Excavator bucket tooth tip trajectory planning method and device and excavator

Abstract

The present application relates to the technical field of mechanical engineering, and provides a kind of excavator bucket tooth tip trajectory planning method, device and excavator, wherein the method comprises: obtaining the target joint angle of the current time length of the excavator in the target excavation area operation process and the target point cloud map corresponding to the target joint angle, and the target terrain feature corresponding to the target excavation area;Target joint angle, target point cloud map and target terrain feature are input into bucket tooth tip action prediction model, and the target bucket tooth tip action sequence prediction result output by bucket tooth tip action prediction model is obtained;Trajectory planning is carried out based on target bucket tooth tip action sequence prediction result, and the target bucket tooth tip trajectory planning result of current time length is determined.The present application can not only significantly improve the efficiency and precision of excavation operation, reduce energy consumption and the labor intensity of operator, but also can greatly improve the accuracy and reliability of bucket tooth tip trajectory planning, has important practical value and broad market prospect.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical engineering, and in particular to a method and device for planning the trajectory of an excavator bucket tooth tip, and an excavator. Background Art

[0002] As we all know, bucket tooth tip trajectory planning is crucial to the efficiency and quality of excavator operations. Traditionally, bucket tooth tip trajectory planning for excavators relies primarily on manual control based on the operator's experience. This is labor-intensive and limits efficiency and accuracy. Therefore, improving the automation level and efficiency of excavator operations while reducing energy consumption has become a key issue that needs to be addressed.

[0003] In the related technology, the optimal trajectory planning method based on spline curves can be used to plan the bucket tooth tip trajectory. Although this method can combine spline curves, nonlinear programming and iterative methods to plan time-smooth trajectories, its entire planning process not only relies on pre-set rules, but also has a complex calculation process; alternatively, trajectory optimization and model predictive control methods can be used to plan the bucket tooth tip trajectory. This method needs to consider various types of physical constraints on the controlled object, as well as the uncertainty of the model constructed for the controlled object; thus, the existing bucket tooth tip trajectory planning method has low intelligence and operating efficiency. Summary of the Invention

[0004] The present invention provides an excavator bucket tooth tip trajectory planning method, device and excavator, which are used to solve the defects of the existing bucket tooth tip trajectory planning method in terms of low intelligence and operating efficiency. It can not only significantly improve the efficiency and accuracy of excavation operations, reduce energy consumption and operator labor intensity, but also greatly improve the accuracy and reliability of bucket tooth tip trajectory planning. It has important practical value and broad market prospects.

[0005] The present invention provides a method for planning a tooth tip trajectory of an excavator bucket, comprising:

[0006] Obtaining a target joint angle of the excavator during operation in a target excavation area for a current length of time, a target point cloud map corresponding to the target joint angle, and target terrain features corresponding to the target excavation area;

[0007] Inputting the target joint angle, the target point cloud map, and the target terrain features into a bucket tooth tip motion prediction model, and obtaining a target bucket tooth tip motion sequence prediction result output by the bucket tooth tip motion prediction model;

[0008] Performing trajectory planning based on the target bucket tooth tip motion sequence prediction result to determine the target bucket tooth tip trajectory planning result of the current time length;

[0009] Among them, the bucket tooth tip motion prediction model is obtained by training the Transformer learning model to predict the bucket tooth tip motion sequence based on the sample joint angles of the sample excavator during operation in the sample excavation area for a historical period of time, the sample point cloud map corresponding to the sample joint angles, and the sample terrain features corresponding to the sample excavation area.

[0010] According to the excavator bucket tooth tip trajectory planning method provided by the present invention, the number of the target joint angles is multiple, and the method of obtaining the target joint angles for the current time length during the excavator's operation in the target excavation area includes:

[0011] Acquire initial joint angles with timestamps collected for different preset joints of the excavator within the current time length;

[0012] Preprocessing each of the initial joint angles respectively, and inputting each of the preprocessed joint angles into a joint angle enhancement model to obtain each of the target joint angles output by the joint angle enhancement model;

[0013] The joint angle enhancement model is obtained by training a generative adversarial network model based on the preprocessed sample joint angles, and the generator of the generative adversarial network model includes an LSTM network.

[0014] According to the excavator bucket tooth tip trajectory planning method provided by the present invention, the training process of the joint angle enhancement model includes:

[0015] The generative adversarial network model is trained based on the preprocessed sample joint angles, and a Wasserstein loss value based on the Wasserstein distance between the joint angle enhancement result output by the intermediate network model after a preset number of trainings and the corresponding true joint angle is obtained;

[0016] When it is determined that the Wasserstein loss value is less than or equal to a preset loss threshold, the training is stopped, and the intermediate network model corresponding to the time when the training is stopped is determined as the joint angle enhancement model.

[0017] According to the present invention, a method for planning the trajectory of an excavator bucket tooth tip is provided, wherein the training process of the bucket tooth tip motion prediction model includes:

[0018] For each sample excavator and sample excavation area corresponding to each historical time length, determine a sample time step sequence in a process in which the sample excavator operates from a first position to a second position in the sample excavation area, and determine a sample accumulation reward corresponding to the sample time step sequence;

[0019] Based on the sample joint angles and the sample point cloud maps corresponding to each of the historical time lengths, as well as the sample terrain features and the sample accumulated rewards, the Transformer learning model is trained to predict the bucket tooth tip motion sequence until the output bucket tooth tip motion sequence prediction result meets the preset accuracy requirement, and the bucket tooth tip motion prediction model is determined.

[0020] According to the excavator bucket tooth tip trajectory planning method provided by the present invention, the process of obtaining the target terrain features corresponding to the target excavation area includes:

[0021] Obtaining an initial three-dimensional terrain point cloud map with a timestamp collected by a laser radar for the target excavation area;

[0022] Preprocessing the initial three-dimensional terrain point cloud map;

[0023] Performing raster processing on map difference data between the pre-processed three-dimensional terrain point cloud map and the target three-dimensional engineering map to obtain a grid map to be constructed in the target excavation area;

[0024] Feature extraction is performed on the grid map to be constructed to obtain the target terrain features.

[0025] According to the present invention, a method for planning the tooth tip trajectory of an excavator bucket is provided, the method further comprising:

[0026] Reacquiring post-operation terrain data of the target excavation area while the excavator is operating the target excavation area;

[0027] The target bucket tooth tip trajectory planning result is adjusted based on the post-operation terrain data and the terrain change difference and potential excavation deviation determined by the preset operation plan.

[0028] The present invention also provides an excavator bucket tooth tip trajectory planning device, comprising:

[0029] a data acquisition unit, configured to acquire a target joint angle of the excavator during operation in a target excavation area for a current length of time, a target point cloud map corresponding to the target joint angle, and target terrain features corresponding to the target excavation area;

[0030] a motion prediction unit, configured to input the target joint angle, the target point cloud map, and the target terrain features into a bucket tooth tip motion prediction model, and obtain a target bucket tooth tip motion sequence prediction result output by the bucket tooth tip motion prediction model;

[0031] A trajectory planning unit is configured to perform trajectory planning based on the target bucket tooth tip action sequence prediction result, and determine a target bucket tooth tip trajectory planning result for the current time length.

[0032] The bucket tooth tip action prediction model is obtained by performing bucket tooth tip action sequence prediction training on a Transformer learning model based on sample joint angles of a sample excavator in a sample excavation area during a historical time length, a sample point cloud map corresponding to the sample joint angles, and sample terrain features corresponding to the sample excavation area.

[0033] The application further provides an excavator comprising the excavator bucket tooth tip trajectory planning device.

[0034] The application further provides an electronic device comprising 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 excavator bucket tooth tip trajectory planning method according to any one of the preceding embodiments when executing the program.

[0035] The application further provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executable on a processor to implement the steps of the excavator bucket tooth tip trajectory planning method according to any one of the preceding embodiments.

[0036] The excavator bucket tooth tip trajectory planning method, device, and excavator provided by the application can realize high-precision and high-efficiency automatic control of the excavation operation without defining an accurate physical model or complex rules in advance, can significantly improve the efficiency and precision of the excavation operation, reduce energy consumption and the labor intensity of the operator, and can greatly improve the accuracy and reliability of the bucket tooth tip trajectory planning by combining the reinforcement learning capability of the Transformer learning model, and thus has important practical value and broad market prospects. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0038] Figure 1 It is a flow chart of the excavator bucket tooth tip trajectory planning method provided by the present invention;

[0039] Figure 2 This is a block diagram of the decision-making principle of the Transformer learning model provided by the present invention;

[0040] Figure 3 It is a schematic diagram of the structure of the depth-separable convolutional network provided by the present invention;

[0041] Figure 4 It is a structural schematic diagram of the excavator bucket tooth tip trajectory planning device provided by the present invention;

[0042] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention; DETAILED DESCRIPTION

[0043] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0044] In the embodiments of the present invention, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. In the textual description of the present invention, the character " / " generally indicates that the previous and next associated objects are in an "or" relationship. In addition, it should be noted that the serial numbers themselves for the objects described in the present invention, such as "first", "second", etc., are only used to distinguish the objects described, and do not have any order or technical meaning.

[0045] Currently, traditional excavator bucket tip trajectory planning relies primarily on manual control based on the operator's experience. This is labor-intensive and limits efficiency and accuracy. Therefore, improving the automation level and efficiency of excavator operations while reducing energy consumption has become a key issue that needs to be addressed.

[0046] In related technologies, an optimal trajectory planning method based on spline curves can be used to plan the bucket tooth tip trajectory. This method aims to convert the excavator's excavation trajectory into a topologically equivalent path that is fast, smooth, and dynamically feasible. These path points are connected using spline curves, and the trajectory is reparameterized. Finally, a minimum time-smooth trajectory that meets the kinematic constraints is generated through nonlinear programming and iteration. Although this method can generate a fast, smooth, and dynamically feasible trajectory and fully utilize the dynamic feasibility of the excavator, its entire planning process not only relies on pre-set rules but also requires complex computational processes to achieve trajectory reparameterization and optimization, which has significant limitations in environments with insufficient computing resources.

[0047] Alternatively, an industry-proven trajectory optimization and model predictive control approach can be employed. This approach focuses on using trajectory optimization and model predictive control (MPC) to plan the motion of a hydraulic excavator in real time, while also considering four types of physical constraints: force / torque limits, power limits, cylinder displacement limits, and pump flow limits. Furthermore, the uncertainty of the model constructed for the controlled object must be considered, and disturbance estimation is introduced to achieve disturbance perception. Finally, the MPC problem is reformulated through feedback linearization to ensure the real-time applicability of the proposed local trajectory planning. While this approach can plan trajectories in real time and achieves strong adaptability and robustness by considering multiple physical and operational constraints, it places high demands on the accuracy of the dynamic model, requires precise models and disturbance estimation, and has a relatively high computational burden.

[0048] In summary, the existing bucket tooth tip trajectory planning methods are not very intelligent and have low operating efficiency.

[0049] In order to solve the above technical problems, the present invention provides an excavator bucket tooth tip trajectory planning method, device and excavator.

[0050] The following combination Figures 1 to 5The present invention describes an excavator bucket tooth tip trajectory planning method, device, and excavator. The method can be executed by a built-in processor in the excavator or an external device connected to the excavator. Both the processor and the external device have at least information collection, data enhancement, data preprocessing, reinforcement learning, model training, feature extraction, and trajectory planning functions. The processor can be a central processing unit (CPU) and can be connected to a display module, which ensures that the corresponding display displays information to the user. The external device can include, but is not limited to, other devices such as personal computers (PCs), portable devices, laptops, smartphones, tablets, and portable wearable devices. The present invention does not limit the specific form of the processor or external device. Furthermore, the method can also be applied to an excavator bucket tooth tip trajectory planning device located in the processor or external device. The excavator bucket tooth tip trajectory planning device can be implemented using software, hardware, or a combination of both. The following describes the excavator bucket tooth tip trajectory planning method by taking the excavator's built-in processor as an example.

[0051] To facilitate understanding of the excavator bucket tooth tip trajectory planning method provided by the embodiments of the present invention, the following exemplary embodiments will be used to describe the excavator bucket tooth tip trajectory planning method provided by the present invention in detail. It is understood that the following exemplary embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0052] Reference Figure 1 , which is a flow chart of the excavator bucket tooth tip trajectory planning method provided by the present invention, as shown Figure 1 As shown, the excavator bucket tooth tip trajectory planning method includes the following steps 110 to 130.

[0053] Step 110 : Obtain the target joint angle of the excavator during the current time length of operation in the target excavation area, the target point cloud map corresponding to the target joint angle, and the target terrain features corresponding to the target excavation area.

[0054] Among them, the target excavation area can be the area where the excavation operation is performed, and the target joint angles can include but are not limited to the joint angles of the excavator's arm, dipper arm, bucket, rotation and other joints, and each joint angle carries a timestamp, and each timestamp can be used to represent the time when the corresponding joint angle is collected; the number of target joint angles is multiple.

[0055] The target point cloud map can specifically be a point cloud map generated based on multiple target joint angles, and based on the target point cloud map, not only the current motion state of the excavator within the current time length can be known, but also the oil pressure parameters of the excavator within the current time length can be known, such as the large chamber pressure of the hydraulic cylinder.

[0056] The target terrain features may specifically be key terrain features of the three-dimensional terrain to be constructed in the target excavation area.

[0057] It should be noted that, for the same time length (such as the current time length), the number of the excavator's current motion states is the same as the number of target terrain features and the number of target point cloud maps, and they correspond one to one.

[0058] It can be understood that the built-in processor of the excavator obtains the target joint angle. When angle sensors are set on other joints of the excavator such as the boom, dipper arm, bucket and rotation, the processor can obtain the joint angle reported by each angle sensor in real time based on the data reporting protocol agreed in advance with each angle sensor.

[0059] The processor obtains the target point cloud map, and can process the three-dimensional data set of each target angle data in the three-dimensional space into a map form to obtain a point cloud map.

[0060] The processor obtains target terrain features, and can determine the three-dimensional terrain data to be constructed based on the target excavation area, and then convert the three-dimensional terrain data into a raster map to extract key terrain features.

[0061] Step 120 : Input the target joint angle, the target point cloud map, and the target terrain features into the bucket tooth tip motion prediction model to obtain the target bucket tooth tip motion sequence prediction result output by the bucket tooth tip motion prediction model.

[0062] Among them, the bucket tooth tip motion prediction model is obtained by training the Transformer learning model to predict the bucket tooth tip motion sequence based on the sample joint angles of the sample excavator during operation in the sample excavation area for a historical period of time, the sample point cloud map corresponding to the sample joint angles, and the sample terrain features corresponding to the sample excavation area.

[0063] It is understandable that the Transformer learning model is trained using a large amount of historical mining operation data, so that the Transformer learning model learns from each historical mining operation data how to predict the bucket tooth tip motion sequence based on the sample joint angles, sample point cloud maps, and sample terrain features of the historical time length. Training is stopped when the error between the predicted bucket tooth tip motion sequence and the corresponding actual bucket tooth tip motion sequence is minimized, and the intermediate Transformer learning model corresponding to the training stop is determined to be the bucket tooth tip motion prediction model. In this way, in actual mining operations, the target terrain features, target joint angles, and target point cloud maps obtained in real time can be input into the bucket tooth tip motion prediction model to quickly predict the bucket tooth tip motion at each time step, thereby obtaining the target bucket tooth tip motion sequence prediction result for the current time length.

[0064] It should be noted that among a large amount of historical mining operation data, each historical mining operation data corresponds to a different historical time length, and each historical mining operation data may include but is not limited to the sample joint angle of the corresponding sample excavator in the operation process of the corresponding sample mining area for the corresponding historical time length and the sample point cloud map corresponding to the sample joint angle, and the sample terrain features corresponding to the sample mining area.

[0065] Step 130 : Perform trajectory planning based on the target bucket tooth tip motion sequence prediction result to determine the target bucket tooth tip trajectory planning result for the current time length.

[0066] Specifically, when the target bucket tooth tip motion sequence prediction result output by the bucket tooth tip motion prediction model is obtained, trajectory planning can be performed in combination with the current underground state of the target excavation area and the pre-set excavation target until the target bucket tooth tip trajectory planning result of the current time length is obtained.

[0067] The excavator bucket tooth tip trajectory planning method provided by the present invention obtains a bucket tooth tip motion prediction model after pre-training a Transformer learning model based on a large amount of historical excavation operation data. By utilizing the bucket tooth tip motion prediction model to predict the bucket tooth tip motion sequence based on the real-time input target terrain features, target joint angles and target point cloud maps, and planning the bucket tooth tip trajectory within the current time length based on the predicted bucket tooth tip motion sequence, the purpose of high-precision and high-efficiency automated control of the excavation operation is achieved. There is no need to manually define precise physical models or complex rules in advance. Not only can the efficiency and accuracy of the excavation operation be significantly improved, energy consumption and the labor intensity of the operator can be reduced, but the accuracy and reliability of the bucket tooth tip trajectory planning can also be greatly improved by combining the reinforcement learning ability of the Transformer learning model. It has important practical value and broad market prospects.

[0068] Based on the above Figure 1 In one exemplary embodiment of the excavator bucket tooth tip trajectory planning method shown, considering that a small amount of data can affect the accuracy of model prediction results, the collected initial joint angles can be preprocessed and then enhanced to construct more target joint angles. Based on this, if there are multiple target joint angles, step 110 obtains the target joint angles for the current duration of the excavator's operation in the target excavation area. The specific implementation process may include:

[0069] First, the initial joint angles with timestamps collected for different preset joints of the excavator are obtained within the current time length; further, each initial joint angle is preprocessed respectively, and the preprocessed joint angles are input into the joint angle enhancement model to obtain the target joint angles output by the joint angle enhancement model.

[0070] Among them, the joint angle enhancement model is obtained by training the generative adversarial network model based on the preprocessed sample joint angles, and the generator of the generative adversarial network model includes an LSTM network.

[0071] Specifically, when angle data acquisition modules are respectively set at the excavator's arm, dipper arm, bucket, rotation and other joints, the initial joint angle reported by each angle data module can be obtained within the current time length according to the data acquisition protocol agreed upon in advance with each angle data acquisition module, and then a timestamp is added to each obtained initial joint angle, and each timestamp is used to represent the acquisition moment of the corresponding initial joint angle; then, each initial joint angle carrying a timestamp is preprocessed by other operations such as denoising and filtering, and the preprocessed joint angles are input into the joint angle enhancement model to mine more joint angle data, thereby obtaining a large number of target joint angles output by the joint angle enhancement model.

[0072] It should be noted that, when the LSTM network is included in the generator of the generative adversarial network model, the generative adversarial network model is trained by using a large number of preprocessed sample joint angles, so that the generative adversarial network model performs data enhancement training on each preprocessed sample joint angle, so as to mine more joint angle data; the training is stopped until the mined joint angle data is sufficient and / or the model loss after multiple trainings is small enough, and the model corresponding to the time when the training is stopped is determined as the joint angle enhancement model. In this way, in the actual mining operation, all the preprocessed joint angles obtained in real time can be input into the joint angle enhancement model to quickly construct and mine more joint angle data, thereby obtaining all the target joint angles output by the joint angle enhancement model.

[0073] Based on the above Figure 1In the excavator bucket tooth tip trajectory planning method shown in FIG, in an exemplary embodiment, the training process of the joint angle enhancement model includes:

[0074] First, the generative adversarial network model is trained based on the preprocessed sample joint angles, and the Wasserstein loss value based on the Wasserstein distance between the joint angle enhancement result output by the intermediate network model after a preset number of training and the corresponding true joint angle is obtained; then, the training is stopped when it is determined that the Wasserstein loss value is less than or equal to the preset loss threshold, and the intermediate network model corresponding to the time when the training is stopped is determined as the joint angle enhancement model.

[0075] Among them, the real joint angle can be used to characterize the sample joint angle participating in the model training and its training result is the joint angle enhancement result.

[0076] Specifically, when a large number of preprocessed sample joint angles are present, the large number of preprocessed sample joint angles can be divided into multiple batches, each batch containing at least one preprocessed sample joint angle. The number of sample joint angles contained in adjacent batches can be the same or different. This is not specifically limited here.

[0077] In this way, by performing preset batch training on the generative adversarial network model, an intermediate network model after a preset number of trainings can be obtained. All sample joint angles in each batch participate in one training, and it can be considered that one training is completed.

[0078] At this time, the Wasserstein loss value based on the Wasserstein distance between the joint angle enhancement result output by the intermediate network model after a preset number of training and the corresponding true joint angle can be calculated, and the Wasserstein loss value can be compared with the set loss threshold that represents a sufficiently high training accuracy of the model.

[0079] When it is determined that the Wasserstein loss value is lower than a preset loss threshold, training can be stopped, and the intermediate network model corresponding to the time of training stop can be determined as the joint angle enhancement model. Conversely, when it is determined that the Wasserstein loss value is greater than the preset loss threshold, the next preset batch of sample joint angles can be selected from a large number of preprocessed sample joint angles, and the intermediate network model corresponding to the time when the Wasserstein loss value is greater than the preset loss threshold can be trained again, until a joint angle enhancement model that meets the training stop condition is obtained.

[0080] It should be noted that the preset batch of sample joint angles selected again can be the same as or different from the previous preset batch of sample joint angles. Furthermore, the selection of the next preset batch of sample joint angles from a large number of pre-processed sample joint angles can be done by randomly sorting all the pre-processed sample joint angles, or by selecting from the remaining sample joint angles that have not participated in training. The selection method can be sequential selection or intermittent selection. The present invention does not impose any specific limitations on this.

[0081] In addition, it should be noted that to better generate trajectory data, this invention improves traditional generative adversarial networks (GANs): 1) The generator includes a long short-term memory (LSTM) network to generate sequence data, and the generated sequence data is batched, with each LSTM unit outputting multiple time points to improve temporal correlation; 2) It supports the training and generation of edge-length sequences; 3) It supports fixed traversals (attributes) that do not change over time, which is common in time series data; 4) It supports per-case scaling of continuous variables to handle data with a large dynamic range; and 5) It uses Wasserstein loss and gradient penalty to reduce prone mode collapse and improve the training process. As a result, compared to traditional GANs, this invention demonstrates significant advantages in processing and generating complex data with temporal correlation, and is particularly capable of learning and generating data with temporal correlation at different scales.

[0082] For example, to address the issues of unstable training and prone to mode collapse in traditional generative adversarial networks, the present invention proposes a loss function based on the Wasserstein distance, which can achieve a more stable training process and higher-quality generated data. The key to this loss function is to calculate the Wasserstein distance between the generated data (i.e., the joint angle enhancement result) and the real data (i.e., the real joint angle). For the discriminator (D) and the generator (G), the Wasserstein loss value can be defined as:

[0083] \[L=\mathbb{E}_{x\sim\mathbb{P}_{r}}[D(x)]-\mathbb{E}_{\tilde{x}\sim\mathbb{P}_{g}}[D(G(\tilde{x}))]\]

[0084] Among them, \(\mathbb{E}_{x\sim\mathbb{P}_{r}}[D(x)]\) represents the expected value of the discriminator's score on the real data \(x\),

[0085] \(\mathbb{E}_{\tilde{x}\sim\mathbb{P}_{g}}[D(G(\tilde{x}))]\) represents the expected value of the discriminator's score on the generated data \(G(\tilde{x})\).

[0086] For the Wasserstein distance to be effective, the discriminator (evaluator) must be 1-Lipschitz continuous. This can be achieved by imposing constraints on the discriminator weights. The most common methods are weight clipping or gradient penalty. Weight clipping can be done by limiting the weights to a fixed range, such as \([-c,c]\), where c is an integer. Gradient penalty: A penalty term is added to constrain the norm of the gradient to ensure 1-Lipschitz continuity.

[0087] Based on the above Figure 1 In the excavator bucket tooth tip trajectory planning method shown in FIG, in an exemplary embodiment, the training process of the bucket tooth tip motion prediction model includes:

[0088] First, for the sample excavators and sample excavation areas corresponding to each historical time length, the sample time step sequence of the sample excavator in the process of operating from the first position to the second position in the sample excavation area is determined, and the sample accumulated reward corresponding to the sample time step sequence is determined; then, based on the sample joint angles and sample point cloud maps corresponding to each historical time length, as well as the sample terrain features and sample accumulated rewards, the Transformer learning model is trained to predict the bucket tooth tip motion sequence, and the bucket tooth tip motion prediction model is determined when the output bucket tooth tip motion sequence prediction result meets the preset accuracy requirement.

[0089] Specifically, for the large amount of historical mining operation data obtained, each historical mining operation data is operation data generated after a corresponding sample excavator operated in a corresponding sample mining area for a corresponding historical length of time. Among all the sample excavators covered by all the historical mining operation data, at least two sample excavators may be the same, or all sample excavators may be different. This is not specifically limited here.

[0090] In order to efficiently clear each area to be constructed (such as each sample excavation area), the volume of each area to be constructed can be calculated, and the negative value of each volume can be used as a reward. Furthermore, the sum of the expected rewards that can be obtained from a given historical time step to the end of the corresponding sample time step sequence can be used as the sample cumulative reward (Return To Go, RTG) corresponding to the sample time step sequence.

[0091] For example, for a sample time step sequence t, the corresponding sample accumulation reward Rt It can be calculated by formula (1).

[0092] (1)

[0093] In formula (1), T represents the end time step of the corresponding sample time step sequence, represents the instantaneous reward of the sample obtained at a given sample time step K; Represents a pre-set discount factor and is a natural number between 0 and 1, used to measure the present value of future rewards; when , which means no discount on future rewards.

[0094] Each historical time length corresponds to a large number of sample joint angles. These sample joint angles are obtained after preprocessing with denoising and filtering, followed by joint angle data mining using a joint angle enhancement model. These sample joint angles for each historical time length can then be used as the decision action for that historical time length.

[0095] The sample point cloud maps corresponding to each historical time length are obtained by processing the point cloud maps of a large number of sample joint angles corresponding to the corresponding historical time length. And from each sample point cloud map, the historical hydraulic cylinder cavity pressure and historical motion state of the corresponding sample excavator within the corresponding historical time length can be determined. In addition, the sample terrain features corresponding to each historical time length are obtained by first scanning the corresponding sample excavation area for the corresponding historical time length to obtain a high-precision sample three-dimensional terrain point cloud map with a timestamp, then performing 2.5-dimensional raster processing and preprocessing on the difference map between the sample three-dimensional terrain point cloud map and the corresponding expected three-dimensional engineering map, and finally performing feature extraction on the processed sample grid map to be constructed.

[0096] At this time, the decision state State corresponding to each historical time length can be constructed based on the combination of the historical hydraulic cylinder large chamber pressure historical movement state, sample terrain characteristics and sample point cloud map.

[0097] So far, refer to Figure 2 The decision principle block diagram of the Transformer learning model shown in the figure is to accumulate the reward R of the sample corresponding to each historical time length. t , decision-making state State (can be recorded as a t ) and decision action Action (which can be recorded as S t) is combined into a state-action-reward (SAR), and used as the input data for training the Transformer learning model. The Transformer learning model is trained to predict the bucket tooth tip motion sequence until the bucket tooth tip motion sequence prediction result output by the intermediate Transformer learning model after a preset number of trainings meets the preset accuracy requirement. The training is stopped when the intermediate Transformer learning model corresponding to the time when the training stops is determined as the bucket tooth tip motion prediction model.

[0098] It should be noted that the bucket tooth tip motion sequence prediction result output by the intermediate Transformer learning model after training for a preset number of times meets the preset accuracy requirements, and the error value between the actual bucket tooth tip motion sequence in the historical mining operation data participating in the model training and the bucket tooth tip motion sequence prediction result is minimized or lower than the preset error threshold.

[0099] For example, by minimizing the cross entropy loss between the predicted bucket tooth tip motion sequence and the actual bucket tooth tip motion sequence, training is stopped until the cross entropy loss value reaches a minimum or falls below a preset loss threshold. This ensures that the model learns to make decisions that lead to higher rewards.

[0100] In addition, it should be noted that after each input data enters the Transformer learning model, the Transformer learning model first uses the vector state linear layer or the frame-based state CNN encoding layer to perform token embedding, and then uses the self-attention characteristics of the causal self-attention mask to predict the bucket tooth tip action at each time step in the corresponding historical time length, until the bucket tooth tip action sequence prediction result of the corresponding historical time length is obtained.

[0101] Exemplarily, the Transformer learning model can represent each input data consisting of a state-action-reward (SAR) tuple as a tag sequence, which is then embedded into a continuous vector using an embedding layer; position encoding is then added to the embedded input vector to capture the relative position of the input tag; the multi-head attention mechanism of the multi-head self-attention layer is further used to calculate the attention score of each input tag to capture the dependency between different tags in the input sequence; after the multi-head self-attention layer, the continuous output vector is mapped back to the action space through the output layer after entering the feedforward neural network layer, residual connection, normalization layer and stacking layer to generate a predicted action sequence, which is processed into a bucket tooth tip action sequence of corresponding historical time length and then output.

[0102] It can be seen that in order to generate future decisions that achieve specified expected returns, the present invention innovatively proposes a Transformer learning model specifically designed for tasks involving step-by-step decision-making. This Transformer learning model is good at receiving information sequences and generating action sequences, and can make wise decisions in a structured and continuous manner.

[0103] This innovative approach relies on an autoregressive model conditioned on the expected reward, past states, and actions. By using generative trajectory modeling to predict the combined pattern of situations, actions, and rewards, the method simplifies the reinforcement learning process, bypassing the traditional reward-maximizing reinforcement learning process and directly generating a sequence of future actions that will achieve the expected reward.

[0104] The excavator bucket tooth tip trajectory planning method provided by the present invention can process long time series data by reinforcing the training of the Transformer learning model, which is particularly critical for tasks such as mining operations that need to consider historical operation information to make current decisions; furthermore, because the reinforcement learning Transformer model is based on a large amount of historical mining operation data, the trained bucket tooth tip motion prediction model has better generalization ability, so that it can adapt to changing working environments and different mining tasks.

[0105] Based on the above Figure 1 In the excavator bucket tooth tip trajectory planning method shown in FIG, in an exemplary embodiment, the process of obtaining target terrain features corresponding to the target excavation area in step 110 includes:

[0106] First, an initial three-dimensional terrain point cloud map with a timestamp collected by the lidar for the target excavation area is obtained. Then, the initial three-dimensional terrain point cloud map is preprocessed. The map difference data between the preprocessed three-dimensional terrain point cloud map and the target three-dimensional engineering map is further rasterized to obtain a grid map of the target excavation area to be constructed. Feature extraction is then performed on the grid map to obtain the target terrain features.

[0107] Specifically, when a laser radar is installed on the excavator, a terrain data acquisition protocol can be pre-set with the laser radar so that during the operation of the excavator, a high-precision initial three-dimensional terrain point cloud map with a timestamp determined after the laser radar scans the target excavation area can be obtained in real time.

[0108] To facilitate the bucket tooth tip motion prediction model in identifying the terrain of the target excavation area, the initial 3D terrain point cloud map acquired in real time can be subtracted from the expected target 3D engineering map to obtain a 3D map to be constructed. The 3D map to be constructed is then subjected to 2.5D raster processing to obtain a raster map to be constructed of the target excavation area. Target terrain features representing key terrain features can then be extracted from the raster map to be constructed.

[0109] It should be noted that in order to design a simple, efficient, and computationally inefficient convolutional neural network, the present invention uses a depthwise separable convolutional network for feature extraction. This network consists of a depthwise convolutional layer that filters the input channels and a pointwise convolutional layer that combines them to create new features. In the case of depthwise convolution, with an input feature map of dimension DF*DF and M kernels of channel size 1, the total computational cost can be 1 / M of that of a standard convolution, where M is a positive integer.

[0110] Since depthwise convolution is only used to filter input channels, it cannot produce new features by combining them. Therefore, an additional layer called a pointwise convolution layer is created, which uses 1×1 convolution to calculate the linear combination of the depthwise convolution output.

[0111] For example, refer to Figure 3 The structural diagram of the depth-separable convolutional network shown in Figure 3 In the figure, the 3×3 Depthwise Conv layer is a 3×3 depthwise convolution layer, and the 1×1 Conv layer is a 1×1 pointwise convolution layer. Each convolution layer is followed by a batch normalization (BN) layer and a ReLU layer, which uses the ReLU function for linear rectification. Furthermore, a final average pooling operation is introduced before the fully connected layer to reduce the spatial dimension to 1. This shows that the depthwise separable convolutional network architecture provided by this invention is sufficiently small and computationally inefficient.

[0112] Based on the above Figure 1 In an exemplary embodiment of the excavator bucket tooth tip trajectory planning method shown, to ensure higher trajectory planning accuracy, the bucket tooth tip trajectory can be dynamically adjusted based on terrain data updated during actual excavation operations. Based on this, after step 130, the excavator bucket tooth tip trajectory planning method provided by the present invention can further include:

[0113] First, while the excavator is operating in the target excavation area, the post-operation terrain data of the target excavation area is reacquired. Then, based on the post-operation terrain data and the terrain change differences and potential excavation deviations determined by the preset operation plan, the target bucket tooth tip trajectory planning results are adjusted.

[0114] Specifically, during the excavation process, not only can the progress of the excavation operation and the status of the excavator be continuously monitored, but the post-operation terrain data of the target excavation area can also be rescanned. By comparing the latest scanned post-operation terrain data with the target operation plan set in advance for the target excavation area, the terrain change differences and potential excavation deviations can be determined. Based on the terrain change differences and potential excavation deviations, the bucket tooth tip action sequence is recalculated through the bucket tooth tip action prediction model, so that the bucket tooth tip trajectory can be replanned based on the recalculated bucket tooth tip action sequence, thereby obtaining the adjusted target bucket tooth tip trajectory planning result to adapt to terrain changes and actual operation requirements, and ensure the efficient and accurate completion of the excavation task.

[0115] It should be noted that during the actual operation process, the optimal bucket tooth tip trajectory can also be calculated based on real-time monitoring of the excavator's position, posture and bucket status, as well as the real-time updated three-dimensional terrain point cloud map and the trained bucket tooth tip motion prediction model, and the optimal bucket tooth tip trajectory can be sent to the excavator's control system so that the control system can guide the excavator to perform precise excavation actions based on the optimal bucket tooth tip trajectory.

[0116] In addition, it should be noted that for the enhanced training process of the Transformer learning model, if the processor has a performance evaluation function, it can also evaluate the performance of the mining operation and obtain performance evaluation results based on indicators such as the quality, efficiency, and energy consumption of the mining completion. The performance evaluation results are then used as feedback to optimize the parameters and training process of the Transformer learning model, thereby continuously improving the overall performance of the system.

[0117] Furthermore, it should be noted that if the excavator has a built-in display module and an external display, the operator can monitor the real-time status of the excavation operation through the display, view the target bucket tip trajectory planning results for the excavator bucket, and manually adjust the target bucket tip trajectory if the operator determines based on their experience that the target bucket tip trajectory planning results are inaccurate or have large deviations. Furthermore, the display provides a visual display of performance evaluation results, helping operators and administrators understand operational efficiency and system performance.

[0118] The excavator bucket tooth tip trajectory planning method provided by the present invention dynamically adjusts the trajectory planning of the bucket tooth tip by receiving real-time feedback on the working environment and excavator status, thereby ensuring flexible response to various unexpected situations that may be encountered during the operation, such as sudden obstacles or changes in terrain, thereby greatly improving the flexibility and dynamic adaptability of the excavator bucket tooth tip trajectory planning.

[0119] The excavator bucket tooth tip trajectory planning device provided by the present invention is described below. The excavator bucket tooth tip trajectory planning device described below and the excavator bucket tooth tip trajectory planning method described above can be referenced to each other.

[0120] Reference Figure 4 , which is a structural diagram of the excavator bucket tooth tip trajectory planning device provided by the present invention, as shown Figure 4 As shown, the excavator bucket tooth tip trajectory planning device 400 may include: a data acquisition unit 410 , an action prediction unit 420 and a trajectory planning unit 430 .

[0121] The data acquisition unit 410 is used to acquire the target joint angle of the excavator during the current time length of operation in the target excavation area, the target point cloud map corresponding to the target joint angle, and the target terrain features corresponding to the target excavation area;

[0122] The motion prediction unit 420 is used to input the target joint angle, the target point cloud map and the target terrain features into the bucket tooth tip motion prediction model, and obtain the target bucket tooth tip motion sequence prediction result output by the bucket tooth tip motion prediction model;

[0123] The trajectory planning unit 430 is used to perform trajectory planning based on the target bucket tooth tip motion sequence prediction result and determine the target bucket tooth tip trajectory planning result of the current time length;

[0124] Among them, the bucket tooth tip motion prediction model is obtained by training the Transformer learning model to predict the bucket tooth tip motion sequence based on the sample joint angles of the sample excavator during operation in the sample excavation area for a historical period of time, the sample point cloud map corresponding to the sample joint angles, and the sample terrain features corresponding to the sample excavation area.

[0125] Optionally, the data acquisition unit 410 is specifically configured to acquire initial joint angles with timestamps collected for different preset joints of the excavator within a current time period; preprocess each initial joint angle, and input the preprocessed joint angles into a joint angle enhancement model to obtain target joint angles output by the joint angle enhancement model;

[0126] Among them, the joint angle enhancement model is obtained by training the generative adversarial network model based on the preprocessed sample joint angles, and the generator of the generative adversarial network model includes an LSTM network.

[0127] Optionally, the excavator bucket tooth tip trajectory planning device provided by the present invention may further include a model training unit, which is used to train the generative adversarial network model based on the preprocessed sample joint angles, and obtain the Wasserstein loss value based on the Wasserstein distance between the joint angle enhancement result output by the intermediate network model after a preset number of trainings and the corresponding true joint angle; the training is stopped when it is determined that the Wasserstein loss value is less than or equal to a preset loss threshold, and the intermediate network model corresponding to the time when the training is stopped is determined as the joint angle enhancement model.

[0128] Optionally, the model training unit is further used to determine, for the sample excavators and sample excavation areas corresponding to each historical time length, a sample time step sequence of the sample excavator in the process of operating from the first position to the second position in the sample excavation area, and determine the sample accumulation reward corresponding to the sample time step sequence; based on the sample joint angles and sample point cloud maps corresponding to each historical time length, as well as the sample terrain features and sample accumulation rewards, the Transformer learning model is trained to predict the bucket tooth tip motion sequence, and the bucket tooth tip motion prediction model is determined until the output bucket tooth tip motion sequence prediction result meets the preset accuracy requirement.

[0129] Optionally, the data acquisition unit 410 is specifically used to obtain an initial three-dimensional terrain point cloud map with a timestamp collected by the laser radar for the target excavation area; preprocess the initial three-dimensional terrain point cloud map; perform raster processing on the map difference data between the preprocessed three-dimensional terrain point cloud map and the target three-dimensional engineering map to obtain a raster map to be constructed in the target excavation area; perform feature extraction on the raster map to be constructed to obtain target terrain features.

[0130] Optionally, the excavator bucket tooth tip trajectory planning device provided by the present invention may also include a trajectory adjustment unit, which is used to re-acquire the post-operation terrain data of the target excavation area while the excavator is operating in the target excavation area; and adjust the target bucket tooth tip trajectory planning result based on the post-operation terrain data and the terrain change differences and potential excavation deviations determined by the preset operation plan.

[0131] The excavator bucket tooth tip trajectory planning device 400 provided in an embodiment of the present invention can implement the technical solution in any embodiment of the above-mentioned excavator bucket tooth tip trajectory planning method, which will not be described in detail here.

[0132] The present invention also provides an excavator, which includes the excavator bucket tooth tip trajectory planning device described in the aforementioned embodiment. The specific implementation process involved can refer to the technical solution in any embodiment of the above-mentioned excavator bucket tooth tip trajectory planning method, and will not be repeated here.

[0133] Figure 4 An example of a physical structure diagram of an electronic device is shown below. Figure 4 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the excavator bucket tooth tip trajectory planning method, which includes:

[0134] Obtaining the target joint angle of the excavator during the current time length of operation in the target excavation area, the target point cloud map corresponding to the target joint angle, and the target terrain features corresponding to the target excavation area;

[0135] Inputting the target joint angle, target point cloud map and target terrain features into the bucket tooth tip motion prediction model, and obtaining the target bucket tooth tip motion sequence prediction result output by the bucket tooth tip motion prediction model;

[0136] Perform trajectory planning based on the target bucket tooth tip motion sequence prediction result to determine the target bucket tooth tip trajectory planning result for the current time length;

[0137] Among them, the bucket tooth tip motion prediction model is obtained by training the Transformer learning model to predict the bucket tooth tip motion sequence based on the sample joint angles of the sample excavator during operation in the sample excavation area for a historical period of time, the sample point cloud map corresponding to the sample joint angles, and the sample terrain features corresponding to the sample excavation area.

[0138] Furthermore, the logic instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0139] On the other hand, the present invention further provides a computer program product, comprising a computer program stored on a non-transitory computer-readable storage medium, wherein the computer program comprises program instructions. When the program instructions are executed by a computer, the computer is capable of executing the excavator bucket tooth tip trajectory planning method provided by the above methods, the method comprising:

[0140] Obtaining the target joint angle of the excavator during the current time length of operation in the target excavation area, the target point cloud map corresponding to the target joint angle, and the target terrain features corresponding to the target excavation area;

[0141] Inputting the target joint angle, target point cloud map and target terrain features into the bucket tooth tip motion prediction model, and obtaining the target bucket tooth tip motion sequence prediction result output by the bucket tooth tip motion prediction model;

[0142] Perform trajectory planning based on the target bucket tooth tip motion sequence prediction result to determine the target bucket tooth tip trajectory planning result for the current time length;

[0143] Among them, the bucket tooth tip motion prediction model is obtained by training the Transformer learning model to predict the bucket tooth tip motion sequence based on the sample joint angles of the sample excavator during operation in the sample excavation area for a historical period of time, the sample point cloud map corresponding to the sample joint angles, and the sample terrain features corresponding to the sample excavation area.

[0144] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the above-mentioned excavator bucket tooth tip trajectory planning method, the method comprising:

[0145] Obtaining the target joint angle of the excavator during the current time length of operation in the target excavation area, the target point cloud map corresponding to the target joint angle, and the target terrain features corresponding to the target excavation area;

[0146] Inputting the target joint angle, target point cloud map and target terrain features into the bucket tooth tip motion prediction model, and obtaining the target bucket tooth tip motion sequence prediction result output by the bucket tooth tip motion prediction model;

[0147] Perform trajectory planning based on the target bucket tooth tip motion sequence prediction result to determine the target bucket tooth tip trajectory planning result for the current time length;

[0148] Among them, the bucket tooth tip motion prediction model is obtained by training the Transformer learning model to predict the bucket tooth tip motion sequence based on the sample joint angles of the sample excavator during operation in the sample excavation area for a historical period of time, the sample point cloud map corresponding to the sample joint angles, and the sample terrain features corresponding to the sample excavation area.

[0149] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0150] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for planning the tooth tip trajectory of an excavator bucket, characterized in that: include: Obtaining a target joint angle of the excavator during operation in a target excavation area for a current length of time, a target point cloud map corresponding to the target joint angle, and target terrain features corresponding to the target excavation area; Inputting the target joint angle, the target point cloud map, and the target terrain features into a bucket tooth tip motion prediction model, and obtaining a target bucket tooth tip motion sequence prediction result output by the bucket tooth tip motion prediction model; Performing trajectory planning based on the target bucket tooth tip motion sequence prediction result to determine the target bucket tooth tip trajectory planning result of the current time length; The bucket tooth tip motion prediction model is obtained by training a Transformer learning model for bucket tooth tip motion sequence prediction based on the sample joint angles of the sample excavator during operation in the sample excavation area over a historical period of time, the sample point cloud map corresponding to the sample joint angles, and the sample terrain features corresponding to the sample excavation area. There are multiple target joint angles, and obtaining the target joint angles for the current time length of the excavator's operation in the target excavation area includes: Acquire initial joint angles with timestamps collected for different preset joints of the excavator within the current time length; Preprocessing each of the initial joint angles respectively, and inputting each of the preprocessed joint angles into a joint angle enhancement model to obtain each of the target joint angles output by the joint angle enhancement model; The joint angle enhancement model is obtained by training a generative adversarial network model based on the preprocessed sample joint angles, and the generator of the generative adversarial network model includes an LSTM network; The training process of the joint angle enhancement model includes: The generative adversarial network model is trained based on the preprocessed sample joint angles, and a Wasserstein loss value based on the Wasserstein distance between the joint angle enhancement result output by the intermediate network model after a preset number of trainings and the corresponding true joint angle is obtained; When it is determined that the Wasserstein loss value is less than or equal to a preset loss threshold, the training is stopped, and the intermediate network model corresponding to the time when the training is stopped is determined as the joint angle enhancement model.

2. The excavator bucket tooth tip trajectory planning method according to claim 1, characterized in that: The training process of the bucket tooth tip motion prediction model includes: For each sample excavator and sample excavation area corresponding to each historical time length, determine a sample time step sequence in a process in which the sample excavator operates from a first position to a second position in the sample excavation area, and determine a sample accumulation reward corresponding to the sample time step sequence; Based on the sample joint angles and the sample point cloud maps corresponding to each of the historical time lengths, as well as the sample terrain features and the sample accumulated rewards, the Transformer learning model is trained to predict the bucket tooth tip motion sequence until the output bucket tooth tip motion sequence prediction result meets the preset accuracy requirement, and the bucket tooth tip motion prediction model is determined.

3. The excavator bucket tooth tip trajectory planning method according to claim 1, characterized in that: The process of obtaining the target terrain features corresponding to the target excavation area includes: Obtaining an initial three-dimensional terrain point cloud map with a timestamp collected by a laser radar for the target excavation area; Preprocessing the initial three-dimensional terrain point cloud map; Performing raster processing on map difference data between the pre-processed three-dimensional terrain point cloud map and the target three-dimensional engineering map to obtain a grid map to be constructed in the target excavation area; Feature extraction is performed on the grid map to be constructed to obtain the target terrain features.

4. The excavator bucket tooth tip trajectory planning method according to claim 1, characterized in that: The method further comprises: Reacquiring post-operation terrain data of the target excavation area while the excavator is operating the target excavation area; The target bucket tooth tip trajectory planning result is adjusted based on the post-operation terrain data and the terrain change difference and potential excavation deviation determined by the preset operation plan.

5. An excavator bucket tooth tip trajectory planning device using the excavator bucket tooth tip trajectory planning method according to any one of claims 1 to 4, characterized in that: include: a data acquisition unit, configured to acquire a target joint angle of the excavator during operation in a target excavation area for a current length of time, a target point cloud map corresponding to the target joint angle, and target terrain features corresponding to the target excavation area; a motion prediction unit, configured to input the target joint angle, the target point cloud map, and the target terrain features into a bucket tooth tip motion prediction model, and obtain a target bucket tooth tip motion sequence prediction result output by the bucket tooth tip motion prediction model; A trajectory planning unit, configured to perform trajectory planning based on the target bucket tooth tip motion sequence prediction result, and determine the target bucket tooth tip trajectory planning result of the current time length; Among them, the bucket tooth tip motion prediction model is obtained by training the Transformer learning model to predict the bucket tooth tip motion sequence based on the sample joint angles of the sample excavator during operation in the sample excavation area for a historical period of time, the sample point cloud map corresponding to the sample joint angles, and the sample terrain features corresponding to the sample excavation area.

6. An excavator, characterized in that: It includes the excavator bucket tooth tip trajectory planning device as described in claim 5.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the excavator bucket tooth tip trajectory planning method according to any one of claims 1 to 4 are implemented.

8. A non-transitory 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 excavator bucket tooth tip trajectory planning method according to any one of claims 1 to 4 are implemented.

Images