A robot driving method, system, device and storage medium

Abstract

The application discloses a robot driving method, system, device and storage medium. Sequence language information is acquired, the sequence language information is input into a path generation network to obtain a target reference path, the path generation network is trained according to historical movement data of a robot and is a network for predicting a target reference path of the robot; a target speed matched with the target reference path is determined based on a current speed of the robot and an acceleration limit value of the robot, and the robot is driven by using the reference path and the target speed. That is, a part of the reference path is generated by using a pre-trained path generation network, a running speed most suitable for a current state of the robot is determined based on the upper and lower limit values of the current speed and the acceleration, and the robot is driven to move stably according to the determined speed and the route. Therefore, the robot can stably run without collision in a complex indoor environment with many obstacles, and the stability of the robot running is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a robot driving method, system, device, and storage medium. Background Technology

[0002] The ability to navigate within its environment is crucial for any mobile device. Firstly, it's essential to avoid dangerous situations such as collisions and unsafe conditions (temperature, radiation, weather exposure, etc.), but if the robot's objective relates to specific locations within the robotic environment, then finding those locations is necessary. Deep reinforcement learning has shown great potential for applications in robot navigation.

[0003] Existing deep reinforcement learning navigation methods aim to train end-to-end navigation strategies, directly mapping the robot's observations of the environment to its motion commands, such as linear and angular velocities. In these methods, the robot's navigation strategy only needs to respond to the current environment, such as navigation methods based on asynchronous advantage action evaluation algorithms, without considering smooth movement over a period of time. However, in real-world applications, in addition to avoiding collisions, the smoothness of robot navigation is also crucial.

[0004] Therefore, how to improve the smoothness of robot movement without collisions is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a robot driving method, system, device, and storage medium to improve the stability of the robot driving process.

[0006] To address the above problems, the technical solutions provided in this application are as follows:

[0007] The first aspect of this application provides a robot driving method, including:

[0008] Obtain sequence language information, which is used to describe environmental information acquired by the robot in the target environment within a preset time period;

[0009] The sequence language information is input into the path generation network to obtain the target reference path. The path generation network is trained based on the robot's historical movement data and is used to predict the robot's target reference path.

[0010] The target speed matching the target reference path is determined based on the robot's current speed and the robot's acceleration limit, wherein the acceleration limit is used to characterize the robot's maximum acceleration value and / or minimum acceleration value.

[0011] Drive the robot along the reference path and at the target speed.

[0012] Optionally, the method for determining the sequence language information includes:

[0013] The robot acquires environmental information observed by the robot, the relative coordinates of the target point with respect to the robot, and the reference path of the completed movement within a first preset time period, which is earlier than the current time.

[0014] The acquired content is encoded to obtain sequence language information.

[0015] Optionally, encoding the acquired content to obtain sequence language information includes:

[0016] The acquired content is input into the multilayer perceptron corresponding to each category to obtain sequence language information. The multilayer perceptron is used to represent the content of each category in a unified language.

[0017] Optionally, the reference path can be represented by an ordered list of several included angles, where each included angle is the angle between the line connecting two points at a preset distance in the target reference path and the line directly in front of the robot.

[0018] Optionally, determining the target speed matching the target reference path includes:

[0019] The velocity range for the next moment is calculated based on the robot's current speed and the robot's acceleration limit.

[0020] The velocity range is uniformly sampled to obtain several velocity samples;

[0021] Each speed sample is evaluated according to the preset evaluation indicators, and the speed sample that meets the preset speed requirements is selected as the target speed.

[0022] Optionally, the training process of the path generation network includes:

[0023] Obtain historical datasets, which include observation data of the robot during each run to the target point within a second preset time period, the relative coordinates of the target point relative to the robot, and the completed reference paths.

[0024] A preliminary path generation network is trained using historical datasets until the accuracy of the reference paths output by the trained network reaches a preset threshold, thereby generating the path generation network.

[0025] A second aspect of this application provides a robot drive system, comprising:

[0026] An acquisition unit is used to acquire sequence language information, which is used to describe environmental information acquired by the robot in the target environment within a preset time period.

[0027] The path determination unit is used to input the sequence language information into the path generation network to obtain the target reference path. The path generation network is trained based on the robot's historical movement data and is used to predict the robot's target reference path.

[0028] A velocity determination unit is used to determine a target velocity that matches the target reference path based on the robot's current velocity and the robot's acceleration limit, wherein the acceleration limit is used to characterize the robot's maximum acceleration value and / or minimum acceleration value.

[0029] A drive unit for driving the robot along the reference path and at the target speed.

[0030] Optionally, the speed determination unit includes:

[0031] The speed range calculation unit is used to calculate the speed range for the next moment based on the robot's current speed and the robot's acceleration limit.

[0032] A sampling unit is used to uniformly sample the velocity range to obtain several velocity samples;

[0033] The target speed determination unit is used to evaluate each speed sample according to preset evaluation indicators, and select the speed sample that meets the preset speed requirements as the target speed.

[0034] A third aspect of this application provides 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 computer program, it implements the robot driving method described in any one of the first aspects above.

[0035] A fourth aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a terminal device, cause the terminal device to perform the robot driving method as described in any of the preceding first aspects.

[0036] Compared with the prior art, this application has the following beneficial effects:

[0037] The process involves acquiring sequence language information, which describes environmental information obtained by the robot within a preset time period in a target environment. This sequence language information is then input into a path generation network to obtain a target reference path. This path generation network is trained based on the robot's historical movement data and is used to predict the robot's target reference path. A target speed matching the target reference path is determined based on the robot's current speed and its acceleration limit, where the acceleration limit characterizes the robot's maximum and / or minimum acceleration value. The robot is then driven using the reference path and the target speed. This approach uses a pre-trained path generation network to generate a partial reference path, rather than simply determining the running route based on environmental information and generating motion commands end-to-end as in existing technologies. The solution in this application determines the most suitable running speed for the robot's current state based on the upper and lower limits of the current speed and acceleration, and drives the robot to move smoothly according to the determined speed and route. Therefore, the solution provided in this application allows the robot to operate smoothly without collisions in complex indoor environments with many obstacles, improving the stability of robot operation. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings used in the description of the embodiment or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a flowchart of a robot driving method provided in an embodiment of this application;

[0040] Figure 2 This is an example diagram illustrating the simulation of a robot in an indoor scene provided in an embodiment of this application;

[0041] Figure 3 This is an example diagram illustrating the simulation of a robot in an indoor scene provided in an embodiment of this application;

[0042] Figure 4 This is a structural diagram of a robot drive system provided in an embodiment of this application. Detailed Implementation

[0043] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0044] To facilitate understanding of the technical solutions provided in the embodiments of this application, the background technology involved in the embodiments of this application will be described below.

[0045] As mentioned earlier, robots are increasingly used in daily life. Most existing deep reinforcement learning navigation methods aim to train end-to-end navigation strategies, directly mapping the robot's observations of the environment to its motion commands. That is, they generate only corresponding movement commands based on the environment to drive the robot, such as generating control commands containing linear and angular velocities to instruct the robot to move. In these methods, the navigation strategy only needs to respond to the current environment, such as navigation methods based on asynchronous advantage action evaluation algorithms, without considering smooth movement over a period of time.

[0046] To address this issue, embodiments of this application provide a robot driving method, system, device, and storage medium. The method involves acquiring sequence language information, which describes environmental information obtained by the robot within a preset time period in a target environment; inputting the sequence language information into a path generation network to obtain a target reference path, wherein the path generation network is trained based on the robot's historical movement data to predict the robot's target reference path; determining a target speed matching the target reference path based on the robot's current speed and its acceleration limit, wherein the acceleration limit characterizes the robot's maximum and / or minimum acceleration value; and driving the robot with the reference path and the target speed. This solution uses a pre-trained path generation network to generate a partial reference path, instead of simply determining the running route based on environmental information and generating motion commands end-to-end as in existing technologies. Based on this, the solution determines the most suitable running speed for the robot's current state. This running speed prevents the robot from experiencing abrupt changes during movement, and the robot moves smoothly according to the determined speed and route. Therefore, the solution provided by this application enables the robot to run smoothly without collisions in complex indoor environments with many obstacles, improving the stability of robot operation.

[0047] To facilitate understanding of the robot driving method provided in the embodiments of this application, the following scenario examples of this application are described.

[0048] The following embodiment illustrates a robot driving method provided in this application. See also... Figure 1 The figure is a flowchart of a robot driving method provided in an embodiment of this application. The execution subject of this method can be a server, or more specifically, a robot driving system in a cloud server. The method includes:

[0049] S101: Obtain sequence language information.

[0050] The sequence language information is used to describe the environmental information obtained by the robot in the target environment within a preset time period.

[0051] In one possible implementation, the determination of the sequence language information includes steps A1-A2:

[0052] A1: Obtain environmental information observed by the robot, the relative coordinates of the target point with respect to the robot, and the reference path of the completed movement within the first preset time period.

[0053] The first preset time period is earlier than the current time. The acquired content serves as the network input sequence, which includes the expected reward (RTG) that may be obtained in the future, the observed laser information, the relative coordinates of the target point, and a portion of the reference path from previous times.

[0054] A2: Encode the acquired content to obtain sequence language information.

[0055] In practical applications, encoding can be processed in different ways. In one possible implementation, the acquired content can be input into the multilayer perceptron corresponding to each category to obtain sequence language information. The multilayer perceptron is used to represent the content of each category in a unified language.

[0056] That is, for each different input information, a multilayer perceptron can be applied to encode the different information into a unified language representation, forming the network's input sequence language. These multilayer perceptron networks will be trained together with the main network to obtain the best language representation.

[0057] S102: Input the sequence language information into the path generation network to obtain the target reference path.

[0058] The path generation network is trained based on the robot's historical movement data and is used to predict the robot's target reference path.

[0059] The reference path is represented as an ordered list of angles, where each angle is the angle between the line connecting two points at a predetermined distance on the target reference path and the front of the robot. In practical applications, some reference paths are often represented as an ordered list of coordinates of several points on the path. However, this representation introduces significant complexity and makes network training difficult. Therefore, this application simplifies this part further. For example, the distance between points on the reference path can be fixed at 0.2 meters, and the angle between the line connecting two points and the front of the robot can be calculated sequentially. This ordered list of angles represents a portion of the reference path. This encoding method greatly simplifies the output complexity and improves output stability. The distance values ​​mentioned above are only illustrative; in practical applications, the distance values ​​can be adaptively adjusted according to actual needs and path conditions.

[0060] In one possible implementation, the training process of the path generation network includes steps B1-B2:

[0061] Step B1: Obtain historical datasets.

[0062] The historical dataset includes observation data of the robot during each run to the target point within the second preset time period, the relative coordinates of the target point relative to the robot, and the completed reference path.

[0063] In practical applications, open-source 2D simulators available on GitHub can be used to simulate robot operation in indoor environments and collect runtime logs to form a training dataset. See also... Figure 2 and Figure 3 , Figure 2 and Figure 3 These are all example images of robot simulation in indoor scenes provided in the embodiments of this application. Figure 2 and Figure 3 The small square in the diagram represents the robot, whose radius and shape can be adjusted according to the needs of the actual scene. The pentagram represents the robot's target point. The surrounding circles and rectangles represent static obstacles. The dots with varying shades of color in the diagram represent the paths the robot has already traversed. The trajectory marked between the pentagram and the small square represents the path the robot will move on, i.e., a partial reference path. During training, it is necessary to collect the observation data of the 180-degree 2D planar laser on the robot at each moment during each run to the target point, the relative coordinates of the target point relative to the robot, and the partial reference path as training data. Collecting several of the above contents forms a historical dataset. For example, 60,000 training data points can constitute a historical dataset. The number of data points here can be adjusted according to actual needs.

[0064] Step B2: Train the preliminary path generation network using historical datasets until the accuracy of the reference path output by the trained network reaches a preset threshold, and then generate the path generation network.

[0065] Using the dataset obtained above, a path generation network is prepared. The main body of this network consists of GPT-2. In actual training, the batch size can be 512, using an adaptive moment estimation optimizer with weight decay and cross-entropy as the loss function. A total of 30,000 batches of data were trained. It should be noted that the training batch size and the applied algorithm can be adjusted according to actual needs. In one possible implementation, the multilayer perceptron network mentioned in the previous steps can be trained together with this network to obtain the best language representation.

[0066] S103: Determine the target speed that matches the target reference path based on the robot's current speed and the robot's acceleration limit.

[0067] The acceleration limits are used to characterize the robot's maximum and / or minimum acceleration values. For determining the target velocity, a path-following algorithm can be used. This algorithm can be viewed as solving an optimization problem to find the control command that best fits the path the robot is following. The requirement for smooth movement adds an additional constraint to this problem.

[0068] In one possible implementation, determining the target speed matching the target reference path includes steps C1-C3:

[0069] Step C1: Calculate the speed range for the next moment based on the robot's current speed and the robot's acceleration limit.

[0070] The acceleration limits are used to characterize the robot's maximum acceleration and / or minimum acceleration, i.e., the upper and lower limits of acceleration. This takes into account both whether the moving speed exceeds the robot's maximum load and the current operating speed of the robot, preventing sudden and significant acceleration / deceleration and making the robot's movement more stable.

[0071] Step C2: Uniformly sample the velocity range to obtain several velocity samples.

[0072] Uniform sampling is performed within the velocity range to obtain several velocity samples. The sampling frequency and the number of samples can be adjusted according to actual needs and are not limited here.

[0073] Step C3: Evaluate each speed sample according to the preset evaluation index, and select the speed sample that meets the preset speed requirements as the target speed.

[0074] The preset evaluation criteria serve as screening conditions for whether speed samples meet the criteria. An algorithm can be applied to evaluate each sample, and these criteria can include the difference between the current speed and the speed at which the sample is being followed, the deviation from the target path caused by traveling at this speed, and so on. The sample that best meets the preset evaluation criteria is taken as the final output of the algorithm. Our proposed method explicitly considers the constraints of smooth movement and, through sampling, can quickly obtain optimal results.

[0075] S104: Drive the robot along the reference path and at the target speed.

[0076] Based on the reference path determined in the above steps and the evaluated target speed, drive the current robot to move smoothly.

[0077] The above embodiments propose a novel deep learning navigation framework, which uses deep reinforcement learning methods to generate a partial reference path, and then uses a path-following algorithm to drive the robot to move smoothly. Specific algorithms are proposed for both parts of the framework. This enables the robot to operate smoothly and efficiently in complex indoor environments with many obstacles.

[0078] The above are some specific implementations of the robot driving method provided in the embodiments of this application. Based on this, this application also provides a corresponding system for robot driving. The system provided in the embodiments of this application will be described below from the perspective of functional modularity. Figure 4 This is a structural diagram of a robot drive system provided in an embodiment of this application.

[0079] The system includes:

[0080] Acquisition unit 201 is used to acquire sequence language information, which is used to describe the environmental information acquired by the robot in the target environment within a preset time period;

[0081] The path determination unit 202 is used to input the sequence language information into the path generation network to obtain the target reference path. The path generation network is trained based on the robot's historical movement data and is used to predict the robot's target reference path.

[0082] The speed determination unit 203 is used to determine a target speed that matches the target reference path based on the robot's current speed and the robot's acceleration limit, wherein the acceleration limit is used to characterize the robot's maximum acceleration value and / or minimum acceleration value.

[0083] The drive unit 204 is used to drive the robot along the reference path and at the target speed.

[0084] Optionally, the speed determination unit includes:

[0085] The speed range calculation unit is used to calculate the speed range for the next moment based on the robot's current speed and the robot's acceleration limit.

[0086] A sampling unit is used to uniformly sample the velocity range to obtain several velocity samples;

[0087] The target speed determination unit is used to evaluate each speed sample according to preset evaluation indicators, and select the speed sample that meets the preset speed requirements as the target speed.

[0088] Optionally, the training process of the path generation network includes:

[0089] Obtain historical datasets, which include observation data of the robot during each run to the target point within a second preset time period, the relative coordinates of the target point relative to the robot, and the completed reference paths.

[0090] A preliminary path generation network is trained using historical datasets until the accuracy of the reference paths output by the trained network reaches a preset threshold, thereby generating the path generation network.

[0091] This application also provides corresponding devices and computer storage media for implementing the robot driving method provided in this application.

[0092] The device includes a memory and a processor. The memory is used to store instructions or code, and the processor is used to execute the instructions or code to cause the device to perform the robot driving method described in any embodiment of this application.

[0093] The computer storage medium stores code, and when the code is run, the device running the code implements the robot driving method described in any embodiment of this application.

[0094] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0095] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0096] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0097] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0098] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A robot driving method, characterized in that, include: Obtain sequence language information, which is used to describe environmental information acquired by the robot in the target environment within a preset time period; The method for acquiring the sequence language information includes: acquiring 180-degree 2D planar laser data observed by the robot, the expected reward RTG to be obtained in the future, the relative coordinates of the target point relative to the robot, and the reference path of the completed movement within a first preset time period; inputting the 180-degree 2D planar laser data, the expected reward RTG, the relative coordinates, and the reference path of the completed movement into a multilayer perceptron corresponding to each type for encoding to obtain a unified language representation, forming the sequence language information, wherein the multilayer perceptron is used to represent the content of each type in a unified language; The sequence language information is input into the path generation network to obtain the target reference path. The path generation network is trained based on the robot's historical movement data. It is trained on the GPT-2 architecture and uses an adaptive moment estimation optimizer with weight decay and a cross-entropy loss function to predict the robot's target reference path. The target reference path is represented by an ordered list of the angles between the lines connecting continuous points at fixed distances on the path and the front of the robot. A path-following algorithm is employed, based on the robot's current speed, acceleration limit, and preset evaluation indicators. A target speed matching the target reference path and meeting preset speed requirements is determined through sampling. The determination process includes: calculating the speed range for the next moment based on the robot's current speed and acceleration limit; uniformly sampling the speed range to obtain several speed samples; evaluating each speed sample according to the preset evaluation indicators; and selecting the speed sample that best meets the preset evaluation indicators as the target speed. The acceleration limit is used to characterize the robot's maximum and / or minimum acceleration values. The preset evaluation indicators include the difference between the speed sample and the current speed, and the offset from the target reference path when traveling according to the speed sample, to explicitly consider smooth movement constraints. Drive the robot along the reference path and at the target speed; The multiple multilayer perceptrons and the path generation network are obtained through joint training. The joint training enables the multilayer perceptrons to encode various types of input information into a unified language representation. The target reference path is a partial reference path used for local path following.

2. The method according to claim 1, characterized in that, The methods for determining the sequence language information include: The robot acquires environmental information observed by the robot, the relative coordinates of the target point with respect to the robot, and the reference path of the completed movement within a first preset time period, which is earlier than the current time. The acquired content is encoded to obtain sequence language information.

3. The method according to claim 2, characterized in that, The process of encoding the acquired content to obtain sequence language information includes: The acquired content is input into the multilayer perceptron corresponding to each category to obtain sequence language information. The multilayer perceptron is used to represent the content of each category in a unified language.

4. The method according to claim 1, characterized in that, The reference path is represented by an ordered list of several included angles, where each included angle is the angle between the line connecting two points at a preset distance in the target reference path and the line directly in front of the robot.

5. The method according to claim 1, characterized in that, Determining the target speed that matches the target reference path includes: The velocity range for the next moment is calculated based on the robot's current speed and the robot's acceleration limit. The velocity range is uniformly sampled to obtain several velocity samples; Each speed sample is evaluated according to the preset evaluation indicators, and the speed sample that meets the preset speed requirements is selected as the target speed.

6. The method according to claim 1, characterized in that, The training process of the path generation network includes: Obtain historical datasets, which include observation data of the robot during each run to the target point within a second preset time period, the relative coordinates of the target point relative to the robot, and the completed reference paths. A preliminary path generation network is trained using historical datasets until the accuracy of the reference paths output by the trained network reaches a preset threshold, thereby generating the path generation network.

7. A robot drive system, characterized in that, The system includes: An acquisition unit is used to acquire sequence language information, which describes environmental information acquired by the robot in the target environment within a preset time period. The acquisition method of the sequence language information includes: acquiring 180-degree 2D planar laser data observed by the robot, the expected reward RTG to be obtained in the future, the relative coordinates of the target point relative to the robot, and the reference path of the completed movement within the first preset time period; inputting the 180-degree 2D planar laser data, the expected reward RTG, the relative coordinates, and the reference path of the completed movement into a multilayer perceptron corresponding to each type for encoding to obtain a unified language representation, forming the sequence language information; the multilayer perceptron is used to represent the content of each type in a unified language. The path determination unit is used to input the sequence language information into the path generation network to obtain the target reference path. The path generation network is trained based on the robot's historical movement data, and is trained using the GPT-2 architecture and an adaptive moment estimation optimizer with weight decay and a cross-entropy loss function. It is used to predict the robot's target reference path. The target reference path is represented by an ordered list of the angles between the lines connecting continuous points at fixed distances on the path and the front of the robot. The speed determination unit is used to determine a target speed that matches the target reference path and meets the preset speed requirements by using a path following algorithm, based on the robot's current speed, the robot's acceleration limit, and preset evaluation indicators, through sampling. The determination process includes: calculating the speed range for the next moment based on the robot's current speed and the robot's acceleration limit; uniformly sampling the speed range to obtain several speed samples; evaluating each speed sample according to the preset evaluation indicators; and selecting the speed sample that best meets the preset evaluation indicators as the target speed. The acceleration limit is used to characterize the robot's maximum and / or minimum acceleration values. The preset evaluation indicators include the difference between the speed sample and the current speed, and the offset from the target reference path when traveling according to the speed sample, so as to explicitly consider smooth movement constraints. A drive unit is configured to drive the robot along the reference path and at the target speed; The multiple multilayer perceptrons and the path generation network are obtained through joint training. The joint training enables the multilayer perceptrons to encode various types of input information into a unified language representation. The target reference path is a partial reference path used for local path following.

8. The system according to claim 7, characterized in that, The speed determination unit includes: The speed range calculation unit is used to calculate the speed range for the next moment based on the robot's current speed and the robot's acceleration limit. A sampling unit is used to uniformly sample the velocity range to obtain several velocity samples; The target speed determination unit is used to evaluate each speed sample according to preset evaluation indicators, and select the speed sample that meets the preset speed requirements as the target speed.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the robot driving method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a terminal device, cause the terminal device to perform the robot driving method as described in any one of claims 1-6.

Images