Motion control method, device and equipment of foot type robot and medium

Abstract

The invention provides a motion control method and device of a foot type robot, equipment and a medium. The method comprises the steps that state information, coding information and historical actions of the foot type robot at the previous moment and first sensing information at the current moment are obtained; inputting the state information of the previous moment, the coding information and the historical actions into a world model to obtain the state information of the current moment output by the world model; inputting the state information at the current moment and the first sensing information into a motion strategy model to obtain a target action at the current moment output by the motion strategy model; and controlling the movement of the foot type robot according to the target action at the current moment. Any terrain environment can be accurately sensed, so that the stability of the foot type robot in the movement process is ensured, the problems of falling or movement faults and the like are avoided, and the movement control effect of the foot type robot is improved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of robots, and in particular to a motion control method, device, equipment and medium for a legged robot. BACKGROUND

[0002] A legged robot is a robot that can simulate the movement mode of human or animal feet, and can be applied to different fields such as industrial inspection, express delivery and home service, so as to replace humans to enter dangerous or difficult-to-reach areas to perform tasks, reduce personnel risks and improve work efficiency.

[0003] Generally, a legged robot needs to perceive the surrounding environment, and then control the motion according to the perceived surrounding environment. In related technologies, a legged robot perceives the surrounding environment through data obtained by different sensors in a built-in sensor system of the legged robot, so as to control the motion of the legged robot. For example, the legged robot obtains parameters such as pose, angular velocity, acceleration and gravity projection of the legged robot through an inertial measurement unit (IMU), and obtains parameters such as angle and angular velocity of each joint of the legged robot through a joint encoder, and so on. The built-in sensor system mentioned above refers to a system without sensors such as vision and laser radar.

[0004] The above motion control method can effectively cope with common terrains such as slopes and flat ground, but for complex terrain environments such as narrow passages, deep pits or deep holes, the legged robot cannot accurately perceive the above complex terrains, resulting in errors in motion control, causing the legged robot to fall down or even motion failure, and affecting the motion control effect of the legged robot. SUMMARY

[0005] Embodiments of the present application provide a motion control method, device, equipment and medium for a legged robot, which can accurately perceive any terrain environment, thereby ensuring the stability of the legged robot during motion, avoiding problems such as falling down or motion failure, and improving the motion control effect of the legged robot.

[0006] In a first aspect, embodiments of the present application provide a motion control method for a legged robot, comprising:

[0007] obtaining state information, encoding information and historical actions of the legged robot at a previous time, and first perception information at a current time;

[0008] inputting the state information, the encoding information and the historical actions at the previous time into a world model to obtain state information at the current time output by the world model;

[0009] input the state information and the first perception information of the current moment into a motion strategy model to obtain a target action of the current moment output by the motion strategy model;

[0010] control motion of the legged robot according to the target action of the current moment.

[0011] In a second aspect, an embodiment of the present application provides a motion control device of a legged robot, comprising:

[0012] an information acquisition module configured to acquire state information, encoding information and historical actions of the legged robot at a previous moment, and first perception information of the legged robot at a current moment;

[0013] an information determination module configured to input the state information, the encoding information and the historical actions of the previous moment into a world model to obtain state information of the current moment output by the world model;

[0014] an action determination module configured to input the state information and the first perception information of the current moment into a motion strategy model to obtain a target action of the current moment output by the motion strategy model;

[0015] a motion control module configured to control motion of the legged robot according to the target action of the current moment.

[0016] In a third aspect, an embodiment of the present application provides an electronic device, comprising:

[0017] a processor and a memory, the memory being configured to store a computer program, and the processor being configured to invoke and run the computer program stored in the memory to execute the motion control method of the legged robot according to the first aspect and the implementation manners.

[0018] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium configured to store a computer program, and the computer program causes a computer to execute the motion control method of the legged robot according to the first aspect and the implementation manners.

[0019] In a fifth aspect, an embodiment of the present application provides a computer program product comprising program instructions, and the program instructions cause an electronic device to execute the motion control method of the legged robot according to the first aspect and the implementation manners when the program instructions are run on the electronic device.

[0020] The technical solution disclosed in this application acquires the state information, encoding information, and historical actions of a legged robot at the previous moment, as well as the first perception information of the legged robot at the current moment. The state information, encoding information, and historical actions of the previous moment are input into a world model to obtain the state information of the current moment output by the world model. Then, the state information and the first perception information of the current moment are input into a motion strategy model to obtain the target action of the current moment output by the motion strategy model. Then, the movement of the legged robot is controlled according to the target action of the current moment. In this way, any terrain environment can be accurately perceived, thereby ensuring the stability of the legged robot during movement, avoiding problems such as falls or movement failures, and improving the motion control effect of the legged robot. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0022] Figure 1 This is a schematic diagram of the structure of a legged robot provided in an embodiment of this application;

[0023] Figure 2 A schematic diagram of a world model training scenario provided in an embodiment of this application;

[0024] Figure 3 A schematic diagram of a world model training process provided in this application embodiment;

[0025] Figure 4 A schematic diagram of a network structure for a world model provided in this application embodiment;

[0026] Figure 5 This is a schematic diagram of another world model training process provided in an embodiment of this application;

[0027] Figure 6 This is a schematic diagram of a motion strategy model training process provided in an embodiment of this application;

[0028] Figure 7 This application provides a schematic diagram illustrating the structure for determining the motion actions of a legged robot based on a world model and a motion strategy model, as shown in the embodiments of this application.

[0029] Figure 8 A flowchart illustrating a motion control method for a legged robot provided in an embodiment of this application;

[0030] Figure 9A schematic diagram of determining a target action of a legged robot according to a world model and a motion strategy model is provided in the embodiments of the present application;

[0031] Figure 10 A flowchart of a process of obtaining encoding information of a legged robot at a previous time is provided in the embodiments of the present application;

[0032] Figure 11 A structural schematic diagram of an encoding module in a world model is provided in the embodiments of the present application;

[0033] Figure 12 A schematic block diagram of a motion control device of a legged robot is provided in the embodiments of the present application;

[0034] Figure 13 A schematic block diagram of an electronic device is provided in the embodiments of the present application. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0036] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product, or device.

[0037] In the embodiments of the present application, the words "exemplary" or "for example" are used to mean serving as an example, instance, or illustration, any embodiment or aspect described in the embodiments of the present application as "exemplary" or "for example" should not be construed as being more preferred or advantageous than other embodiments or aspects. Rather, the use of "exemplary" or "for example" is intended to present concepts in a concrete manner.

[0038] In the description of the embodiments of the present application, "multiple" means two or more, that is, at least two, unless otherwise specified. "At least one" means one or more. "Any" means any one or any combination.

[0039] In the related art, the foot-type robot perceives the surrounding environment through the data obtained by different sensors in the internal sensor system of the robot, so as to control the movement of the robot. The internal sensor system refers to a system without vision, laser radar and other sensors. However, the above-mentioned movement control method can effectively deal with common terrains such as slopes and flat ground, but for complex terrain environments such as narrow passages, deep pits or deep holes, the robot cannot accurately perceive the complex terrain, resulting in errors in movement control, causing the robot to fall down or even movement failure, and affecting the movement control effect of the robot.

[0040] To solve the above technical problems, the present application provides a movement control method, device, equipment and medium for a foot-type robot, to solve the problem that the robot cannot accurately perceive complex terrains such as narrow passages, deep pits or deep holes, resulting in errors in movement control, causing the robot to fall down or even movement failure, and affecting the movement control effect of the robot.

[0041] It should be understood that the foot-type robot in the present application can include at least one leg mechanism, and each leg mechanism includes at least one joint, and each joint can be provided with a driving motor. That is, the foot-type robot can be a single-foot robot, a two-foot robot, a four-foot robot, a six-foot robot, etc., and the present application does not limit the number of leg mechanisms of the foot-type robot.

[0042] For example, as shown in FIG. 1, assuming that the foot-type robot is a four-foot robot, the four-foot robot 1000 can include a robot body 110, two front foot leg mechanisms 120 and two rear foot leg mechanisms 130. The two front foot leg mechanisms 120 can include a left front foot leg mechanism and a right front foot leg mechanism, and the two rear foot leg mechanisms 130 can include a left rear foot leg mechanism and a right rear foot leg mechanism. Figure 1 The above-mentioned two front foot leg mechanisms 120 can include a thigh mechanism 1201 and a calf mechanism 1202, and the above-mentioned two rear foot leg mechanisms 130 can include a thigh mechanism 1301 and a calf mechanism 1302. Each thigh mechanism 1201, 1301 is connected to the robot body 110 through a hip joint 140, and each calf mechanism 1202, 1302 is connected to the corresponding thigh mechanism 1201, 1301 through a knee joint 150.

[0043]

[0044] ​And, each of the hip joints 140 and each of the knee joints 150 are respectively provided with a driving motor, that is, each of the hip joints 140 corresponds to a hip joint driving motor, and each of the knee joints 150 corresponds to a knee joint driving motor, so that the torque can be output to the corresponding joint by the driving motor to drive the change of at least one of the hip joint and the knee joint, to realize the motion control of the foot-type robot.

[0045] After introducing the physical structure of the foot-type robot, the technical solutions provided by the embodiments of the present application will be described in detail through some embodiments. The embodiments described below can be combined with each other, and the same or similar concepts or processes can not be described in detail in some embodiments.

[0046] It should be understood that the technical solutions provided by the embodiments of the present application can include a model training part and a model application part. The models involved in the embodiments of the present application include a world model and a motion strategy model.

[0047] The model training process involved in the embodiments of the present application will be described in detail below. First, the training process of the world model is described in detail, Figure 2 a world model training scene provided by the embodiments of the present application, Figure 3 a world model training flowchart provided by the embodiments of the present application. In the embodiments of the present application, the world model training can be performed by the motion control device of the foot-type robot. In some embodiments, the motion control device of the foot-type robot can be composed of hardware and / or software, and can be integrated in an electronic device. In the present application, the electronic device can be selected from, but not limited to: a foot-type robot and a hardware device for controlling the foot-type robot, such as a tablet computer, a desktop computer, a simulator, etc., and the type of hardware device for controlling the foot-type robot is not limited herein.

[0048] As Figure 3 shown, the method can include the following steps:

[0049] S101, obtaining first training data, wherein the first training data includes training first perception information, training second perception information and training action information.

[0050] In the present application, the training first perception information, the training second perception information and the training action information respectively refer to the data required for training the world model.

[0051] The training first perception information can be the proprioceptive perception information, that is, the perception information collected by different sensors in the self-built sensor system of the foot-type robot, which is a system without vision, laser radar and other sensors.

[0052] The training second perception information can be a depth image collected by a depth camera on the legged robot. That is, the legged robot in the present application is also provided with at least one depth camera 160 to collect a depth image for perceiving the surrounding environment through the depth camera 160. As shown in Figure 1 The depth camera 160 can be arranged on the robot body 110 of the legged robot.

[0053] In some optional embodiments, when the training first perception information, the training second perception information and the training action information are acquired, at least one of the following methods can be used:

[0054] Firstly, the legged robot is controlled to move in a real environment, and the first perception information is acquired through different sensors in the self-built sensor system of the legged robot during the movement, and the first perception information is taken as the training first perception information. The second perception information is acquired through the depth camera on the legged robot, and the second perception information is taken as the training second perception information. And the movement information of the legged robot in the real environment is acquired, and the movement information is taken as the training action information.

[0055] Secondly, a simulation environment (i.e. a simulated environment) is constructed in a simulation software, and scenes under different illuminations, materials and visual angles are simulated to generate corresponding depth images, and the depth images are taken as the training second perception information. Meanwhile, the movement of the legged robot is simulated in the simulation environment, and the first perception information is acquired through different sensors in the self-built sensor system of the legged robot during the movement, and the first perception information is taken as the training first perception information. And the movement information of the legged robot during the movement is acquired, and the movement information is taken as the training action information.

[0056] In addition to the above two methods, other methods can also be used to acquire the training first perception information, the training second perception information and the training action information, which are not limited herein.

[0057] After the training first perception information, the training second perception information and the training action information are acquired, the first training data can be obtained according to the training first perception information, the training second perception information and the training action information.

[0058] It should be noted that the first training data in the present application can include multiple groups of training data, and each group of training data includes the training first perception information, the training second perception information and the training action information. That is, each group of training data includes one training first perception information, one training second perception information corresponding to the training first perception information and one training action information.

[0059] S102, according to the first training data, the initial world model is trained to obtain a world model.

[0060] In the present application, the initial world model refers to a world model whose network parameters are not adjusted. That is, the initial world model is a world model in a training state. The initial world model can be any network model that supports generating quadruped robot state information.

[0061] Optionally, training the initial world model to obtain the world model can include the following steps:

[0062] Step one, input the first training data into the initial world model to obtain the result information output by the initial world model.

[0063] Step two, train the initial world model according to the result information to obtain the world model.

[0064] In some optional embodiments, after the initial world model obtains the first training data, the initial world model can process the first training data and output a corresponding processing result. Then, the present application can calculate a first loss value according to the processing result. Then, it is compared whether the first loss value is less than a first preset value. If the first loss value is less than the first preset value, it means that the processing effect of the initial world model on the first training data is good, that is, the initial world model achieves the expected effect, at which time the training operation on the initial world model can be ended, and the initial world model can be used as the final world model. If the first loss value is greater than or equal to the first preset value, it means that the processing effect of the initial world model on the first training data is poor, that is, the initial world model does not achieve the expected effect, at which time it is determined that the training of the initial world model has not ended, and then the initial world model can be trained according to the first training data to reduce the first loss value by increasing the number of iterations or adjusting the training parameters, such as the number of first training data, so that the first loss value corresponding to the latest trained world model is less than the first preset value.

[0065] In the present application, the first preset value is an adjustable parameter, which can be set according to the prediction accuracy of the world model, and is not limited herein.

[0066] In some optional embodiments, during the training of the initial world model, the first loss value corresponding to the trained world model can tend to a stable value after multiple iterations, that is, the change of the first loss value is very small or fluctuates within a certain range. At this time, it is determined that the training of the initial world model tends to converge, and then the latest trained world model can be determined as the final world model.

[0067] It should be understood that the training of the initial world model in the present application specifically refers to the adjustment of the network parameters in the initial world model, such as the adjustment of the weight and other parameters in the initial world model.

[0068] The technical solution disclosed in this application obtains first training data for training a world model, and trains the initial world model based on the first training data to obtain a world model. Based on the world model, the state information of the legged robot can be obtained, and the execution action of the legged robot can be determined based on the state information. Based on the execution action, the movement of the legged robot can be accurately controlled, thereby improving the motion control effect of the legged robot.

[0069] In some embodiments, such as Figure 4 As shown, the initial world model in this application may include: a loop module, an encoding module, a prediction module, and a decoding module. The output of the loop module is connected to the input of the encoding module, the input of the prediction module, and the input of the decoding module; the output of the encoding module is connected to the input of the loop module and the input of the decoding module.

[0070] So, in Figure 4 Based on the network model structure shown, such as Figure 5 As shown, step S102 above may include the following steps S102-1 to S102-5:

[0071] S102-1, Input the training action information from the first training data into the loop module to obtain the status information output by the loop module.

[0072] The loop module is used to determine the state information of the legged robot at different times. In this application, the state information can be represented as h.

[0073] In some optional embodiments, the training action information in the first training data can be input as input data to the loop module, so that the loop module processes the training action information to obtain the state information corresponding to the training action information.

[0074] Considering that the first training data may include multiple sets of training data, each set of training data may correspond to different times, and the implementation principle of the training process for each set of training data is the same, for ease of explanation, this application may optionally use the training action information in one set of training data in the first training data as an example for illustration. In some optional embodiments, during each training process, a training action information from one set of training data can be input into the initial world model or the trained world model, and after the training of that set of training data is completed, a training action information from the next set of training data can be input to start training.

[0075] In other words, this application allows for the input of training action information from a set of training data at a time when training the world model, so that the loop module can process the training action information from the set of training data to obtain the state information corresponding to the training action information.

[0076] S102-2, Input the state information output by the loop module, the training first perception information and the training second perception information in the first training data into the encoding module to obtain the encoding information output by the encoding module, and input the state information output by the loop module into the prediction module to obtain the prediction encoding information output by the prediction module.

[0077] The encoding module is used to determine random state information. Specifically, the encoding module outputs random state information based on the state information output by the loop module, the first perceptual information trained in the first training data, and the second perceptual information trained in the first training data. In this application, the random state information can be represented as z.

[0078] The prediction module is used to dynamically predict random state information; that is, based on the state information output by the loop module, the prediction module outputs predicted random state information. In this application, the predicted random state information can be represented as...

[0079] In some optional embodiments, the encoding module obtains encoded information based on the state information output by the loop module, the training first perception information and the training perception information in the first training data, which can be achieved by the following formula (1):

[0080] z t ~q φ (·|h t o t )......................(1)

[0081] Where, z t This represents the encoded information at time t output by the encoding module; ~ represents the information from q. φ (·|h t o t Sampling from q φ (·|h t o t Obtain the encoded information z at time t from ) t ;q φ Represents the coding distribution function; · represents the coding distribution; h t Represents the state information at time t and o t This represents the fused sensory information at time t, and the state information h. t This is the information output by the loop module; it is the fused sensing information. t It can be obtained by fusing the training first-perceptual information and the training perceptual information at time t in the first training data, that is... in d represents the training of the first perceptual information at time t. tThis indicates that the second perceptual information is trained at time t.

[0082] It should be understood that the fusion of sensory information... t This can also be referred to as fused observation information, and no restrictions are placed on it here.

[0083] In some alternative embodiments, the prediction module obtains the prediction encoding information based on the state information output by the loop module, which can be achieved by the following formula (2):

[0084]

[0085] in, This represents the predicted coding information at time t output by the prediction module, and ~ indicates the information from p. φ (·|h t Sampling from p φ (·|h t Obtain the prediction coding information at time t from ) p φ Represents the predicted distribution function; · represents the predicted distribution; h t This represents the state information at time t, and the state information h t This is the information output by the loop module.

[0086] S102-3, Input the status information output by the loop module and the encoding information output by the encoding module into the decoding module to obtain the decoding information output by the decoding module.

[0087] The decoding module is used to reconstruct data based on the state information output by the loop module and the encoded information output by the encoding module, to obtain the reconstructed sensing information. That is, the decoded information output by the decoding module is the reconstructed sensing information. In this application, the reconstructed sensing information can be represented as...

[0088] In some optional embodiments, the decoding module obtains the decoding information based on the state information output by the loop module and the encoding information output by the encoding module, which can be achieved by the following formula (3):

[0089]

[0090] in, This represents the decoded information at time t output by the decoding module, and ~ represents the information from p. φ (·|h t , z t Sampling from p φ (·|h t , z t Obtain the decoding information at time t from ) p φRepresents the predicted distribution function; · represents the predicted distribution; h t This represents the state information at time t, and the state information h t This is the information output by the loop module; z t This represents the encoded information output by the encoding module at time t.

[0091] In this application, the aforementioned decoding information at time t... in This indicates that the first perceived information is restored at time t. This indicates that the second sensory information is restored at time t.

[0092] S102-4, Based on the decoding information output by the decoding module, the encoding information output by the encoding module, and the prediction encoding information output by the prediction module, the initial world model is trained to obtain the world model.

[0093] In some optional embodiments, a first loss value can be calculated based on the decoding information output by the decoding module, the encoding information output by the encoding module, and the prediction encoding information output by the prediction module. Then, the initial world model is back-trained based on the first loss value to adjust the network parameters in the initial world model to obtain the final world model.

[0094] Optionally, the first loss value can be calculated using the following formula (4) based on the decoding information output by the decoding module, the encoding information output by the encoding module, and the prediction encoding information output by the prediction module:

[0095]

[0096] Where L(φ) represents the first loss value; E represents the expectation function; q φ The code represents the distribution function; t represents time t. This represents the decoded information at time t output by the decoding module; β represents the weight value, which is an adjustable parameter; KL (Kullback-Leibler Divergence) represents the KL divergence function; z t This represents the encoded information output by the encoding module at time t. This represents the prediction encoding information at time t output by the prediction module.

[0097] Based on the aforementioned formulas (1) to (3), formula (4) can be transformed to obtain formula (5):

[0098]

[0099] In some optional embodiments, this application performs reverse training on the initial world model based on the first loss value to adjust the network parameters in the initial world model to obtain the final world model, which may include the following steps:

[0100] The first step is to determine whether the first loss value is less than the first preset value. If it is greater than or equal to the first preset value, then proceed to the second step. If it is less than the first preset value, then proceed to the third step.

[0101] The second step involves reversing the training process. If the first loss value is greater than or equal to the first preset value, it indicates that the initial world model's output is poor, meaning the training of the world model is not yet complete. In this case, the world model can be back-trained based on the first loss value, and the new first loss value can be determined again to be greater than or equal to the first preset value. If the new first loss value is still greater than or equal to the first preset value, this second step is repeated until the training stopping condition is met. If the latest first loss value is less than the first preset value, it indicates that the latest trained world model's output has achieved the expected results, and the latest trained world model can be determined as the final world model.

[0102] In this embodiment of the application, the training stopping condition can be that the first loss value is less than a first preset value, the number of training iterations is greater than a preset number, or the first loss value tends to any stable value.

[0103] The first preset value and the preset number of times are both adjustable parameters. They can be flexibly set according to the effect or accuracy of the output processing results of the world model. No restrictions are imposed on them here.

[0104] Third, if the first loss value is less than the first preset value, it means that the processing result output by the initial world model is good and achieves the expected effect. At this time, the training operation of the world model can be ended, and the world model can be used as the final world model.

[0105] It should be noted that after the world model training is completed, the prediction and decoding modules in the world model will cease to function and will therefore be ineffective during the use of the world model. In other words, the prediction and decoding modules in the aforementioned world model only function during the world model training process (training state), and are not needed during the world model usage process (prediction state). Put simply, when the model is in the prediction state, the data output by the prediction and decoding modules is not considered.

[0106] After a detailed explanation of the training process for the world model, it can be combined with... Figure 6 The training process of the motion strategy model involved in the embodiments of this application will be described in detail. Figure 6This is a schematic diagram illustrating a motion strategy model training process provided in an embodiment of this application. In this embodiment, the motion strategy model training can be performed by a motion control device of a legged robot. In some embodiments, the motion control device of the legged robot can consist of hardware and / or software and can be integrated into an electronic device. In this application, the electronic device may be, but is not limited to, a legged robot and hardware devices for controlling the legged robot, such as tablet computers, desktop computers, simulators, etc. No limitation is placed on the type of hardware device used to control the legged robot.

[0107] like Figure 6 As shown, the method may include the following steps:

[0108] S201, Obtain second training data, wherein the second training data includes training first perception information and training state information.

[0109] S202, Based on the second training data, the initial motion strategy model is trained to obtain the motion strategy model.

[0110] In this application, the training first perception information can be the training first perception information in the first training data when training the world model in the foregoing embodiments, or it can be other training first perception information besides the training first perception information in the first training data, etc. There are no specific limitations on it here.

[0111] Training state information can refer to the state information of a legged robot used to train a motion strategy model.

[0112] The initial motion strategy model refers to a network model whose network parameters have not been adjusted in any way. That is, the initial motion strategy model is a motion strategy model in a state of being to be trained. It should be understood that the initial motion strategy model in this application can be any network model that supports the generation of motion strategies for legged robots.

[0113] In some embodiments, a simulation environment can be constructed in simulation software, and the movement of the legged robot can be simulated in the simulation environment to obtain the state information of the legged robot during the movement process. The obtained state information can then be used as training state information. Alternatively, the legged robot can be controlled to move in a real environment to obtain the state information of the legged robot during the movement process, and the obtained state information can then be used as training state information. There are no restrictions on the implementation method of obtaining training state information, as long as the training state information can be obtained.

[0114] After obtaining the second training data, this application can input the training first perception information and training state information from the second training data into the initial motion policy model, so that the initial motion policy model processes the input data to obtain the output result. Then, a second loss value is calculated based on the output result, and the initial motion policy model is trained based on the second loss value to obtain the motion policy model.

[0115] In some alternative embodiments, training the initial motion policy model based on the second loss value to obtain the motion policy model may include the following steps:

[0116] Step 1: Determine whether the second loss value is greater than or equal to the second preset value. If it is greater than or equal to the second preset value, proceed to Step 2; otherwise, proceed to Step 3.

[0117] Step 2: If the second loss value is determined to be greater than or equal to the second preset value, it indicates that the training of the motion strategy model is not yet complete, meaning the motion strategy model has not achieved the expected results. At this point, the motion strategy model is back-trained based on the second loss value, and the new second loss value is again determined to be greater than or equal to the second preset value. If the new second loss value is still greater than or equal to the second preset value, Step 2 is not executed until the training stop condition is met. If the latest second loss value is less than the second preset value, it indicates that the output result of the latest trained motion strategy model has achieved the expected results. At this point, the training operation on the motion strategy model is stopped, and the latest trained motion strategy model is determined as the final motion strategy model.

[0118] In this embodiment of the application, the training stopping condition can be that the second loss value is less than a second preset value, the number of training iterations is greater than a preset number, or the second loss value tends to any stable value.

[0119] The second preset value and the preset number of times are both adjustable parameters. They can be flexibly set according to the output effect or accuracy of the motion strategy model. No restrictions are imposed on them here.

[0120] Step 3: If the second loss value is determined to be less than the second preset value, it means that the training of the motion strategy model has ended, that is, the output result of the motion strategy model has achieved the expected effect, and at this time the motion strategy model can be used as the final motion strategy model.

[0121] This application embodiment obtains a motion strategy model by acquiring second training data and training the initial motion strategy model based on the first perception information and training state information in the second training data. Based on this motion strategy model, the optimal motion strategy for controlling the motion of the legged robot can be obtained, and the motion of the legged robot can be accurately and effectively controlled based on the optimal motion strategy.

[0122] The training process of the world model and the motion policy model has been explained in detail above. The following section will combine... Figure 7 and Figure 8 This paper provides a detailed explanation of how to determine the motion actions of a legged robot using a world model and a motion strategy model, and how to control the movement of the legged robot based on these motion actions.

[0123] Figure 7 This is a schematic diagram illustrating a method for determining the motion of a legged robot based on a world model and a motion strategy model, as provided in an embodiment of this application. Figure 8 This is a flowchart illustrating a motion control method for a legged robot provided in this application embodiment. In other words, this application embodiment mainly introduces the process of determining the motion information of the legged robot when performing a task using the world model and motion strategy model obtained from the above training.

[0124] like Figure 8 As shown, the motion control method for this legged robot may include the following steps:

[0125] S301, acquire the legged robot's state information, encoded information, and historical actions at the previous moment, as well as the first perception information at the current moment.

[0126] S302, input the state information, encoding information and historical actions of the previous moment into the world model, and obtain the state information of the current moment output by the world model.

[0127] In some optional embodiments, when the legged robot performs any task in a real environment, it can acquire the legged robot's state information, encoding information, and historical actions at each moment, as well as the first perceptual information (i.e., proprioception) collected by different sensors in the legged robot's built-in sensor system. Therefore, this application can acquire the legged robot's state information, encoding information, and historical actions at the previous moment, and the first perceptual information at the current moment collected by different sensors in the legged robot's built-in sensor system.

[0128] Furthermore, the state information, encoding information, and historical actions of the previous moment can be input into the world model to obtain the state information of the current moment output by the world model.

[0129] according to Figure 4 As can be seen, the world model includes a loop module. Therefore, this application inputs the state information, encoding information and historical actions of the previous moment into the world model. Specifically, it inputs the state information, encoding information and historical actions of the previous moment into the loop module of the world model so that the loop module can process the state information, encoding information and historical actions of the previous moment to obtain the state information of the current moment.

[0130] In other words, this application obtains the current state information output by the loop module by inputting the previous state information, encoding information and historical actions into the loop module.

[0131] In some alternative embodiments, the loop module obtains the current state information based on the previous state information, encoding information, and historical actions, which can be achieved through the following formula (6):

[0132] h t =f φ (h t-1 , z t-1 a t-1 )........................(6)

[0133] Among them, h t This represents the current state information output by the loop module; f φ h represents the processing function. t-1 This represents the state information of the previous moment; z t-1 This represents the encoded information from the previous moment; a t-1 It indicates a historical action taken at a previous moment.

[0134] It should be noted that the encoded information z from the previous moment mentioned above... t-1 It is based on the state information h from the previous moment. t-1 Generated.

[0135] S303, input the current state information and the first perception information into the motion strategy model to obtain the target action of the current moment output by the motion strategy model.

[0136] In some optional embodiments, the current state information output by the loop module in the world model, along with the current first perception information collected by different sensors in the legged robot's built-in sensor system, can be input into the motion strategy model as input data. This allows the motion strategy model to process the current state information and the first perception information to obtain the optimal motion strategy for controlling the legged robot's movement. Furthermore, based on the optimal motion strategy, the target action for controlling the legged robot at the current moment can be obtained.

[0137] Optionally, the above motion strategy model processes the current state information and the first perception information to obtain the optimal motion strategy for controlling the legged robot's motion, which can be achieved through the following formula (7):

[0138]

[0139] Where π represents the optimal motion strategy. h represents the first perceived information at the current moment.t This represents the state information at the current moment, and sg() represents the stopping gradient function.

[0140] Furthermore, the target action of the legged robot at the current moment, obtained based on the optimal motion strategy, can be achieved through the following formula (8):

[0141]

[0142] Among them, a t This indicates the target action at the current moment.

[0143] S304 controls the movement of the legged robot based on the target action at the current moment.

[0144] After obtaining the target action of the legged robot at the current moment, this application can control each joint in each leg mechanism of the legged robot to perform corresponding movements according to the target action at the current moment, so as to execute the target action at the current moment, thereby realizing the control of the movement of the legged robot.

[0145] In some optional embodiments, the legged robot in this application may include at least one leg mechanism, wherein each leg mechanism includes at least one joint, and each joint corresponds to a drive motor. Therefore, controlling the movement of the legged robot according to the target action at the current moment can be achieved by determining the target torque corresponding to each joint in each leg mechanism based on the target action at the current moment. Then, each drive motor is controlled to output the corresponding target torque to each joint, driving each joint to rotate by a corresponding angle, so that other joints and related mechanisms associated with each joint can generate a certain amount of movement in space, thereby achieving motion control of the legged robot.

[0146] Specifically, the target torque corresponding to each joint in each leg mechanism is determined based on the target action at the current moment. This can be achieved by using the PD control law to determine the target torque corresponding to each joint in each leg mechanism based on the target action at the current moment.

[0147] In some alternative embodiments, the above-mentioned determination of the target torque corresponding to each joint in each leg mechanism based on the target action at the current moment according to the PD control law can be achieved by the following formula (9):

[0148]

[0149] Where τ represents the target torque corresponding to each joint, and K P q represents the position control gain. b K represents the desired position of each joint, q represents the actual position of each joint, and K represents the desired position of each joint. d Indicates speed control gain. This represents the expected velocity of each joint. This represents the actual speed of each joint. Wherein, K... P and K d All parameters are adjustable.

[0150] The above q b =q stand +a t , where q stand This indicates the default position of the standing joint of a legged robot, a t This indicates the target action at the current moment.

[0151] Furthermore, the above The desired position q of each joint can be determined. d The difference is obtained by performing a differential operation, and It can be obtained by performing differential processing on the actual position q of each joint.

[0152] It should be understood that PD control law refers to PD control algorithm. This PD control algorithm is a simplified form of PID control algorithm, which adjusts the system output value and rate of change to ensure stable system operation. Compared to PID control algorithm, PD control algorithm is simpler, clearer, and easier to implement.

[0153] Typically, a PD control algorithm consists of two parts: a proportional component and a derivative component. The proportional component calculates the difference between the current error and the setpoint, multiplying it by a proportional coefficient Kp to obtain the first output signal. The derivative component calculates the rate of change between the current error and the previous error, multiplying it by a derivative coefficient Kd to obtain the second output signal. Finally, the first and second output signals are added together to obtain the final output signal.

[0154] The technical solution disclosed in this application acquires the state information, encoding information, and historical actions of a legged robot at the previous moment, as well as the first perception information of the legged robot at the current moment. The state information, encoding information, and historical actions of the previous moment are input into a world model to obtain the state information of the current moment output by the world model. Then, the state information and the first perception information of the current moment are input into a motion strategy model to obtain the target action of the current moment output by the motion strategy model. Then, the movement of the legged robot is controlled according to the target action of the current moment. In this way, any terrain environment can be accurately perceived, thereby ensuring the stability of the legged robot during movement, avoiding problems such as falls or movement failures, and improving the motion control effect of the legged robot.

[0155] Based on the above embodiments, considering that the movement speed of legged robots is not very fast and the hardware processing capabilities are limited, determining the target action at every moment may increase the data processing burden and affect the movement performance of the legged robot. Therefore, this application can determine the target action at preset time intervals, thereby avoiding target action prediction at every moment. This ensures stable operation of the legged robot in any terrain environment while improving the data processing efficiency of the legged robot, thus improving the movement performance of the legged robot.

[0156] In this application, the aforementioned preset time interval can be flexibly set according to the movement speed and other movement performance of the legged robot, and no specific restrictions are imposed here.

[0157] It should be noted that when determining the target action according to the preset time interval, this application obtains the state information and encoding information of the legged robot at the previous moment, specifically the state information and encoding information of the legged robot at the previous preset time interval, and the corresponding historical actions of the legged robot at the previous moment, specifically all historical actions during the previous preset time interval.

[0158] For example, assuming a preset time interval of k, meaning the target action is determined every k time intervals, then determining the target action of a legged robot based on a world model and motion strategy can be done as follows: Figure 9 As shown.

[0159] The following is combined Figure 10 ,right Figure 8 In step S301, the encoded information of the legged robot from the previous moment is obtained for further interpretation and explanation. For example... Figure 10 As shown, step S301 above may include the following steps S301-1 to S301-2:

[0160] S301-1, acquire the first and second perception information of the legged robot at the previous moment.

[0161] Optionally, this application can acquire the first perception information from the previous moment through different sensors in the legged robot's built-in sensor system. Furthermore, based on... Figure 1 As can be seen from the structure of the legged robot shown, the legged robot includes a depth camera, so it can collect second perception information from the previous moment through the depth camera.

[0162] S301-2, input the previous moment's state information, first perception information and second perception information into the encoding module to obtain the previous moment's encoded information output by the encoding module.

[0163] according to Figure 4As shown, the world model in this application may include an encoding module. Therefore, this application can input the state information, first perception information and second perception information obtained at the previous moment into the encoding module of the world model, so that the encoding module can process the state information, first perception information and second perception information at the previous moment to obtain the encoded information at the previous moment.

[0164] In some alternative embodiments, such as Figure 11 As shown, the above-mentioned encoding module may include a first processing unit and a second processing unit. Then, the state information, first perception information and second perception information of the previous moment are input into the encoding module to obtain the encoding information of the previous moment output by the encoding module. Alternatively, the first perception information and second perception information of the previous moment can be input into the first processing unit to obtain the fused perception information output by the first processing unit. Then, the fused perception information and the state information of the previous moment are input into the second processing unit to obtain the encoding information of the previous moment output by the second processing unit.

[0165] It should be understood that the fused sensing information output by the first processing unit can be represented as o t-1 ,and in, d represents the first perceived information from the previous moment. t-1 This represents the second sensory information from the previous moment.

[0166] In some optional embodiments, the fused sensing information and the state information of the previous moment are input into the second processing unit to obtain the encoded information of the previous moment output by the second processing unit, which can be achieved by the following formula (10):

[0167] z t-1 ~q φ (·|h t-1 ,o t-1 ).........................(10)

[0168] Among them, z t-1 This represents the encoded information at time t-1 output by the second processing unit; ~ represents the information from q φ (·|h t-1 o t-1 Sampling from q φ (·|h t-1 o t-1 Obtain the encoded information z at time t-1 from ) t-1 ;q φ Represents the coding distribution function; · represents the coding distribution; h t-1 This represents the state information at time t-1 and o t-1 This represents the fused sensing information at time t-1.

[0169] In summary, the encoding information obtained in this application for the previous moment is determined by the state information, the first perception information, and the second perception information of the previous moment.

[0170] The following is a reference to the appendix. Figure 12 This application describes a motion control device for a legged robot based on an embodiment. Figure 12 As shown, the motion control device 400 of the legged robot includes: an information acquisition module 410, an information determination module 420, an action determination module 430, and a motion control module 440.

[0171] Among them, the information acquisition module 410 is used to acquire the legged robot's state information, encoding information and historical actions at the previous moment, as well as the first perception information at the current moment;

[0172] The information determination module 420 is used to input the previous moment's state information, encoding information, and historical actions into the world model to obtain the current moment's state information output by the world model;

[0173] The action determination module 430 is used to input the current state information and the first perception information into the motion strategy model to obtain the target action at the current moment output by the motion strategy model.

[0174] The motion control module 440 is used to control the movement of the legged robot according to the target action at the current moment.

[0175] In one optional implementation of this application embodiment, the world model includes an encoding module, and the information acquisition module 410 includes: an acquisition unit and a processing unit;

[0176] The acquisition unit is used to acquire the first and second perception information of the legged robot at the previous moment.

[0177] The processing unit is used to input the previous moment's state information, first sensing information, and second sensing information into the encoding module to obtain the previous moment's encoded information output by the encoding module.

[0178] In one optional implementation of this application, the encoding module includes: a first processing unit and a second processing unit, wherein the processing unit is specifically used for:

[0179] The first and second perception information from the previous moment are input into the first processing unit to obtain the fused perception information output by the first processing unit.

[0180] The fused sensing information and the previous moment's state information are input into the second processing unit to obtain the encoded information of the previous moment output by the second processing unit.

[0181] In one optional implementation of this application, the legged robot includes a depth camera, and the second perception information is a depth image.

[0182] In one optional implementation of this application embodiment, the world model further includes a loop module, and the information determination module 420 is specifically used for:

[0183] The previous state information, encoding information, and historical actions are input into the loop module to obtain the current state information output by the loop module.

[0184] In one optional implementation of this application embodiment, the legged robot further includes at least one leg mechanism, wherein each leg mechanism includes at least one joint, and each joint corresponds to a drive motor, and the motion control module 440 is specifically used for:

[0185] Based on the target action at the current moment, determine the target torque corresponding to each joint in each leg mechanism; control each drive motor to output the corresponding target torque to each joint.

[0186] In one optional implementation of this application, the motion control device 400 for the legged robot further includes: a first acquisition module and a first training module;

[0187] The first acquisition module is used to acquire first training data, which includes training first perception information, training second perception information, and training action information.

[0188] The first training module is used to train the initial world model based on the first training data to obtain the world model.

[0189] In one optional implementation of this application, the motion control device 400 for the legged robot further includes: a second acquisition module and a second training module;

[0190] The second acquisition module is used to acquire second training data, which includes training first perception information and training state information.

[0191] The second training module is used to train the initial motion strategy model based on the second training data to obtain the motion strategy model.

[0192] It should be understood that the device embodiments and the foregoing method embodiments can correspond to each other, and similar descriptions can be referred to the method embodiments. To avoid repetition, further details will not be provided here. Specifically, Figure 12 The device 400 shown can perform Figure 8 The corresponding method embodiments, and the foregoing and other operations and / or functions of each module in device 400 are respectively implemented to achieve Figure 8 For the sake of brevity, the corresponding processes in each method are not described in detail here.

[0193] The apparatus 400 of this application embodiment has been described above from the perspective of functional modules in conjunction with the accompanying drawings. It should be understood that this functional module can be implemented in hardware, in software instructions, or in a combination of hardware and software modules. Specifically, the steps of the first aspect method embodiment in this application can be completed by integrated logic circuits in the processor's hardware and / or by software instructions. The steps of the first aspect method disclosed in this application embodiment can be directly embodied as being executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. Optionally, the software module can be located in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the first aspect method embodiment described above.

[0194] Figure 13 This is a schematic block diagram of an electronic device provided as an embodiment of this application. Figure 13 As shown, the electronic device 500 may include a memory 510 and a processor 520. The memory 510 stores computer programs and transmits the program code to the processor 520. In other words, the processor 520 can call and run the computer program from the memory 510 to implement the motion control method for the legged robot described in the first aspect of this application embodiment.

[0195] For example, the processor 520 can be used to execute the above-described motion control method for the legged robot according to instructions in the computer program.

[0196] In some embodiments of this application, the processor 520 may include, but is not limited to:

[0197] General-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0198] In some embodiments of this application, the memory 510 includes, but is not limited to:

[0199] Volatile memory and / or non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DR RAM).

[0200] In some embodiments of this application, the computer program may be divided into one or more modules, which are stored in the memory 510 and executed by the processor 520 to complete the motion control method for the legged robot provided in this application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the electronic device.

[0201] like Figure 13 As shown, the electronic device 500 may further include:

[0202] Transceiver 530, which can be connected to processor 520 or memory 510.

[0203] The processor 520 can control the transceiver 530 to communicate with other devices; specifically, it can send information or data to other devices or receive information or data sent by other devices. The transceiver 530 may include a transmitter and a receiver. The transceiver 530 may further include antennas, and the number of antennas may be one or more.

[0204] It should be understood that the various components in the electronic device are connected through a bus system, which includes a data bus, a power bus, a control bus, and a status signal bus.

[0205] This application also provides a computer storage medium storing a computer program thereon, which, when executed by a computer, enables the computer to perform the motion control method for the legged robot described in the first aspect above.

[0206] This application also provides a computer program product containing program instructions that, when executed on an electronic device, cause the electronic device to perform the motion control method for the legged robot described in the first aspect.

[0207] When implemented using software, it can be implemented entirely or partially as a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video disc (DVD)), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0208] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0209] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0210] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. For example, the functional modules in the various embodiments of this application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0211] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0212] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A motion control method for a legged robot, characterized in that, include: Acquire the legged robot's state information, encoded information, and historical actions at the previous moment, as well as its first perception information at the current moment; The previous state information, encoding information, and historical actions are input into the world model to obtain the current state information output by the world model. The current state information and the first perception information are input into the motion strategy model to obtain the target action at the current moment output by the motion strategy model. The movement of the legged robot is controlled according to the target action at the current moment.

2. The method according to claim 1, characterized in that, The world model includes an encoding module, and the acquisition of the legged robot's encoded information at the previous moment includes: Acquire the first and second sensory information of the legged robot at the previous moment; The previous state information, the first perception information, and the second perception information are input into the encoding module to obtain the encoding information of the previous moment output by the encoding module.

3. The method according to claim 2, characterized in that, The encoding module includes a first processing unit and a second processing unit. The step of inputting the previous moment's state information, first sensing information, and second sensing information into the encoding module to obtain the encoded information of the previous moment output by the encoding module includes: The first and second perception information from the previous moment are input into the first processing unit to obtain the fused perception information output by the first processing unit. The fused sensing information and the previous moment's state information are input into the second processing unit to obtain the encoded information of the previous moment output by the second processing unit.

4. The method according to claim 2, characterized in that, The legged robot includes a depth camera, and the second sensing information is a depth image.

5. The method according to claim 1, characterized in that, The world model further includes a loop module, which inputs the previous moment's state information, encoding information, and historical actions into the world model to obtain the current moment's state information output by the world model, including: The previous state information, encoding information, and historical actions are input into the loop module to obtain the current state information output by the loop module.

6. The method according to claim 1, characterized in that, The legged robot further includes at least one leg mechanism, wherein each leg mechanism includes at least one joint, and each joint corresponds to a drive motor. Controlling the movement of the legged robot according to the target action at the current moment includes: Based on the target action at the current moment, determine the target torque corresponding to each joint in each leg mechanism; Each of the drive motors is controlled to output the corresponding target torque to each of the joints.

7. The method according to any one of claims 1-6, characterized in that, Also includes: Acquire first training data, which includes first perceptual information, second perceptual information, and action information. The world model is obtained by training the initial world model based on the first training data.

8. The method according to claim 7, characterized in that, Also includes: Acquire second training data, which includes training first perceptual information and training state information; Based on the second training data, the initial motion strategy model is trained to obtain the motion strategy model.

9. A motion control device for a legged robot, characterized in that, include: The information acquisition module is used to acquire the legged robot's state information, encoding information, and historical actions in the previous moment, as well as the first perception information in the current moment. The information determination module is used to input the previous moment's state information, encoding information, and historical actions into the world model to obtain the current moment's state information output by the world model. The action determination module is used to input the current state information and the first perception information into the motion strategy model to obtain the target action at the current moment output by the motion strategy model. The motion control module is used to control the movement of the legged robot according to the target action at the current moment.

10. An electronic device, characterized in that, include: A processor and a memory, the memory for storing a computer program, the processor for calling and running the computer program stored in the memory to perform the method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, Used to store a computer program that causes a computer to perform the method as described in any one of claims 1 to 8.

12. A computer program product containing program instructions, characterized in that, When the program instructions are executed on the electronic device, the electronic device causes the electronic device to perform the method as described in any one of claims 1 to 8.

Images