Physical constraint fused robot trajectory planning method, device, equipment, medium and product

Abstract

The invention discloses a robot trajectory planning method and device fusing physical constraints, equipment, a medium and a product, and relates to the technical field of trajectory planning, and the method comprises the steps: obtaining a current trajectory state quantity sequence of a robot; constructing an LTSM-RDNN network based on a long short-term memory network and a cyclic deep neural network, training the LTSM-RDNN network by using the optimal trajectory data set, and introducing trajectory state quantity constraint loss and controlled quantity constraint loss by using a PINN theory in the training process of the LTSM-RDNN network to obtain a trajectory prediction model based on the PINN network; based on the trajectory state quantity sequence, determining the optimal control quantity of the robot at the current time point by adopting a trajectory prediction model based on a PINN network; and the robot is controlled according to the optimal control quantity at the current time point, and real-time trajectory planning of the robot is achieved. Real-time trajectory planning can be carried out according to the physical constraint of the robot in the actual scene, and the feasibility of the trajectory is improved.

Description

Technical Field

[0001] The present application relates to the technical field of trajectory planning, and particularly to a robot trajectory planning method, device, equipment, medium and product that integrates physical constraints. Background Art

[0002] Currently, the academic and industrial communities adopt trajectory planning methods based on deep neural networks (DNNs). Deep neural networks have shown great potential in improving computational efficiency and real-time performance, especially achieving initial success in complex task scenarios such as hypersonic vehicle reentry, spacecraft celestial body landing, and autonomous parking of unmanned vehicles. Through data-driven methods, DNNs can quickly learn trajectory datasets related to tasks and generate control signals through real-time inference to achieve a rapid response to trajectory planning. The core idea of this method is to generate a set of trajectory datasets that meet task objectives through trajectory optimization methods and train neural networks on these trajectory datasets so that they can learn the non-linear mapping relationship between system states and optimal controls. However, despite the advantages of DNNs in computational efficiency and real-time performance, relying solely on data-driven training methods still has certain limitations. The DNN model only learns the numerical mapping relationship in the data and ignores many physical constraints, such as upper and lower limit constraints on state variables and control variables. This may cause the trained model to not meet the physical constraint requirements in actual applications, thereby affecting the feasibility of the trajectory. Summary of the Invention

[0003] The purpose of the present application is to provide a robot trajectory planning method, device, equipment, medium and product that integrates physical constraints, which can perform trajectory planning according to the physical constraints of the robot in the actual scenario and improve the feasibility of the trajectory.

[0004] To achieve the above purpose, the present application provides the following solutions:

[0005] In the first aspect, the present application provides a robot real-time trajectory planning method that integrates physical constraints, including:

[0006] Obtain the current trajectory state quantity sequence of the robot; wherein, the trajectory state quantity sequence is composed of trajectory state quantities corresponding to a continuous plurality of time points including the current time point; the trajectory state quantity includes the position, speed, steering angle, and azimuth angle of the robot;

[0007] Construct an optimal trajectory dataset according to the end point and start point of the target scenario;

[0008] Construct an LTSM-RDNN network based on long short-term memory networks and recurrent deep neural networks, train the LTSM-RDNN network using an optimal trajectory dataset, and introduce a trajectory state quantity constraint loss and a control quantity constraint loss based on the PINN theory during the training process of the LTSM-RDNN network to obtain a trajectory prediction model based on the PINN network;

[0009] Based on the trajectory state quantity sequence, use the trajectory prediction model based on the PINN network to determine the optimal control quantity of the robot at the current time point; wherein, the optimal control quantity includes acceleration and angular velocity;

[0010] Control the robot according to the optimal control quantity at the current time point to achieve real-time trajectory planning of the robot.

[0011] In a second aspect, the present application provides a robot real-time trajectory planning device integrating physical constraints, including:

[0012] An acquisition module, configured to acquire the current trajectory state quantity sequence of the robot; wherein, the trajectory state quantity sequence is composed of trajectory state quantities corresponding to a continuous plurality of time points including the current time point; the trajectory state quantity includes the position, speed, steering angle, and azimuth angle of the robot;

[0013] A dataset construction module, configured to construct an optimal trajectory dataset according to the end point and the start point in the target scenario;

[0014] A training module, configured to construct an LTSM-RDNN network based on long short-term memory networks and recurrent deep neural networks, train the LTSM-RDNN network using the optimal trajectory dataset, and introduce a trajectory state quantity constraint loss and a control quantity constraint loss based on the PINN theory during the training process of the LTSM-RDNN network to obtain a trajectory prediction model based on the PINN network;

[0015] A trajectory online planning module, configured to, based on the trajectory state quantity sequence, use the trajectory prediction model based on the PINN network to determine the optimal control quantity of the robot at the current time point; wherein, the optimal control quantity includes acceleration and angular velocity;

[0016] A control module, configured to control the robot according to the optimal control quantity at the current time point to achieve real-time trajectory planning of the robot.

[0017] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned robot trajectory planning method integrating physical constraints.

[0018] Fourthly, the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned robot trajectory planning method integrating physical constraints is realized.

[0019] Fifthly, the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, the above-mentioned robot trajectory planning method integrating physical constraints is realized.

[0020] According to the specific embodiments provided by the present application, the following technical effects are disclosed:

[0021] The present application provides a robot trajectory planning method, device, equipment, medium and product integrating physical constraints. By constructing an optimal trajectory data set according to the end point and start point of the target scene, and using the optimal trajectory data set to train the LTSM-RDNN network, and introducing two physical constraints, namely trajectory state quantity constraint and control quantity constraint, into the training process for optimization based on the PINN theory, the trained trajectory prediction model based on the PINN network can perform trajectory planning in the target scene by combining the physical constraints in the actual scene, so that the planned trajectory meets the requirements of the robot in the actual scene and improves the feasibility of the trajectory. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0023] Figure 1 It is an application environment diagram of a robot trajectory planning method integrating physical constraints in an embodiment of the present application;

[0024] Figure 2 It is a schematic flow chart of a robot trajectory planning method integrating physical constraints provided in an embodiment of the present application;

[0025] Figure 3 It is a schematic diagram of the principle of real-time trajectory planning of a robot provided in an embodiment of the present application;

[0026] Figure 4 It is a schematic diagram of the optimal trajectory solving process provided in an embodiment of the present application;

[0027] Figure 5 It is a structural diagram of an LTSM unit provided in an embodiment of the present application;

[0028] Figure 6Schematic diagram of the training process of the LTSM-RDNN network provided by an embodiment of the present application;

[0029] Figure 7 Schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0030] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0031] To make the objectives, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0032] The real-time trajectory planning method for a robot integrating physical constraints provided by the embodiments of the present application can be applied to, for example Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set separately, integrated on the server 104, placed on the cloud or other servers. The terminal 102 can send the current trajectory state quantity sequence of the robot to the server 104. After receiving the current trajectory state quantity sequence of the robot, for the current trajectory state quantity sequence of the robot, the server 104 obtains the current trajectory state quantity sequence of the robot; constructs an optimal trajectory data set according to the end point and start point of the target scenario; constructs an LTSM-RDNN network based on the long short-term memory network and the recurrent deep neural network, trains the LTSM-RDNN network with the optimal trajectory data set, and introduces a trajectory state quantity constraint loss and a control quantity constraint loss based on the PINN theory during the training of the LTSM-RDNN network to obtain a trajectory prediction model based on the PINN network; based on the trajectory state quantity sequence, determines the optimal control quantity at the current time point of the robot by using the trajectory prediction model based on the PINN network; controls the robot according to the optimal control quantity at the current time point to realize the real-time trajectory planning of the robot. The server 104 can feedback the optimal control quantity at the current time point of the robot to the terminal 102. In addition, in some embodiments, the method for real-time trajectory planning of a robot integrating physical constraints can also be implemented separately by the server 104 or the terminal 102. For example, the terminal 102 can directly process the current trajectory state quantity sequence of the robot, or the server 104 can obtain the current trajectory state quantity sequence of the robot from the data storage system and process the current trajectory state quantity sequence of the robot.

[0033] Among them, the terminal 102 can be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0034] In an exemplary embodiment, as Figure 2 shown, a method for real-time trajectory planning of a robot integrating physical constraints is provided. This method is executed by a computer device, and can be specifically executed separately by a computer device such as a terminal or a server, or jointly executed by a terminal and a server. In the embodiments of the present application, taking this method applied to Figure 1 the server 104 in it as an example for illustration, it includes the following steps 201 to step 205. Among them:

[0035] Step 201, obtain the current trajectory state quantity sequence of the robot; wherein, the trajectory state quantity sequence is composed of trajectory state quantities corresponding to a continuous plurality of time points including the current time point; the trajectory state quantity includes the position, speed, steering angle and azimuth angle of the robot.

[0036] Specifically, the position of the robot can be represented by the abscissa and ordinate.

[0037] Step 202, construct an optimal trajectory data set according to the end point and start point of the target scene.

[0038] Step 203, construct an LTSM-RDNN network based on the long short-term memory network and the recurrent deep neural network, train the LTSM-RDNN network using the optimal trajectory data set, and introduce the trajectory state quantity constraint loss and control quantity constraint loss based on the PINN theory during the training process of the LTSM-RDNN network to obtain a trajectory prediction model based on the PINN network.

[0039] Step 204, based on the trajectory state quantity sequence, use the trajectory prediction model based on the PINN network to determine the optimal control quantity of the robot at the current time point; wherein, the optimal control quantity includes the acceleration and angular velocity.

[0040] As Figure 3 shown, the obtained current trajectory state quantity sequence of the robot is the trajectory state quantity sequence from the 1st time point to the Tth time point. Based on this, use the trajectory prediction model based on the PINN network to predict the optimal control quantity at the (T + 1)th time point.

[0041] Step 205, control the robot according to the optimal control quantity at the current time point to achieve real-time trajectory planning of the robot.

[0042] Input the optimal control quantity at the current time into the robot to obtain the trajectory state quantity at the next time point. Based on the trajectory state quantity at the next time point, update the current trajectory state quantity sequence to obtain a trajectory state quantity sequence including the trajectory state quantity at the next time point. Based on this, use the trajectory prediction model based on the PINN network again to determine the optimal control quantity of the robot at the next time point, and repeat the process to complete the trajectory planning of the robot from the start point to the end point.

[0043] Implementing the above steps 201 to 205 can achieve real-time trajectory planning of the robot in a specific scenario, making the planned trajectory more feasible.

[0044] In an exemplary embodiment, step 202 specifically includes steps 301 to 307:

[0045] Step 301: Determine the target neighborhood based on the starting point; where the target neighborhood includes multiple position points.

[0046] For the starting point in the target scenario, determine the target neighborhood. There are multiple position points in the target neighborhood, and use these multiple position points to perturb the starting point.

[0047] Step 302: Determine the kinematic constraint and the collision avoidance constraint according to the ending point and the starting point.

[0048] The expression of the kinematic constraint f kinematics is as follows:

[0049]

[0050] where is the derivative information of the abscissa of the robot's position at time τ, is the derivative information of the ordinate of the robot's position at time τ, v(τ) is the speed of the robot at time τ, is the derivative information of the speed of the robot at time τ, φ(τ) is the front wheel steering angle of the robot at time τ, is the derivative information of the front wheel steering angle of the robot at time τ, is the derivative information of the heading angle of the robot at time τ, L B is the wheelbase of the robot, ω(τ) is the angular velocity of the robot at time τ, and a(τ) is the acceleration of the robot at time τ.

[0051] The expression of the collision avoidance constraint f collision-avoidance is as follows:

[0052] d(A1,A2)>d0;

[0053] where d(·) represents the distance between two points, A1 and A2 are the geometric center coordinates of the robot and the obstacle respectively, and d0 is the set distance threshold.

[0054] Step 303: Construct a kinematic constraint penalty function according to the kinematic constraint, and construct a collision avoidance constraint penalty function according to the collision avoidance constraint.

[0055] The kinematic constraint and the collision avoidance constraint, as non - linear constraints, often lead to a slow solution process. Therefore, in this embodiment, the kinematic constraint and the collision avoidance constraint are respectively abstracted as f kinematics and f collision-avoidance , and they are softened into external penalty functions, namely the kinematic constraint penalty function and the collision avoidance constraint penalty function.

[0056] The expression of the kinematic constraint penalty function is:

[0057]

[0058] The expression of the collision avoidance constraint penalty function is as follows:

[0059]

[0060] where P1 is the kinematic constraint penalty function and P2 is the collision avoidance constraint penalty function; is the derivative information of the trajectory state variable at time point τ, τ is the integration time, and f kinematics is the kinematic constraint, and f collision-avoidance is the collision avoidance constraint, and T τ is the total time from the starting point to the ending point of the target scenario.

[0061] Step 304: With the goal of minimizing the total time from the starting point to the ending point, construct an objective function based on the kinematic constraint penalty function and the collision avoidance constraint penalty function.

[0062]

[0063] where J new is the objective function, ω p is the weight coefficient, i = 1 or 2. When i = 1, P1 is the kinematic constraint penalty function, and when i = 2, P2 is the collision avoidance constraint penalty function.

[0064] Step 305: Use the interior point method to solve the objective function to obtain the task trajectory of the target scenario corresponding to multiple position points.

[0065] Step 306: Determine the optimal trajectory according to the comparison result between the infeasibility degree of the task trajectory and the infeasibility degree threshold.

[0066] The interior point method adopted in this embodiment is a numerical optimizer. The kinematic constraint penalty function and the collision avoidance constraint penalty function are integrated into the objective function, making the entire solution of the optimization problem simplified to a problem only containing box constraints. Since the constraints become the simplest linear form at this time, the solution process will become very fast, thus ensuring the efficiency of data generation.

[0067] Based on multiple position points in the target neighborhood, solve the optimal trajectory from each position point to the ending point. Specifically, taking a position point in the target neighborhood as an example, solve the optimal trajectory from this position point to the ending point. As Figure 4 shown, use the iter() function to create an iterator object and initialize it iter = 0. Figure 4 in is the infeasibility degree, that is The calculation formula of the infeasibility degree is as follows:

[0068]

[0069] When the threshold of infeasibility is reached, P1 - P2 also tends to 0, indicating that during the iterative solution process, the optimization problem can satisfy all softened non - linear constraints. In Figure 3 , the threshold of infeasibility is denoted as In addition, with a sufficient number of iterations, a better solution will be obtained in each iteration. When the infeasibility of the task trajectory is less than the threshold of infeasibility, the current task trajectory is the optimal solution from this position point to the end point, that is, the optimal trajectory from this position point to the end point.

[0070] Step 307, construct the optimal trajectory dataset by using the optimal trajectories corresponding to multiple position points.

[0071] In an exemplary embodiment, as Figure 5 and Figure 6 shown, in the LSTM - RDNN network, using the optimal control sequence in as the network input, predict the optimal control quantity at the current time point, that is:

[0072]

[0073] where, is the trajectory state quantity at the k - th time point, is the optimal control quantity at the k - th time point, Ε u is a non - linear function representing the input - output relationship of the LSTM - RDNN network.

[0074] The LTSM - RDNN network has N L layers, and each layer has N n neurons; at time t k , the output of the g - th neuron in the j - th layer can be written as:

[0075]

[0076] where, j = {1,..., N L}, g = {1,..., N n}, I (·) 、F (·) 、O (·) are the input gate, forget gate, and output gate respectively, are the candidate state, cell state, and hidden state respectively, is at time t kThe input of the g-th neuron in the j-th layer at time point and are respectively the input gate output, forget gate output, output gate output, cell state, candidate state, and hidden state of the g-th neuron in the j-th layer at time t k . and are respectively the cell state and hidden state of the g-th neuron in the j-th layer at time t k-1 . σ is the sigmoid function, tanh is the hyperbolic tangent function, and W i , W f , W o and W c are respectively the input gate weight matrix, forget gate weight matrix, output gate weight matrix, and candidate state weight matrix. b i , b f , b o and b c are respectively the input gate bias vector, forget gate bias vector, output gate bias vector, and candidate state bias vector.

[0077] Define a performance metric to train the LTSM-RDNN network and update the weight matrix and bias vector. In this embodiment, the mean square error is used to measure the approximation ability of the LTSM-RDNN network, which is defined as the basic error loss, and the expression is:

[0078]

[0079] where N tr is the total number of samples in the optimal trajectory dataset, η = t f / N t , t f is the task duration of the optimal trajectory, N t is the number of time points, N i is the N i -th sample in the optimal trajectory dataset, o(t kη ) and u * (t kη ) are respectively the predicted optimal control quantity and the optimal control quantity in the optimal trajectory dataset at time point t kη .

[0080] In addition, in order to add the two physical constraints of the trajectory state quantity constraint and the control quantity constraint to the loss term, it is necessary to perform ReLU regularization on the trajectory state quantity constraint and the control quantity constraint, which are respectively the control quantity constraint loss and the trajectory state quantity constraint loss, as follows:

[0081] The control quantity constraint loss is:

[0082]

[0083] The trajectory state quantity constraint loss is as follows:

[0084]

[0085] where u z is the z-th optimal control quantity, u max,z and u min,z are the upper and lower limits corresponding to u z respectively, x s is the s-th trajectory state quantity, x max,s and x min,s are the upper and lower limits corresponding to x s respectively, m and n are the total numbers of the control quantity and the trajectory state quantity respectively, ReLU is regularization, in this embodiment, the value of m is 5 and the value of n is 2.

[0086] Therefore, the final loss function can be described as the weighted sum of the basic error loss, the control quantity constraint loss, and the trajectory state quantity constraint loss, and the expression is:

[0087] L total = λ1L1 + λ2L control + λ3L state ;

[0088] where λ1, λ2, and λ3 are the weighted sum coefficients.

[0089] There are different loss terms in the total loss function, and the orders of magnitude of different loss terms may be inconsistent. To avoid that smaller loss terms may be ignored and larger loss terms will dominate the optimization process, making it difficult for the trajectory prediction model based on the PINN network to simultaneously satisfy all physical constraints and affecting the overall performance of the trajectory prediction model based on the PINN network, this embodiment uses a multi-loss term scale adaptive balance mechanism to achieve balanced optimization of different loss terms during training, ensuring that physical constraints and other loss terms can act effectively at the same time, and the final loss function in the above formula is transformed into:

[0090]

[0091] where L′ total is the total loss, ReLU is regularization, ω1, ω2, and ω3 are the loss weight coefficients.

[0092] By simply adjusting p or q, different physical constraint loss terms can be adjusted to roughly the same order of magnitude, which allows the optimization algorithm to optimize both simultaneously instead of favoring one of them. During the training process, according to the actual order of magnitude, by adjusting p or q, the optimization of the corresponding loss term can be suppressed or accelerated, so as to achieve the desired effect. For training the LTSM-RDNN network, this embodiment plans to use the stochastic gradient descent algorithm combined with the adaptive learning rate technique to train the LTSM-RDNN network, and finally obtain a trajectory prediction model based on the PINN network.

[0093] Based on the same inventive concept, an embodiment of the present application also provides a real-time trajectory planning device for a robot that incorporates physical constraints for implementing the above-mentioned real-time trajectory planning method for a robot that incorporates physical constraints. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the real-time trajectory planning device for a robot that incorporates physical constraints provided below can refer to the limitations on the real-time trajectory planning method for a robot that incorporates physical constraints in the above text, and will not be repeated here.

[0094] In an exemplary embodiment, a real-time trajectory planning device for a robot that incorporates physical constraints is provided, including:

[0095] An acquisition module, configured to acquire the current trajectory state quantity sequence of the robot; wherein, the trajectory state quantity sequence is composed of trajectory state quantities corresponding to a continuous plurality of time points including the current time point; the trajectory state quantity includes the position, speed, steering angle, and azimuth angle of the robot.

[0096] A data set construction module, configured to construct an optimal trajectory data set according to the end point and the start point in the target scenario;

[0097] A training module, configured to construct an LTSM-RDNN network based on a long short-term memory network and a recurrent deep neural network, train the LTSM-RDNN network using the optimal trajectory data set, and introduce a trajectory state quantity constraint loss and a control quantity constraint loss based on the PINN theory during the training process of the LTSM-RDNN network to obtain a trajectory prediction model based on the PINN network.

[0098] A trajectory online planning module, configured to determine the optimal control quantity of the robot at the current time point based on the trajectory state quantity sequence by using a trajectory prediction model based on the PINN network; wherein, the optimal control quantity includes acceleration and angular velocity.

[0099] A control module, configured to control the robot according to the optimal control quantity at the current time point to implement the real-time trajectory planning of the robot.

[0100] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as shown in Figure 7 . The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the current trajectory state quantity sequence of the robot. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a real-time trajectory planning method for a robot that integrates physical constraints.

[0101] Those skilled in the art can understand that Figure 7 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0102] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0103] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0104] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0105] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0106] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random-access memories (ReRAM), magnetoresistive random-access memories (MRAM), ferroelectric random-access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0107] The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0108] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0109] In this article, specific examples are used to elaborate on the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A real-time trajectory planning method for a robot integrating physical constraints, characterized in that, The robot trajectory planning method integrating physical constraints includes: Obtain the current trajectory state quantity sequence of the robot; wherein, the trajectory state quantity sequence is composed of trajectory state quantities corresponding to a continuous plurality of time points including the current time point; the trajectory state quantity includes the position, speed, steering angle, and azimuth angle of the robot; Construct an optimal trajectory data set based on the end point and start point of the target scenario; Construct an LTSM-RDNN network based on a long short-term memory network and a recurrent deep neural network, train the LTSM-RDNN network using the optimal trajectory data set, and introduce a trajectory state quantity constraint loss and a control quantity constraint loss based on the PINN theory during the training process of the LTSM-RDNN network to obtain a trajectory prediction model based on the PINN network; Based on the trajectory state quantity sequence, use the trajectory prediction model based on the PINN network to determine the optimal control quantity of the robot at the current time point; wherein, the optimal control quantity includes acceleration and angular velocity; Control the robot according to the optimal control quantity at the current time point to achieve real-time trajectory planning of the robot.

2. The real-time trajectory planning method for a robot integrating physical constraints according to claim 1, wherein Constructing an optimal trajectory data set using the end point and start point of the target scenario includes: Determine a target neighborhood based on the start point; wherein, the target neighborhood includes a plurality of position points; Determine kinematic constraints and collision avoidance constraints according to the end point and start point; Construct a kinematic constraint penalty function according to the kinematic constraints, and construct a collision avoidance constraint penalty function according to the collision avoidance constraints; Taking the shortest total time from the start point to the end point as the goal, construct an objective function based on the kinematic constraint penalty function and the collision avoidance constraint penalty function; Solve the objective function using the interior point method to obtain the task trajectory of the target scenario corresponding to a plurality of position points; Determine the optimal trajectory according to the comparison result between the infeasibility degree of the task trajectory and the infeasibility degree threshold; Construct the optimal trajectory data set using the optimal trajectories corresponding to a plurality of position points.

3. The real-time trajectory planning method for a robot integrating physical constraints according to claim 2, wherein The expression of the kinematic constraint penalty function is: The expression of the collision avoidance constraint penalty function is: Among them, P1 is the kinematic constraint penalty function, and P2 is the collision avoidance constraint penalty function; is the derivative information of the trajectory state variable at time τ, τ is the integration time, f kinematics is the kinematic constraint, f collision-avoidance is the collision avoidance constraint, T τ is the total time from the starting point to the ending point of the target scenario.

4. The real-time trajectory planning method for a robot integrating physical constraints according to claim 3, wherein The expression of the objective function is: Among them, J new is the objective function, ω p is the weight coefficient, i = 1 or 2. When i = 1, P1 is the kinematic constraint penalty function, and when i = 2, P2 is the collision avoidance constraint penalty function.

5. The real-time trajectory planning method for a robot integrating physical constraints according to claim 4, characterized in that The calculation formula of the infeasibility degree is: Among them, is the infeasibility degree.

6. The real-time trajectory planning method for a robot integrating physical constraints according to claim 1, wherein The total loss adopted in the training process of the LTSM-RDNN network includes a basic error loss, a trajectory state quantity constraint loss, and a control quantity constraint loss; The basic error loss is: The control quantity constraint loss is: The trajectory state quantity constraint loss is: The total loss is: Among them, L′ total is the total loss, L1 is the basic error term, L control is the control quantity constraint loss, L state is the trajectory state quantity constraint loss, p and q are variable parameters, N tr is the total number of samples in the optimal trajectory dataset, N i is the N i th sample in the optimal trajectory dataset, η = t f / N t where t f is the task duration of the optimal trajectory, N t is the number of time points, o(t kη ) and u * (t kη ) are the predicted optimal control quantity at the t kη th time point and the optimal control quantity in the optimal trajectory dataset respectively, u z is the zth optimal control quantity, u max,z and u min,z are the upper and lower limits corresponding to u z respectively, x s is the sth trajectory state quantity, x max,s and x min,s are the upper and lower limits corresponding to x s respectively, m and n are the total numbers of the trajectory state quantity and the control quantity respectively, ReLU is regularization, and ω1, ω2 and ω3 are loss weight coefficients.

7. A real-time trajectory planning device for a robot integrating physical constraints, characterized in that, The robot trajectory planning device integrating physical constraints includes: An acquisition module for acquiring the current trajectory state quantity sequence of the robot; wherein, the trajectory state quantity sequence is composed of trajectory state quantities corresponding to a continuous plurality of time points including the current time point; the trajectory state quantity includes the position, speed, steering angle, and azimuth angle of the robot; A data set construction module for constructing an optimal trajectory data set according to the end point and start point under the target scenario; A training module, which is used to construct an LTSM-RDNN network based on a long short-term memory network and a recurrent deep neural network, train the LTSM-RDNN network using an optimal trajectory dataset, and introduce a trajectory state quantity constraint loss and a control quantity constraint loss based on the PINN theory during the training process of the LTSM-RDNN network to obtain a trajectory prediction model based on the PINN network; A trajectory online planning module, which is used to determine the optimal control quantity of the robot at the current time point by using the trajectory prediction model based on the PINN network based on the trajectory state quantity sequence; wherein, the optimal control quantity includes acceleration and angular velocity; A control module, which is used to control the robot according to the optimal control quantity at the current time point to achieve real-time trajectory planning of the robot.

8. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the robot trajectory planning method with fused physical constraints according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the robot trajectory planning method with fused physical constraints according to any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the robot trajectory planning method with fused physical constraints according to any one of claims 1-6.

Images