A method and system for unmanned aerial vehicle path planning

Abstract

The application discloses a kind of unmanned plane path planning method and system, it is related to robot control technical field, method includes: constructing unmanned plane motion model;Unmanned plane path planning model is constructed based on the unmanned plane motion model;Obtain expert demonstration trajectory dataset, the expert demonstration trajectory dataset is segmented, and get obstacle avoidance demonstration trajectory dataset and navigation demonstration trajectory dataset;Obstacle avoidance strategy is obtained based on the obstacle avoidance demonstration trajectory dataset and first adversarial network;Navigation strategy is obtained based on the navigation demonstration trajectory dataset and second adversarial network;Unmanned plane path is real-time planned based on the unmanned plane path planning model, the obstacle avoidance strategy and the navigation strategy, until unmanned plane reaches target point.The application enables unmanned plane to learn from expert demonstration, obtains the control strategy of path planning, realizes unmanned plane in unknown environment Real-time planning from start point to end point of no-collision safe path.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot control, and in particular to a method and system for path planning of a UAV. BACKGROUND

[0002] Path planning is a rapidly developing discipline in recent years. Path planning refers to determining the positional relationship between itself and obstacles in the environment after perceiving the surrounding environment through various sensors, and automatically providing a path to reach the target position while avoiding obstacles. Due to the complexity of application scenarios and operating environments, mobile robots are required to have strong environmental adaptability.

[0003] The classical algorithm principle is simple, and there are many enhanced versions. These algorithms can solve most path planning tasks, so they have very wide applications.

[0004] Edsger W. Dijkstra proposed the Dijkstra algorithm for solving the shortest path. This algorithm performs forward traversal on all nodes to obtain the optimal path. However, when the number of nodes is very large, the algorithm will consume a large amount of time and computing memory, resulting in low efficiency and being unable to solve dynamic environments. A* and D* are improvements of the Dijkstra algorithm, which partially solve the efficiency problem and the dynamic environment problem. However, the efficiency problem is still the most critical point.

[0005] The RRT algorithm is a sample-based algorithm proposed by Lavalle. This algorithm starts from the initial node and continuously expands node 2 through random sampling. In this way, all nodes can be expanded into a random tree. When a node is expanded to the target node, a suitable path can be found among all nodes from the initial position to the target point. However, during the node expansion process, the algorithm will generate many useless nodes, and these useless explorations will slow down the convergence speed of the algorithm.

[0006] Reinforcement learning is an important branch of machine learning and is widely used in various decision-making tasks. The basic idea of reinforcement learning is through the continuous interaction process between the agent and the environment. According to the reward given to the agent's movement, the agent studies to obtain the maximum reward, thereby learning the optimal strategy. Reinforcement learning mainly includes value function-based methods such as Q-learning and policy search methods such as Actor-Critic. On this basis, many algorithms combine deep learning to solve complex tasks, such as DQN and DDPG. Improved algorithms are applied to path planning in dynamic environments. Compared with traditional path planning algorithms, path planning based on reinforcement learning has achieved better results. However, all reinforcement learning algorithms require certain environments and appropriate reward functions, and at the same time, the algorithm requires a large amount of training time.

[0007] The above path planning algorithms all rely on the cognition of the environment. When the robot needs to adapt to an unknown environment in actual application, the use of the above method has great limitations. SUMMARY

[0008] The purpose of the present application is to provide a UAV path planning method and system, so that the UAV can learn from expert demonstrations, get the control strategy of path planning, and realize the real-time planning of collision-free safe path from the starting point to the end point in unknown environment.

[0009] To achieve the above purpose, the present application provides the following scheme:

[0010] A UAV path planning method, comprising:

[0011] constructing a UAV motion model; constructing a UAV path planning model based on the UAV motion model;

[0012] obtaining an expert demonstration trajectory data set, segmenting the expert demonstration trajectory data set to obtain an obstacle avoidance demonstration trajectory data set and a navigation demonstration trajectory data set;

[0013] obtaining an obstacle avoidance strategy based on the obstacle avoidance demonstration trajectory data set and a first adversarial network;

[0014] obtaining a navigation strategy based on the navigation demonstration trajectory data set and a second adversarial network;

[0015] real-time planning of the UAV path based on the UAV path planning model, the obstacle avoidance strategy and the navigation strategy until the UAV reaches the target point.

[0016] Preferably, the UAV motion model is as follows:

[0017]

[0018] In the formula: x represents the x coordinate of the UAV in space, y represents the y coordinate of the UAV in space, ψ represents the heading angle of the UAV, φ represents the tilt angle of the UAV, v represents the speed of the UAV, g represents the acceleration of gravity, δt represents the step length, a represents the control input of the UAV, represents the turning rate.

[0019] Preferably, the obstacle avoidance strategy is obtained based on the obstacle avoidance demonstration trajectory data set and the first adversarial network, and specifically:

[0020] obtaining a position information set and a state information set of the UAV based on the obstacle avoidance demonstration trajectory data set;

[0021] obtaining an initial Markov decision set based on the position information set and the state information set;

[0022] transforming the initial Markov decision set from absolute states into relative states to obtain a Markov decision set;

[0023] training the first adversarial network based on the Markov decision set to obtain the obstacle avoidance strategy.

[0024] Preferably, the UAV path planning model is as follows:

[0025] Q=[S,A,f,O];

[0026] In the formula, S represents a state space of an environment, A is an action space, O is an observation space, f is a state transition function, and represents a probability of state transition after action execution.

[0027] s∈S, s=[x, y, v, ψ, φ], a∈A, a∈{[-40° / s], [40° / s], [-20° / s], [-20° / s], [0° / s]}.

[0028] Preferably, the UAV path is planned in real time based on the UAV path planning model, the obstacle avoidance strategy and the navigation strategy until the UAV reaches the target point, and specifically:

[0029] determining whether there is an obstacle on the path of the UAV to the target point at the current time based on the UAV path planning model;

[0030] when there is, performing obstacle avoidance path planning based on the obstacle avoidance strategy, performing an obstacle avoidance action based on the obstacle avoidance path, and updating the UAV path planning model based on the obstacle avoidance action;

[0031] when there is not, performing navigation path planning based on the navigation strategy, performing a navigation action based on the navigation path, and updating the UAV path planning model based on the navigation action;

[0032] repeating the above process until the UAV reaches the target point.

[0033] The application also provides a UAV path planning system, comprising:

[0034] a model construction module configured to construct a UAV motion model and construct a UAV path planning model based on the UAV motion model;

[0035] an expert demonstration module configured to obtain an expert demonstration trajectory data set and segment the expert demonstration trajectory data set to obtain an obstacle avoidance demonstration trajectory data set and a navigation demonstration trajectory data set;

[0036] an obstacle avoidance strategy module, configured to obtain an obstacle avoidance strategy based on the obstacle avoidance demonstration trajectory dataset and the first generative adversarial network;

[0037] a navigation strategy module, configured to obtain a navigation strategy based on the navigation demonstration trajectory dataset and the second generative adversarial network;

[0038] a path planning module, configured to plan a path for the UAV in real time based on the UAV path planning model, the obstacle avoidance strategy and the navigation strategy until the UAV reaches the target point.

[0039] Preferably, the UAV motion model is as follows:

[0040]

[0041] wherein x represents the x coordinate of the UAV in space, y represents the y coordinate of the UAV in space, ψ represents the heading angle of the UAV, φ represents the tilt angle of the UAV, v represents the speed of the UAV, g represents the acceleration of gravity, δt represents the step length, a represents the control input of the UAV, represents the turning rate.

[0042] Preferably, the obstacle avoidance strategy module specifically comprises:

[0043] obtaining a position information set and a state information set of the UAV based on the obstacle avoidance demonstration trajectory dataset;

[0044] obtaining an initial Markov decision set based on the position information set and the state information set;

[0045] transforming the initial Markov decision set from absolute state to relative state to obtain a Markov decision set;

[0046] training the first generative adversarial network based on the Markov decision set to obtain the obstacle avoidance strategy.

[0047] Preferably, the UAV path planning model is as follows:

[0048] Q = [S, A, f, O];

[0049] wherein S represents the state space of the environment, A represents the action space, O represents the observation space, f represents the state transition function, and represents the probability of state transition after action execution;

[0050] s ∈ S, s = [x, y, v, ψ, φ], a ∈ A, a ∈ {[-40° / s], [40° / s], [-20° / s], [-20° / s], [0° / s]}.

[0051] Preferably, the path planning module specifically comprises:

[0052] determine whether there is an obstacle on the path from the UAV to the target point at the current time based on the UAV path planning model;

[0053] When there is, perform obstacle avoidance path planning based on the obstacle avoidance strategy, perform an obstacle avoidance action based on the obstacle avoidance path, and update the UAV path planning model based on the obstacle avoidance action.

[0054] When there is not, perform navigation path planning based on the navigation strategy, perform a navigation action based on the navigation path, and update the UAV path planning model based on the navigation action.

[0055] Repeat the above process until the UAV reaches the target point.

[0056] According to the specific embodiments of the present application, the following technical effects are disclosed:

[0057] The present application discloses a UAV path planning method and system, which comprises the following steps: constructing a UAV motion model; constructing a UAV path planning model based on the UAV motion model; obtaining an expert demonstration trajectory dataset, segmenting the expert demonstration trajectory dataset to obtain an obstacle avoidance demonstration trajectory dataset and a navigation demonstration trajectory dataset; obtaining an obstacle avoidance strategy based on the obstacle avoidance demonstration trajectory dataset and a first adversarial network; obtaining a navigation strategy based on the navigation demonstration trajectory dataset and a second adversarial network; and planning a UAV path in real time based on the UAV path planning model, the obstacle avoidance strategy, and the navigation strategy until the UAV reaches the target point. The present application enables the UAV to learn from expert demonstrations and obtain a control strategy for path planning, thereby realizing real-time collision-free safe path planning from the starting point to the end point in an unknown environment. BRIEF DESCRIPTION OF DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0059] Figure 1 The flowchart of the UAV path planning method of the present application;

[0060] Figure 2 The first embodiment schematic diagram of the motion trajectory obtained based on the method of the present application;

[0061] Figure 3 The second embodiment schematic diagram of the motion trajectory obtained based on the method of the present application;

[0062] Figure 4 The figure is a structure diagram of the unmanned aerial vehicle path planning system of the present application.

[0063] Symbol explanation: 1, model construction module; 2, expert demonstration module; 3, obstacle avoidance strategy module; 4, navigation strategy module; 5, path planning module. DETAILED DESCRIPTION

[0064] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0065] The purpose of the present application is to provide an unmanned aerial vehicle path planning method and system, so that the unmanned aerial vehicle can learn from expert demonstration, obtain a control strategy of path planning, and realize real-time planning of a collision-free safe path from a starting point to an end point in an unknown environment.

[0066] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0067] Figure 1 The figure is a flow chart of the unmanned aerial vehicle path planning method of the present application. As shown in the figure, the present application provides an unmanned aerial vehicle path planning method, which comprises the following steps: Figure 1

[0068] Step S1, constructing an unmanned aerial vehicle motion model; constructing an unmanned aerial vehicle path planning model based on the unmanned aerial vehicle motion model. The unmanned aerial vehicle motion model is as follows:

[0069]

[0070] In the formula, x represents the x coordinate of the unmanned aerial vehicle in space, y represents the y coordinate of the unmanned aerial vehicle in space, ψ represents the heading angle of the unmanned aerial vehicle, φ represents the tilt angle of the unmanned aerial vehicle, v represents the speed of the unmanned aerial vehicle, g represents the gravitational acceleration, δt represents the step length, a represents the control input of the unmanned aerial vehicle, represents the turning rate, and ψ is between [-π, π].

[0071] In order to solve the optimal strategy, the path planning problem is constructed as a Markov decision process without a reward function, and the unmanned aerial vehicle path planning model is as follows:

[0072] Q=[S,A,f,O];

[0073] ​In the formula, S represents a state space of an environment, A is an action space, O is an observation space, f is a state transition function, and represents a probability of state transition after action execution;

[0074] s∈S, s=[x,y,v,ψ,φ], a∈A, a∈{[-40° / s],[40° / s],[-20° / s],[-20° / s],[0° / s]}.

[0075] Step S2, obtaining an expert demonstration trajectory data set, and segmenting the expert demonstration trajectory data set to obtain an obstacle avoidance demonstration trajectory data set and a navigation demonstration trajectory data set.

[0076] Step S3, obtaining an obstacle avoidance strategy based on the obstacle avoidance demonstration trajectory data set and a first adversarial network.

[0077] Specifically, the step S3 is specifically:

[0078] Obtaining a position information set and a state information set of the unmanned aerial vehicle based on the obstacle avoidance demonstration trajectory data set.

[0079] Obtaining an initial Markov decision set based on the position information set and the state information set.

[0080] Converting the initial Markov decision set from an absolute state to a relative state to obtain a Markov decision set. Preferably, the absolute state s=[x,y,v,ψ,φ] is converted into the relative state is an angle between a line of sight and a heading of the unmanned aerial vehicle, and l is an obstacle distance.

[0081] The Markov decision set is D={τ1,...,τ m} and τ=(s0,a0,s1,a1,...,s T ,a T ).

[0082] Training the first adversarial network based on the Markov decision set to obtain the obstacle avoidance strategy.

[0083] Preferably, the first adversarial network is trained based on adversarial inverse reinforcement learning, an approximate reward function is obtained by inferring an intention of Markov decision demonstration, and then a maximum likelihood is solved, as follows:

[0084] In the formula, p θ (τ) is a probability of a sample τ in a probability function p with a parameter θ, t is time, T is the time length of the sample. This formula makes the reward function r(s, a) parameterized and fixes the dynamic and initial state distribution on the dynamic and initial state distribution of Markov decision.

[0085] Convert the maximum likelihood problem into a generative adversarial network optimization problem:

[0086]

[0087] where f θ,φ is the combination of the approximate reward function r and the shaping term h, f(s, a, s') = r(s, a) + γh(s') - h(s), s is the current state, a is the action, and s' is the next state.

[0088] Train by maximizing R θ,φ (s, a, s') = log(D θ,φ (s, a, s')) - log(1 - D θ,φ (s, a, s')) to get the obstacle avoidance strategy.

[0089] Get human demonstration samples

[0090] Randomly initialize the strategy π and the discriminator D θ,φ ;

[0091] for step t = 1, 2,..., N do;

[0092] Collect trajectories τ i using the strategy π;

[0093] Train D θ,φ by binary logistic regression to distinguish and generate samples τ i ;

[0094] Update the reward function R θ,φ (s, a, s');

[0095] Update the strategy using R θ,φ (s, a, s') as the reward function of the PPO algorithm;

[0096] end for.

[0097] Set the number of training times to 100000, update the discriminator D θ,φ and PPO every 2000 times. The learning rate is set to 0.0003. Choose Adam as the optimizer. The parameters of the optimizer are set to 0.9 and 0.999 respectively. γ in f θ,φ (s, a, s') is 0.995. Use the public platform PyTorch for implementation.

[0098] Step S4, based on the navigation demonstration trajectory data set and the second adversarial network, obtain the navigation strategy. Referring to step S3, the same reasoning applies.

[0099] Step S5, based on the UAV path planning model, the obstacle avoidance strategy and the navigation strategy, real-time planning of the UAV path is performed until the UAV reaches the target point.

[0100] Optionally, the step S5 is specifically:

[0101] Based on the UAV path planning model, it is determined whether there is an obstacle on the path of the UAV to the target point at the current time;

[0102] When there is, based on the obstacle avoidance strategy, the obstacle avoidance path is planned, the obstacle avoidance action is performed based on the obstacle avoidance path, and the UAV path planning model is updated based on the obstacle avoidance action;

[0103] When there is not, based on the navigation strategy, the navigation path is planned, the navigation action is performed based on the navigation path, and the UAV path planning model is updated based on the navigation action;

[0104] The above process is repeated until the UAV reaches the target point.

[0105] The UAV path planning tracking result based on the method of the present application is as shown in Figure 2 and Figure 3 Curve in Figure 2 and Figure 3 represents the curve of the tracking planning path.

[0106] Figure 4 The structure diagram of the UAV path planning system of the present application is shown in Figure 4 The present application provides a UAV path planning system, which comprises a model construction module 1, an expert demonstration module 2, an obstacle avoidance strategy module 3, a navigation strategy module 4 and a path planning module 5.

[0107] The model construction module 1 is used to construct a UAV motion model and construct a UAV path planning model based on the UAV motion model.

[0108] The expert demonstration module 2 is used to obtain an expert demonstration trajectory data set and segment the expert demonstration trajectory data set to obtain an obstacle avoidance demonstration trajectory data set and a navigation demonstration trajectory data set.

[0109] The obstacle avoidance strategy module 3 is used to obtain an obstacle avoidance strategy based on the obstacle avoidance demonstration trajectory data set and a first adversarial network.

[0110] The navigation strategy module 4 is configured to obtain a navigation strategy based on the navigation demonstration trajectory data set and the second generative adversarial network.

[0111] The path planning module 5 is configured to plan a path for the UAV in real time based on the UAV path planning model, the obstacle avoidance strategy and the navigation strategy until the UAV reaches the target point.

[0112] Optionally, the UAV motion model is as follows:

[0113]

[0114] In the formula, x represents the x coordinate of the UAV in space, y represents the y coordinate of the UAV in space, represents the heading angle of the UAV, represents the tilt angle of the UAV, v represents the speed of the UAV, g represents the acceleration of gravity, represents the step length, a represents the control input of the UAV, represents the turning rate.

[0115] Optionally, the obstacle avoidance strategy module 3 specifically comprises:

[0116] The obstacle avoidance demonstration trajectory data set is used to obtain a position information set and a state information set of the UAV.

[0117] The position information set and the state information set are used to obtain an initial Markov decision set.

[0118] The initial Markov decision set is converted from absolute state to relative state to obtain a Markov decision set.

[0119] The first generative adversarial network is trained based on the Markov decision set to obtain the obstacle avoidance strategy.

[0120] Optionally, the UAV path planning model is as follows:

[0121] Q = [S, A, f, O];

[0122] In the formula, S represents the state space of the environment, A represents the action space, O represents the observation space, f represents the state transition function, and represents the probability of state transition after action execution.

[0123] s e S, s = [x, y, v,, ], a e A, a e {[-40° / s], [40° / s], [-20° / s], [-20° / s], [0° / s]}.

[0124] Optionally, the path planning module 5 specifically comprises:

[0125] The UAV path planning model is used to determine whether there is an obstacle on the path of the UAV to the target point at the current time.

[0126] When the obstacle avoidance strategy exists, an obstacle avoidance path is planned based on the obstacle avoidance strategy, an obstacle avoidance action is performed based on the obstacle avoidance path, and the UAV path planning model is updated based on the obstacle avoidance action.

[0127] When the obstacle avoidance strategy does not exist, a navigation path is planned based on the navigation strategy, a navigation action is performed based on the navigation path, and the UAV path planning model is updated based on the navigation action.

[0128] The above process is repeatedly performed until the UAV reaches the target point.

[0129] The various embodiments are described in a progressive manner in the specification, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be mutually referred to. For the system disclosed in the embodiments, the description is relatively simple because it corresponds to the method disclosed in the embodiments. The relevant parts can be referred to in the method part.

[0130] The principles and implementation manners of the present application are described by using specific examples in the present application. The above description of the embodiments is only used to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, the specific implementation manners and application ranges can be changed according to the idea of the present application. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method for unmanned aerial vehicle path planning, characterized in that, The application comprises the following steps: constructing a UAV motion model; constructing a UAV path planning model based on the UAV motion model; obtaining an expert demonstration trajectory data set, and segmenting the expert demonstration trajectory data set to obtain an obstacle avoidance demonstration trajectory data set and a navigation demonstration trajectory data set; obtaining an obstacle avoidance strategy based on the obstacle avoidance demonstration trajectory data set and a first adversarial network; specifically comprising: obtaining a position information set and a state information set of the UAV based on the obstacle avoidance demonstration trajectory data set; obtaining an initial Markov decision set based on the position information set and the state information set; converting the initial Markov decision set from an absolute state to a relative state to obtain a Markov decision set; training the first adversarial network based on the Markov decision set to obtain the obstacle avoidance strategy; s = [x, y, v, ψ, φ] is converted into the relative state s = [x, y, v, ψ, φ] of is the angle between the line of sight and the heading of the UAV, l is the distance of the obstacle; x is the x coordinate of the UAV in space, y is the y coordinate of the UAV in space, ψ is the heading angle of the UAV, φ is the tilt angle of the UAV, and v is the speed of the UAV; obtaining a navigation strategy based on the navigation demonstration trajectory data set and a second adversarial network; real-time planning a UAV path based on the UAV path planning model, the obstacle avoidance strategy and the navigation strategy until the UAV reaches a target point. 2.The UAV path planning method of claim 1, wherein, The UAV motion model is as follows: In the formula: x represents the x coordinate of the unmanned aerial vehicle in space, y represents the y coordinate of the unmanned aerial vehicle in space, represents the heading angle of the unmanned aerial vehicle, represents the tilt angle of the unmanned aerial vehicle, v represents the speed of the unmanned aerial vehicle, g represents the acceleration of gravity, represents the step length, a represents the control input of the unmanned aerial vehicle, represents the turning rate. 3.The UAV path planning method of claim 2, wherein, The UAV path planning model is as follows: Q = [S, A, f, O]; In the formula, S represents the state space of the environment, A is the action space, O is the observation space, f is the state transition function, and represents the probability of state transition after action execution; s∈S, s=[x,y,v,ψ,φ], a∈A, a∈{[-40° / s],[40° / s],[-20° / s],[20° / s],[0° / s]}. 4.The method of claim 1, wherein, The real-time planning of the UAV path based on the UAV path planning model, the obstacle avoidance strategy and the navigation strategy until the UAV reaches the target point is specifically: judging whether there is an obstacle on the path from the current time point of the UAV to the target point based on the UAV path planning model; when there is, planning an obstacle avoidance path based on the obstacle avoidance strategy, performing an obstacle avoidance action based on the obstacle avoidance path, and updating the UAV path planning model based on the obstacle avoidance action; when there is not, planning a navigation path based on the navigation strategy, performing a navigation action based on the navigation path, and updating the UAV path planning model based on the navigation action; repeating the above process until the UAV reaches the target point.

5. An unmanned aerial vehicle path planning system, comprising: The application comprises the following steps: a model construction module for constructing a UAV motion model and constructing a UAV path planning model based on the UAV motion model; an expert demonstration module for obtaining an expert demonstration trajectory data set and segmenting the expert demonstration trajectory data set to obtain an obstacle avoidance demonstration trajectory data set and a navigation demonstration trajectory data set; an obstacle avoidance strategy module for obtaining an obstacle avoidance strategy based on the obstacle avoidance demonstration trajectory data set and a first adversarial network; a navigation strategy module for obtaining a navigation strategy based on the navigation demonstration trajectory data set and a second adversarial network; a path planning module for real-time planning a UAV path based on the UAV path planning model, the obstacle avoidance strategy and the navigation strategy until the UAV reaches a target point.

6. The UAV path planning system of claim 5, wherein, The UAV motion model is as follows: In the formula: x represents the x coordinate of the unmanned aerial vehicle in space, y represents the y coordinate of the unmanned aerial vehicle in space, represents the heading angle of the unmanned aerial vehicle, represents the tilt angle of the unmanned aerial vehicle, v represents the speed of the unmanned aerial vehicle, g represents the acceleration of gravity, represents the step length, a represents the control input of the unmanned aerial vehicle, represents the turning rate.

7. The UAV path planning system of claim 5, wherein, The obstacle avoidance strategy module is specifically: Obtain a position information set and a state information set of the UAV based on the obstacle avoidance demonstration trajectory data set; Obtain an initial Markov decision set based on the position information set and the state information set; Convert the initial Markov decision set from an absolute state to a relative state to obtain a Markov decision set; Train the first adversarial network based on the Markov decision set to obtain the obstacle avoidance strategy.

8. The UAV path planning system of claim 6, wherein, The UAV path planning model is as follows: Q=[S,A,f,O]; In the formula, S represents a state space of an environment, A is an action space, O is an observation space, f is a state transition function, and represents a probability of state transition after action execution; s∈S, s=[x,y,v,ψ,φ], a∈A, a∈{[-40° / s],[40° / s],[-20° / s],[20° / s],[0° / s]}.

9. The UAV path planning system of claim 5, wherein, The path planning module is specifically: Determine whether there is an obstacle on a path from the UAV to a target point at a current time based on the UAV path planning model; When there is an obstacle, perform obstacle avoidance path planning based on the obstacle avoidance strategy, execute an obstacle avoidance action based on the obstacle avoidance path, and update the UAV path planning model based on the obstacle avoidance action; When there is no obstacle, perform navigation path planning based on the navigation strategy, execute a navigation action based on the navigation path, and update the UAV path planning model based on the navigation action; Repeat the above process until the UAV reaches the target point.

Images