obstacle information, and parameter information of other agents within the communication range, using the improved RVO obstacle avoidance algorithm, while avoiding obstacles between unmanned agents and between unmanned agents and obstacles;

[0195] The detection module is used by each agent to detect when it has reached the target. If it has not reached the target, it updates its current position and performs the next step of rolling time domain path planning. If it has reached the target, it ends.

[0196] The present invention can be implemented in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware based on a multi-agent distributed flyable path planning method under communication constraints. Moreover, the present invention can be implemented in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0197] If the distributed flyable path planning method of multi-agents under communication constraints of the present invention is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. Computer-readable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be realized by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals. Among them, the computer storage medium can be any available medium or data storage device that can be accessed by the computer, including but not limited to magnetic storage (such as floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), etc.), optical storage (such as CD, DVD, BD, HVD, etc.), and semiconductor storage (such as ROM, EPROM, EEPROM, non-volatile memory (NANDFLASH), solid-state drive (SSD)), etc.

[0198] In an exemplary embodiment, a computer device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the multi-agent distributed flyable path planning method based on communication constraints when executing the computer program. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.

[0199] In summary, the present invention designs a distributed multi-agent flyable path planning method under communication constraints. The method is executed in each agent in a distributed manner without prior environmental information. Each agent completely relies on its own local perception and local communication to obtain information about surrounding obstacles and adjacent agents. The rolling time domain concept is used to plan a local Dubins path that meets its own dynamic constraints. When planning the path, the designed scene detection decomposition method is used to complete the path planning. When executing the path, the improved RVO is used to avoid dynamic obstacles and surrounding agents. At the same time, the scale of the drone is dynamically variable, the computing requirements are small, the parameters are easy to adjust and deploy, and it has good engineering application value. It can effectively deal with the problem of flyable path planning for dynamic target obstacles in sudden unknown scenes for large-scale unmanned clusters.

[0200] The present invention adopts a distributed path planning method. Each intelligent agent completely relies on its own local perception and local communication, and uses the rolling time domain idea to plan a local Dubins path that meets the dynamic constraints. When executing the path, the improved RVO is used for dynamic obstacle avoidance, thereby effectively solving the problem of unmanned cluster path planning in emergency scenarios with high dynamics of targets and obstacles and the ability to fly with dynamic constraints.

[0201] The above description is only an embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A multi-agent distributed flyable path planning method under communication constraints, characterized in that: The following steps are involved: Step 1: Each agent obtains the local environment information around it through its own sensor and obtains the parameters of the local obstacles around it; Each agent communicates with other agents within the communication range through its own communication device to obtain parameter information of surrounding agents; Step 2: Each agent uses the improved artificial potential field model and its own target information to select the optimal local target point from the surrounding local environment; Step 3: According to the optimal local target point in step 2, each agent combines its own dynamic constraints, the position of the current point and the local target point, performs scene detection decomposition, and completes the local Dubins path planning; Step 4: Each agent moves at a specified time based on its own Dubins path planning, local obstacle information, and parameter information of other agents within the communication range, using the improved RVO obstacle avoidance algorithm, while avoiding obstacles between unmanned agents and between unmanned agents and obstacles; Step 5: Each agent performs arrival target detection; If the target is not reached, update the current position of the device and return to step 1 to perform the next step of rolling time domain path planning; If the target has been reached, end; Step 2 specifically includes the following steps: Step 201: Each agent a i Get the current position of the target p T i =(x T i ,y T i ,), construct the potential field model using the following function: The attractive potential of the point to be selected by the agent is defined as U attract (x), the repulsive potential of these obstacles and the adjacent agents to a certain point to be selected by the agent is defined as U repel (x), the potential field of the point x to be selected around each agent can be expressed as: U(x)=U attract (x)+U repel (x); Among them, k attract represents the gravitational potential energy coefficient, x represents the position of a point to be selected by the current agent, d(x,p T i ) represents the Euclidean distance from the current point to be selected to the target point; K repel represents the repulsive potential energy coefficient, d(x,x′) represents the Euclidean distance from the current point to be selected to the obstacle or other intelligent agent, and d min is the sum of the radius of the current agent and the radius of the corresponding obstacle or other agents; When d(x,x′)≤d min When , it means that if the agent moves to x, it will collide with surrounding obstacles or other agents, so the repulsive potential energy is defined as positive infinity; Step 202: Generate a set of local optimal points to be selected: Each agent obtains the current set running speed v set i , the pose is p i =(x i ,y i ,θ i ), according to the rolling time domain, each path planning needs to plan at least T l Time length path; combined with dynamic constraints, the direction change angle θ of the agent's sampling point can be obtained △ ; The sampling method designed is as follows: take the current agent position as the center of the circle, and set i ×T l The length is the radius, with θ i ±θ △ The direction angle is limited, and the total number of sampling points is N evenly on the corresponding arc. select The sampling of the selected points is expressed as Ω i ; Step 203: Select the optimal local target point: Each agent has a candidate point set Ω obtained in step 202. i For each point in the graph, calculate its corresponding potential energy value. i , and its optimal local target point position is If the current agent is less than the detection radius R D i When , it is considered that the target can be observed, then the optimal local target point position is the final target point, p local i =p T i .

2. The method for distributed flyable path planning of multiple agents under communication constraints according to claim 1, characterized in that: In step 1, each agent obtains the local environment information around it through its own sensors and obtains the parameters of the local obstacles around it, which are as follows: Step 101: Acquisition of surrounding environment information: Each drone agent a i ,i∈{1,…,N u }, the radius is r i , the pose is p i =(x i ,y i ,θ i ), the detection radius of the sensor carried is R D i ; θ i N is the direction angle of the agent at the current moment; u is the number of UAV intelligent energy bodies; Then for any obstacle o k =(x k ,y k ,R k ,v k ),o k ∈{1,…,N o },N o is the number of obstacles; R k represents the radius of the obstacle, v k Indicates the speed of the obstacle; If and only if the following equation is satisfied, it means that the current obstacle o k Can be used by agent a i perceived; So agent a i The set of perceived obstacles is expressed as:

3. The method for distributed flyable path planning of multiple agents under communication constraints according to claim 1, characterized in that: In step 1, each agent communicates with other agents within the communication range through its own communication device to obtain parameter information of surrounding agents, specifically: Each drone agent a i , whose position is p i =(x i ,y i ,θ i ), the communication radius of the communication device carried is R C i , then for any other UAV agent a j , if and only if the following equation is satisfied, it means that agent a i Can obtain agent a j Information j ; m j ={p j ,v j },v j is the current agent a j The speed of i is the direction angle of the agent at the current moment; Agent a i The information set of other agents that can be obtained is expressed as 4. The method for distributed flyable path planning of multiple agents under communication constraints according to claim 1, characterized in that: Step three specifically includes the following steps: Step 301, scene detection decomposition: Each agent sets the direction of the local optimal target point to p i Point to p local i direction, current pose p i and the local optimal target point p local i The line segment formed by connecting the lines is Each agent has an obstacle set O i The position of each obstacle in the circle is the center of the circle, with obstacle R k +Agent i Make a circle with radius, line segment Scenarios that intersect these circles fall into three categories:

1. If they do not intersect, use step 302 to perform ordinary Dubins planning; 2. There is a single circle secant, and the obstacle avoidance Dubins planning is performed using step 303, and the obstacle to be avoided is O3; 3. If there are two or more circles that intersect each other, insert an auxiliary point P between the two intersecting points. A1 ,P A2 , the auxiliary point direction points to the optimal local target point, respectively for p i To P A1 , P A1 To P A2 , P A2 to p local i Perform multiple obstacle avoidance Dubins planning in step 303, and the resulting curve combination is the desired path; Step 302: Common Dubins path planning: Each agent obtains the minimum curvature radius r according to its current speed and its own dynamic constraints. min , in order to perform ordinary Dubins path planning; When the distance between the planned path and the obstacle is less than r i When , the S straight line portion in the Dubins path returns to step 301, the straight line detection starting point is the starting point of the straight line portion of the Dubins path, and the new path obtained replaces the original straight line portion; Step 303: Obstacle avoidance Dubins path planning: Using the shortest obstacle avoidance Dubins path planning method, each agent obtains the minimum curvature radius r according to its current speed and its own dynamic constraints. min According to the scene detection decomposition result in step 301, the starting point P of the obstacle avoidance Dubins path planning is obtained s and the planned end point P e ; Determine the circumnavigation direction based on the azimuth of the line connecting the starting point Ps and the planned end point Pe; Then calculate the starting circle and the ending circle position, whose radius is r min , the formula used is as follows: Among them, the starting point posture is P s (x s ,y s ,θ s ), the end point position is P e (x e ,y e ,θ e ), the center of the starting circle is P cs (x cs ,y cs ), the center of the termination circle is P cf (x cf ,y cf ), when it is flying to the left, the symbol in the equation is "+", and when it is flying to the right, the symbol in the equation is "-"; After that, calculate the starting circle cutting point P l , obstacle circle entry point P n1 , obstacle circle cut-out point P n2 , End circle entry point P m ; Among them, the obstacle circle O1(x o1 ,y o1 ,r o1 ), augmented obstacle circle O1′(x o1 ,y o1 ,r o1 ′), the obstacle circle entry point P n1 (x n1 ,y n1 ), the starting circle cuts out point P l (x l ,y l ), augmented obstacle circle tangent point P n1 ′(x n1 ′,y n1 ′); radius of obstacle circle O1 r o1 is the radius of the obstacle R1 and the radius of the agent r i The sum of the radius r of the augmented obstacle circle O1′ o1 ′=R1+r i +r min ; Similarly, calculate the obstacle circle cut-out point P n2 and the end circle entry point P m ; Among them, the obstacle circle O1(x o1 ,y o1 ,r o1 ), augmented obstacle circle O1′(x o1 ,y o1 ,r o1 ′), the obstacle circle cuts out point P n2 (x n2 ,y n2 ), the end circle cuts into the point P m (x m ,y m ), augmented obstacle circle tangent point P n2 ′(x n2 ′,y n2 ′); radius of obstacle circle O1 r o1 is the radius of the obstacle R1 and the radius of the agent r i The sum of the radius r of the augmented obstacle circle O1′ o1 ′=R1+r i +r min ; The shortest obstacle-avoiding Dubins path is obtained by arc segment Straight line segment Arc Straight line segment Arc composition; The arc segment and arc segments The corresponding radius is r min , arc segment The corresponding radius is r o1 , the path points are uniformly sampled, and the pose of a certain point points to the pose of the next path point; Step 304: Connect the obtained paths end to end to obtain the current position p of each agent i To the optimal local target point p local i A flyable Dubins path that satisfies its own dynamic constraints.

5. The method for distributed flyable path planning of multiple agents under communication constraints according to claim 1, characterized in that: Step 4 specifically includes the following steps: Step 401: Each agent obtains the optimal speed based on the Dubins path and the current position. Each agent is based on its current position p now i , find the path point p closest to the current position from the path obtained in step 4 du1 i , whose next path point is p du2 i , let p now i Point to p du2 i The direction angle is θ path i , then the corresponding preferred velocity direction angle Determined by the following formula: i pref i =λθ du1 i +(1-λ)θ path i ; Among them, θ du1 i is the nearest path point p du1 i The corresponding direction angle, λ is the adjustment coefficient; Set the speed v according to the current agent set i With direction angle The optimal speed of the current agent can be calculated Step 402: Each agent establishes its own speed obstacle space: First, each agent obtains local obstacle information O i , use the following formula to calculate the speed obstacle space of the obstacle part: where p A is the agent position, v A is the current agent speed, v B is the obstacle speed; It is a cone-shaped area; Then, each agent uses the parameter information M of other agents within the communication range to i , use the following formula to calculate the speed obstacle space of its agent: in is the corresponding VO speed obstacle space, v A is the current agent speed, v C is the velocity of the neighboring agent, v′ A is a certain speed in the reciprocal speed obstacle space RVO; Each agent will superimpose all the speed obstacle spaces obtained to obtain the complete speed obstacle space Φ corresponding to each agent. i ; Step 403: Each agent calculates the final speed v final i : Each agent is based on the current posture p now i , movement speed v now i and its own dynamic constraints, at its own maximum and minimum speed [v min ,v max ] range, combined with the current azimuth angle θ now i and the maximum steering angle θ △now i , perform equally spaced velocity sampling in the enclosed fan-shaped area; When the sampled speed set has at least one speed that is not in the speed obstacle space Φ i When the speed is within The speed closest to the Euclidean distance is taken as the current final speed v final i ; Step 404: Each agent executes the final speed within the minimum execution interval τ; If it has been T since the last Dubins path planning e time, then go to step 5; otherwise, return to step 401.

6. The method for distributed flyable path planning of multiple agents under communication constraints according to claim 5, characterized in that: If there is no object that is not in the speed obstacle space Φ i When the speed is within, assuming there are 4 candidate speeds {v cand 1 , v cand 2 , v cand 3 , v cand 4 }, three speed obstacle spaces Φ i ={φ1,φ2,φ3}, φ1~φ3 are VO or RVO, calculated by the corresponding obstacles / agents respectively; Selected speed v cand 1 , v cand 2 , v cand 3 , v cand 4 They are located in the speed obstacle spaces (φ1, φ2), (φ2, φ3), (φ1, φ2), and (φ2) respectively; For a candidate speed of an intelligent agent located in a speed obstacle space, the importance description function of the speed obstacle space to the current intelligent agent is defined as follows: In the formula, p m is the obstacle / agent position corresponding to the current speed obstacle space φ, ‖p m -p i ‖ is the Euclidean distance between the current agent and the corresponding obstacle / agent, R m is the radius of the obstacle / agent, r i is the radius of the current agent; n is the unit direction vector of the current agent pointing to the corresponding obstacle / agent position; A candidate speed The motion characteristics preservation function is described as follows: For an agent a i , its candidate speed The joint calculation function is as follows: In the formula, K mot and K imp They are motion characteristic weight and importance weight respectively; For an intelligent agent a i , joint calculation function The selected speed corresponding to the minimum value is the desired final velocity, 7. A multi-agent distributed flyable path planning system under communication constraints based on the multi-agent distributed flyable path planning method under communication constraints as described in any one of claims 1 to 6, characterized in that: It includes multiple UAV agents, each of which is equipped with sensors, communication modules, planners and actuators; The sensor and communication module are connected to the planner, and the planner is connected to the actuator; The planner includes an obstacle parameter acquisition module, an agent parameter acquisition module, an optimal local target point calculation module, a path planning module, an obstacle avoidance module and a detection module; The obstacle parameter acquisition module is used for each agent to obtain the surrounding local environment information through its own sensor and obtain the surrounding local obstacle parameters; The agent parameter acquisition module is used for each agent to communicate with other agents within the communication range through its own communication device to obtain parameter information of surrounding agents; The optimal local target point calculation module is used for each agent to select the optimal local target point from the surrounding local environment by using the improved artificial potential field model and combining its own target information; The path planning module is used to perform scene detection decomposition based on the optimal local target point, and each agent combines its own dynamic constraints, the current point and the position of the local target point to complete the local Dubins path planning; The obstacle avoidance module is used for each agent to move in time based on its own Dubins path planning, local obstacle information, and parameter information of other agents within the communication range, using the improved RVO obstacle avoidance algorithm, while avoiding obstacles between unmanned agents and between unmanned agents and obstacles; Detection module, used by each agent to detect the arrival target; If the target is not reached, the current position of the vehicle is updated and the next step of rolling time domain path planning is performed; If the target has been reached, the process ends.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images