Potential consistency planning method and system for unmanned aerial vehicle track generation

Abstract

The invention discloses a potential consistency planning method and system for unmanned aerial vehicle track generation, and relates to the technical field of unmanned aerial vehicles, and the method comprises the steps: obtaining unmanned aerial vehicle track sequence data, carrying out the feature extraction based on the unmanned aerial vehicle track sequence data, obtaining a coding random feature and a coding determination feature, determining a decoding reduction trajectory state based on the coding random feature and the coding determination feature; and inputting the coding random feature and the coding determination feature into a pre-established consistency world model based on the decoding restoration trajectory state, and outputting to obtain the state feature of the next time step, so that the intelligent level of autonomous flight of the unmanned aerial vehicle can be improved.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to a potential consistency planning method and system for generating UAV trajectories. Background Art

[0002] With the rapid development of drone technology, its applications in urban logistics, environmental monitoring, disaster relief, and other fields are becoming increasingly widespread. However, autonomous navigation of drones in complex environments still faces many challenges, such as avoiding dynamic obstacles, responding to emergencies, and planning long-term autonomous flight. Current trajectory planning methods often rely on predefined environmental models or rule-guided path optimization, but these methods have difficulty dealing with random factors in drone trajectories, such as sudden obstacles and changes in ambient wind speed. In addition, trajectory prediction methods based on traditional deep learning usually rely on large-scale data training, but their generalization ability in unseen environments is limited, making it difficult to provide sufficiently robust trajectory generation capabilities. Summary of the Invention

[0003] In order to address the deficiencies mentioned in the above background technology, the purpose of the present invention is to provide a potential consistency planning method and system for UAV trajectory generation.

[0004] In a first aspect, the purpose of the present invention can be achieved by the following technical solution: a potential consistency planning method for UAV trajectory generation, the method comprising the following steps:

[0005] Acquire the drone trajectory sequence data, perform feature extraction based on the drone trajectory sequence data to obtain coded random features and coded deterministic features, and determine the decoding and restoration trajectory state based on the coded random features and the coded deterministic features;

[0006] Based on the decoding and restoration of the trajectory state, the encoded random features and the encoded deterministic features are input into the pre-established consistent world model, and the state features of the next time step are output.

[0007] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: extracting features based on the drone trajectory sequence data, the process being as follows:

[0008] The random feature encoder E1 is used to extract the random feature D-Feature from the UAV trajectory as the encoded random feature, and the deterministic feature encoder E2 is used to extract the deterministic feature I-Feature as the encoded deterministic feature.

[0009] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: extracting random features D-Features from the UAV trajectory using the random feature encoder E1, the process comprising:

[0010] In UAV trajectory sequence data In the state s t 、Action a t , reward r t and the next state s t+1 , random feature encoder E1 encodes two adjacent states s t 、s t+1 and action a t To obtain random features between state transitions:

[0011] D-Feature = E1( s t , a t , s t+1 ) (1)

[0012] Among them, E1 captures the nonlinear changes between trajectory states and extracts disturbance information caused by environmental uncertainty or external factors as encoded random features.

[0013] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: extracting the definite feature I-Feature using the definite feature encoder E2, the process including:

[0014] The drone trajectory sequence data also contains long-term stable deterministic patterns, and the feature encoder E2 is responsible for extracting the deterministic patterns:

[0015] I-Feature = E2(s t ) (2)

[0016] Among them, the determined feature I-Feature in the UAV trajectory is encoded according to formula (2) as the coded determined feature.

[0017] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: determining the decoding and restoration trajectory state based on the encoding random feature and the encoding determination feature, the process including:

[0018] The decoder D uses both the encoded random feature D-Feature and the encoded deterministic feature I-Feature to restore the trajectory:

[0019] s′ t =D(D-Feature,I-Feature) (3)

[0020] Among them, s′ t The original state is restored by the decoder, and then written as the following optimization problem:

[0021]

[0022] The L1 error is calculated by reconstructing the state s′ t With the real state s t The mean absolute difference between the two, L2 error is calculated by reconstructing the state s′ t With the real state s t The mean square error between the worst pixel L 2-worst The error is optimized by specifically reconstructing the pixel with the largest error. The three parts of the error are calculated based on the weight coefficients c1, c2, c 2-worst Perform weighted summation to ultimately form the total loss function of the optimization task.

[0023] In combination with the first aspect, in certain implementations of the first aspect, the method further includes: the pre-established consistent world model includes a student world model and a teacher world model.

[0024] In conjunction with the first aspect, in certain implementations of the first aspect, the method further includes: the student world model combines the first t D-Features and the I-Feature to predict the D-Feature at the next time t+1 to predict the overall trajectory direction:

[0025] D′ t+1 =WM(D-Feature <＝t ,I-Feature <＝t ) (5)

[0026] Where WM represents the student world model, D′ t+1 Represents the D-Feature of the next moment. The teacher world model without gradient update capability is used to predict the D-Feature of the next moment t+1 based on the previous t D-Features and I-Feature:

[0027] D_con′ t+1 =WMT(D-Feature <＝t ,I-Feature <＝t ) (6)

[0028] Where WMT represents the teacher world model. According to the prediction formulas (5) and (6) of the consistent world model, the overall optimization task is written as the following optimization problem:

[0029]

[0030] c3 and c4 are the weighting coefficients of prediction loss and consistency loss, respectively.

[0031] In a second aspect, in order to achieve the above-mentioned objectives, the present invention discloses a potential consistency planning system for UAV trajectory generation, comprising:

[0032] A feature extraction module is used to obtain the UAV trajectory sequence data, perform feature extraction based on the UAV trajectory sequence data, obtain the coded random features and the coded deterministic features, and determine the decoding and restoration trajectory state based on the coded random features and the coded deterministic features;

[0033] The state planning module is used to restore the trajectory state based on decoding, input the encoded random features and encoded deterministic features into a pre-established consistent world model, and output the state features of the next time step.

[0034] In another aspect of the present invention, in order to achieve the above-mentioned purpose, a terminal device is disclosed, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores a computer program capable of running on the processor, and when the processor loads and executes the computer program, it adopts the potential consistency planning method for drone trajectory generation as described above.

[0035] In another aspect of the present invention, in order to achieve the above-mentioned purpose, a computer-readable storage medium is disclosed, in which a computer program is stored. When the computer program is loaded and executed by a processor, the potential consistency planning method for drone trajectory generation as described above is adopted.

[0036] Beneficial effects of the present invention:

[0037] This method uses an encoder to extract both random and deterministic features from trajectories and construct a consistent world model. This allows drones to efficiently simulate trajectory changes in a low-dimensional latent space, improving their adaptability to complex environments. Furthermore, this method combines the consistent model with a feature alternation mechanism to reduce the model's overreliance on historical information, optimize the preservation of trajectory details, and achieve efficient and precise trajectory prediction, enhancing the intelligence level of autonomous drone flight. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0039] Figure 1 It is a schematic flow chart of the method of the present invention;

[0040] Figure 2 It is a schematic diagram of the overall process of the present invention;

[0041] Figure 3 This is a schematic diagram of the overall process of the world model constructed by the present invention;

[0042] Figure 4 It is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0044] Example 1:

[0045] like Figure 1 As shown, a potential consistency planning method for UAV trajectory generation includes the following steps:

[0046] S101: Acquire drone trajectory sequence data, perform feature extraction based on the drone trajectory sequence data to obtain coding random features and coding deterministic features, and determine a decoding and restoration trajectory state based on the coding random features and coding deterministic features;

[0047] The feature extraction process based on the UAV trajectory sequence data is as follows:

[0048] The random feature encoder E1 is used to extract the random feature D-Feature from the UAV trajectory as the encoded random feature, and the deterministic feature encoder E2 is used to extract the deterministic feature I-Feature as the encoded deterministic feature.

[0049] Specifically, a random feature encoder E1 extracts random features D-Feature from the drone's trajectory, while a deterministic feature encoder E2 extracts deterministic features I-Feature. Decoder D then combines D-Feature and I-Feature to reconstruct the original trajectory. This encoder-decoder combination is then continuously optimized to obtain both random features for modeling the world and deterministic features that mitigate the world model's reliance on historical random features.

[0050] The random feature encoder E1 is used to extract random features D-Feature from the UAV trajectory. The process includes:

[0051] In UAV trajectory sequence data In the state s t 、Action a t , reward r t and the next state s t+1 , random feature encoder E1 encodes two adjacent states s t 、st+1 and action a t To obtain random features between state transitions:

[0052] D-Feature = E1( s t , a t , s t+1 ) (1)

[0053] Among them, E1 captures the nonlinear changes between trajectory states and extracts disturbance information caused by environmental uncertainty or external factors (such as sudden obstacles and wind speed changes) as encoded random features.

[0054] The process of extracting the determined feature I-Feature by using the determined feature encoder E2 includes:

[0055] The drone trajectory sequence data also contains long-term stable deterministic patterns, and the feature encoder E2 is responsible for extracting the deterministic patterns:

[0056] I-Feature = E2(s t ) (2)

[0057] Among them, the determined feature I-Feature in the UAV trajectory is encoded according to formula (2) as the coded determined feature.

[0058] Decoding trajectory states based on random features and deterministic features

[0059] In order to reconstruct the trajectory state, the decoder D needs to use both the random feature D-Feature and the deterministic feature I-Feature to restore the trajectory:

[0060] s' t =D(D-Feature,I-Feature) (3)

[0061] Among them, s' t is the original state restored by the decoder, then the overall optimization task of this step can be written as the following optimization problem:

[0062]

[0063] Among them, the decoder loss function is designed to comprehensively consider different types of errors, thereby optimizing the encoder's encoding and decoding capabilities. First, the L1 error is calculated by reconstructing the state s' t With the real state s t The mean absolute difference between the two helps the decoder generate smoother prediction results. Secondly, the L2 error is calculated by reconstructing the state s' t With the real state s tThe mean square error between them ensures that the decoder can accurately fit the true trajectory. Finally, the worst pixel L 2-worst The error is optimized specifically to improve the robustness of the decoder to extreme errors. All three parts of the error are calculated according to the weight coefficients c1, c2, c 2-worst Perform a weighted summation to ultimately form the total loss function for the optimization task. Repeat the above steps for all drone trajectory data, gradually reducing the loss and optimizing the encoder and decoder.

[0064] S102: Based on the decoded and restored trajectory state, the encoded random features and the encoded deterministic features are input into a pre-established consistent world model, and the state features of the next time step are output.

[0065] During the world model construction process, D-Feature and I-Feature are used alternately, and a consistent world model is used to optimize trajectory prediction, reducing the detail loss caused by the discrete variable method. This allows the trajectory details to be accurately preserved while enhancing the ability to predict random events.

[0066] The consistent world model mainly consists of the student world model and the teacher world model. The student world model mainly combines the first t D-Features and I-Features to predict the D-Feature at the next time t+1 to predict the overall trajectory trend:

[0067] D' t+1 =WM(D-Feature <＝t ,I-Feature <＝t ) (5)

[0068] Where WM represents the student world model, D' t+1 In order to improve the prediction ability, the teacher world model without gradient update capability is used to predict the D-Feature of the next moment t+1 based on the previous t D-Features and I-Feature:

[0069] D_con' t+1 =WMT(D-Feature <＝t ,I-Feature <＝t ) (6)

[0070] Where WMT represents the teacher world model. According to the prediction formulas (5) and (6) of the consistent world model, the overall optimization task of this step can be written as the following optimization problem:

[0071]

[0072] The final training loss is obtained by weighted summation of prediction loss and consistency loss, where c3 and c4 are weighted coefficients of prediction loss and consistency loss, respectively. This total loss function guides the model to balance consistency and accurate prediction capabilities during training. This application uses formula (7) to update the student world model, and then uses the parameters of the student world model to update the teacher world model. In this way, and by alternating training with D-Feature and I-Feature, the trajectory prediction ability can be continuously optimized, the loss of details caused by discrete variables can be reduced, and the prediction ability of random events can be enhanced.

[0073] Specifically, the present invention will be further described below through examples:

[0074] In order to verify the effectiveness of the extraction method and feedback the extraction effect, the present invention adopts the following evaluation indicators: generated quality evaluation (FID), mean, median and interquartile mean (IQM).

[0075] FID measures the difference between the generated image and the real image. The smaller the value, the better the quality of the generated image. It evaluates the ability of the encoder and decoder.

[0076] The mean, median, and interquartile mean (IQM) are used to evaluate the total reward returned by the drone. These three indicators can measure the consistent world model from three perspectives: overall performance, typical performance, and robust performance after removing extreme values.

[0077] The application areas of this patent cover multiple scenarios related to autonomous flight and intelligent planning of drones, mainly including:

[0078] 1. Urban logistics and drone delivery

[0079] In complex urban environments, logistics drones need to accurately predict their flight trajectories to avoid sudden obstacles (such as buildings, birds, drone traffic, etc.) and improve the safety and efficiency of delivery.

[0080] The patented trajectory optimization method can achieve more intelligent path planning and improve the autonomous flight capability of logistics drones.

[0081] 2. Emergency rescue and search and rescue missions

[0082] In emergencies such as earthquakes, mountain torrents, and fires, drones need to autonomously search for survivors and make real-time route adjustments in complex environments.

[0083] Through the consistent world model and feature alternation mechanism, this patent can enhance the decision-making ability of drones in dynamic environments and increase the success rate of search and rescue missions.

[0084] 3. Precision agriculture and intelligent inspection

[0085] In agricultural plant protection, farmland monitoring and infrastructure inspection (such as power line and pipeline inspection) tasks, drones need to adapt to complex terrain and ensure precise flight paths.

[0086] This application can reduce trajectory deviation and improve the accuracy of route planning, thereby enhancing the intelligence level of drone operations.

[0087] Example 2: The second aspect, as Figure 4 As shown, in order to achieve the above-mentioned purpose, the present invention discloses a potential consistency planning system for UAV trajectory generation, comprising:

[0088] A feature extraction module 11 is used to obtain drone trajectory sequence data, perform feature extraction based on the drone trajectory sequence data, obtain coded random features and coded deterministic features, and determine the decoding and restoration trajectory state based on the coded random features and coded deterministic features;

[0089] The state planning module 12 is used to restore the trajectory state based on decoding, input the encoded random features and the encoded deterministic features into a pre-established consistent world model, and output the state features of the next time step.

[0090] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. 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 gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0091] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium having a computer program stored thereon, which executes the above method when executed by a processor. The storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or component, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component.

[0092] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0093] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present disclosure. Various changes and improvements may be made to the present disclosure without departing from the spirit and scope of the present disclosure, and such changes and improvements shall fall within the scope of the present disclosure.

Claims

1. A potential consistency planning method for UAV trajectory generation, characterized by: The method comprises the following steps: Acquire the drone trajectory sequence data, perform feature extraction based on the drone trajectory sequence data to obtain coded random features and coded deterministic features, and determine the decoding and restoration trajectory state based on the coded random features and the coded deterministic features; Based on the decoding and restoration of the trajectory state, the encoded random features and the encoded deterministic features are input into the pre-established consistent world model, and the state features of the next time step are output.

2. The potential consistency planning method for UAV trajectory generation according to claim 1 is characterized in that: The feature extraction process based on the UAV trajectory sequence data is as follows: The random feature encoder E1 is used to extract the random feature D-Feature from the UAV trajectory as the encoded random feature, and the deterministic feature encoder E2 is used to extract the deterministic feature I-Feature as the encoded deterministic feature.

3. The potential consistency planning method for UAV trajectory generation according to claim 2 is characterized in that: The random feature encoder E1 is used to extract random features D-Feature from the UAV trajectory. The process includes: In UAV trajectory sequence data In the state s t 、Action a t , reward r t and the next state s t+1 , random feature encoder E1 encodes two adjacent states s t 、s t+1 and action a t To obtain random features between state transitions: D-Feature=E1(S t ,a t ,s t+1 ) (1) Among them, E1 captures the nonlinear changes between trajectory states and extracts disturbance information caused by environmental uncertainty or external factors as encoded random features.

4. The potential consistency planning method for UAV trajectory generation according to claim 3 is characterized in that: The process of extracting the determined feature I-Feature by using the determined feature encoder E2 includes: The drone trajectory sequence data also contains long-term stable deterministic patterns, and the feature encoder E2 is responsible for extracting the deterministic patterns: I-Feature = E2(s t ) (2) Among them, the determined feature I-Feature in the UAV trajectory is encoded according to formula (2) as the coded determined feature.

5. The potential consistency planning method for UAV trajectory generation according to claim 1 is characterized in that: The process of determining the decoding and restoration trajectory state based on the coding random feature and the coding determination feature includes: The decoder D uses both the encoded random feature D-Feature and the encoded deterministic feature I-Feature to restore the trajectory: s ′ t =D(D-Feature,I-Feature) (3) Among them, s ′ t The original state is restored by the decoder, and then written as the following optimization problem: The L1 error is calculated by reconstructing the state s ′ t With the real state s t The mean absolute difference between the two, L2 error, is calculated by reconstructing the state s ′ t With the real state s t The mean square error between the worst pixel L 2-worst The error is optimized by specifically reconstructing the pixel with the largest error. The three parts of the error are calculated based on the weight coefficients c1, c2, c 2-worst Perform weighted summation to ultimately form the total loss function of the optimization task.

6. The potential consistency planning method for UAV trajectory generation according to claim 1 is characterized in that: The pre-established consistent world model includes a student world model and a teacher world model.

7. The potential consistency planning method for UAV trajectory generation according to claim 6 is characterized in that: The student world model combines the first t D-Features and I-Features to predict the D-Feature at the next time t+1 to predict the overall trajectory direction: D ′ t+1 =WM(D-Feature <＝t ,I-Feature <＝t ) (5) Among them, WM represents the student world model, D ′ t+1 Represents the D-Feature of the next moment. The teacher world model without gradient update capability is used to predict the D-Feature of the next moment t+1 based on the previous t D-Features and I-Feature: D_con ′ t+1 =WMT(D-Feature <＝t ,I-Feature <＝t ) (6) Where WMT represents the teacher world model. According to the prediction formulas (5) and (6) of the consistent world model, the overall optimization task is written as the following optimization problem: c3 and c4 are the weighting coefficients of prediction loss and consistency loss, respectively.

8. A potential consistency planning system for UAV trajectory generation, characterized by: include: A feature extraction module is used to obtain the UAV trajectory sequence data, perform feature extraction based on the UAV trajectory sequence data, obtain the coded random features and the coded deterministic features, and determine the decoding and restoration trajectory state based on the coded random features and the coded deterministic features; The state planning module is used to restore the trajectory state based on decoding, input the encoded random features and encoded deterministic features into a pre-established consistent world model, and output the state features of the next time step.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: The memory stores a computer program that can be run on a processor. When the processor loads and executes the computer program, it adopts a potential consistency planning method for drone trajectory generation according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, wherein: When the computer program is loaded and executed by the processor, the potential consistency planning method for drone trajectory generation according to any one of claims 1 to 7 is adopted.

Images