Method, device, equipment, medium and product for optimizing motion imitation of intelligent body dexterous hands

Abstract

The embodiment of the present invention provides a method for optimizing the imitation of the action of an intelligent dexterous hand, which can be applied to the field of artificial intelligence technology. The method for optimizing the imitation of the action of the intelligent dexterous hand includes: obtaining the action image data of the target execution action; extracting the corresponding action trajectory data according to the action image data; and optimizing the imitation execution of the target execution action by the dexterous hand through preset action trajectory optimization conditions according to the current execution state and action trajectory data of the dexterous hand. The embodiment of the present invention also provides a device, equipment, storage medium and program product for optimizing the imitation of the action of the intelligent dexterous hand.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, equipment, medium and product for optimizing the motion imitation of the dexterous hands of an intelligent entity. Background Art

[0002] Artificial Intelligence (AI) is an important driving force for the new round of scientific and technological revolution and industrial transformation. It is a new key technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. As an important part of intelligent science, artificial intelligence attempts to understand the essence of intelligence and produce a new intelligent machine (i.e., intelligent agent) that can respond in a similar way to human intelligence.

[0003] With the rapid development of artificial intelligence technology, following ChatGPT, embodied intelligence has become another wave of research in the field of artificial intelligence technology. Embodied intelligence refers to an intelligent system that perceives and acts based on the physical body. It obtains information, understands problems, makes decisions and takes actions through interaction with the environment, thereby generating intelligent behavior and adaptability. Therefore, in essence, embodied intelligence is proactive first-person intelligence that can actively interact with the environment, autonomously plan, make decisions and take actions. It is a highly intelligent robot with execution capabilities. However, the existing embodied intelligence is still at a relatively basic stage of development in terms of execution based on interactive perception, and has not shown a high level of intelligence. Summary of the invention

[0004] In view of at least one of the problems existing in the above-mentioned prior art, such as the low level of intelligence, the embodiments of the present invention aim to provide a method, device, equipment, medium and product for optimizing the motion imitation of the dexterous hands of an intelligent agent, which can efficiently and accurately realize the imitation and reproduction of the actions performed by human hands by the intelligent agent, thereby providing a technical solution that can enable the intelligent agent to accurately and efficiently and completely reproduce the target action after a real human or even other intelligent agent has performed the target action, thereby achieving high-precision real-time motion imitation, in order to achieve a smoother real-time motion imitation effect, and the action execution process is more accurately in line with the real human action execution reaction, with a higher level of intelligence.

[0005] One aspect of an embodiment of the present invention provides a method for optimizing motion imitation of a dexterous hand of an intelligent agent, which includes: acquiring motion image data of a target execution motion; extracting corresponding motion trajectory data based on the motion image data; and optimizing the dexterous hand's imitation execution of the target execution motion through preset motion trajectory optimization conditions based on the current execution state and motion trajectory data of the dexterous hand.

[0006] According to an embodiment of the present invention, acquiring action image data of a target performing an action includes: detecting the target performing an action in response to a visual detection instruction; and acquiring action image data according to an execution detection process of the target performing the action.

[0007] According to an embodiment of the present invention, extracting corresponding motion trajectory data according to motion image data includes: identifying motion motion data of a target motion object in each frame of image data in the motion image data; and generating motion trajectory data according to the motion motion data.

[0008] According to one embodiment of the present invention, before optimizing the dexterous hand's imitation execution of the target execution action through preset motion trajectory optimization conditions based on the dexterous hand's current execution state and motion trajectory data, it also includes: retrieving the product type information of the dexterous hand; and mapping the motion trajectory data to the dexterous hand according to the product type information.

[0009] According to one embodiment of the present invention, in optimizing the dexterous hand's imitation execution of a target execution action through preset motion trajectory optimization conditions based on the dexterous hand's current execution state and motion trajectory data, it includes: minimizing the difference in the dexterous hand's wrist posture and wrist movement speed through preset wrist tracking reward conditions based on the current wrist state of the dexterous hand's current execution state and the hand trajectory data of the motion trajectory data, thereby optimizing the dexterous hand's imitation execution of the target execution action; wherein the preset motion trajectory optimization conditions include preset wrist tracking reward conditions.

[0010] According to one embodiment of the present invention, in optimizing the dexterous hand's imitation execution of a target execution action through preset motion trajectory optimization conditions based on the dexterous hand's current execution state and motion trajectory data, it also includes: minimizing the position difference of the dexterous hand's finger key points through preset finger imitation reward conditions based on the current finger state of the current execution state of the dexterous hand and the finger trajectory data of the motion trajectory data, thereby optimizing the dexterous hand's imitation execution of the target execution action; wherein the preset motion trajectory optimization conditions include preset finger imitation reward conditions.

[0011] According to one embodiment of the present invention, in optimizing the dexterous hand's imitation execution of a target execution action through preset motion trajectory optimization conditions based on the dexterous hand's current execution state and motion trajectory data, it also includes: optimizing the dexterous hand's imitation execution of the target execution action by penalizing the element-by-element product of the dexterous hand's joint speed and torque through preset smoothness reward conditions based on the current joint state of the current execution state of the dexterous hand and the joint trajectory data of the motion trajectory data; wherein the preset motion trajectory optimization conditions include preset smoothness reward conditions.

[0012] According to one embodiment of the present invention, in optimizing the dexterous hand's imitation execution of the target execution action through preset motion trajectory optimization conditions based on the dexterous hand's current execution state and motion trajectory data, it also includes: minimizing the displacement and speed difference between the simulated object and the object trajectory data through preset object tracking reward conditions based on the current execution state of the dexterous hand and the object trajectory data, thereby optimizing the dexterous hand's imitation execution of the target execution action; and / or rewarding the contact force between the dexterous hand's fingertips and the simulated object through preset contact force reward conditions based on the current execution state of the dexterous hand and the hand-object distance data of the motion trajectory data, thereby optimizing the dexterous hand's imitation execution of the target execution action; wherein the preset motion trajectory optimization conditions include preset object tracking reward conditions and preset contact force reward conditions.

[0013] Another aspect of the embodiments of the present invention provides a motion imitation optimization device for a smart dexterous hand, which includes an image acquisition module, a trajectory extraction module and a motion optimization module. The image acquisition module is used to acquire motion image data of a target execution motion; the trajectory extraction module is used to extract corresponding motion trajectory data according to the motion image data; and the motion optimization module is used to optimize the dexterous hand's imitation execution of the target execution motion by presetting the motion trajectory optimization conditions according to the current execution state and motion trajectory data of the dexterous hand.

[0014] Another aspect of an embodiment of the present invention provides an electronic device, comprising one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above-mentioned intelligent agent dexterous hand motion imitation optimization method.

[0015] Another aspect of an embodiment of the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the above-mentioned method for optimizing the motion imitation of the intelligent agent's dexterous hands.

[0016] Another aspect of an embodiment of the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned method for optimizing the motion imitation of the dexterous hand of an intelligent agent.

[0017] The method for optimizing the action imitation of the dexterous hand of an intelligent agent provided by the embodiment of the present invention can at least partially solve the problem of low intelligence level caused by poor action execution and reproduction accuracy of the intelligent agent in the action imitation execution process in the related art, and thus can achieve at least one of the following technical effects:

[0018] After real humans or even other intelligent agents have performed the target action, the intelligent agent can accurately and efficiently reproduce the above target action in its entirety, achieve high-precision real-time action imitation, and avoid the sense of frustration in action execution, ensuring the overall fluency, realism and smoothness of the imitated action. The action execution process is more accurately aligned with the real action execution reaction of humans, and can efficiently and accurately realize the imitation and reproduction of the actions performed by human hands by the intelligent agent, allowing the intelligent agent to autonomously learn human operating actions and have a higher level of intelligence.

[0019] It should be understood that the above general description and the following detailed description are merely exemplary and illustrative and are not intended to limit the scope of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above contents and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0021] Figure 1 A diagram schematically shows an application scenario of a method, device, equipment, medium, and program product for optimizing the motion imitation of an intelligent agent's dexterous hand according to an embodiment of the present invention;

[0022] Figure 2 A flowchart of a method for optimizing the imitation of a dexterous hand movement of an intelligent agent according to an embodiment of the present invention is schematically shown;

[0023] Figure 3A A diagram schematically shows an application scenario of motion image data processing of a motion imitation optimization method for a dexterous hand of an intelligent agent according to an embodiment of the present invention;

[0024] Figure 3B A diagram schematically shows an application scenario of motion trajectory data extraction of a motion imitation optimization method for a dexterous hand of an intelligent agent according to an embodiment of the present invention;

[0025] Figure 3C A diagram schematically shows an application scenario of an object grabbing action of a dexterous hand of a specific product type according to the action imitation optimization method of the dexterous hand of an intelligent agent according to an embodiment of the present invention;

[0026] Figure 4A A diagram schematically shows an application scenario of dexterous hand motion imitation of a method for optimizing motion imitation of a dexterous hand of an intelligent agent according to an embodiment of the present invention;

[0027] Figure 4B A diagram schematically shows another application scenario of dexterous hand motion imitation of the method for optimizing the motion imitation of the dexterous hand of an intelligent agent according to an embodiment of the present invention;

[0028] Figure 5A schematic diagram showing a structural block diagram of a motion imitation optimization device for a smart dexterous hand according to an embodiment of the present invention; and

[0029] Figure 6 A block diagram of an electronic device suitable for implementing a method for optimizing motion imitation of a dexterous hand of an intelligent agent according to an embodiment of the present invention is schematically shown.

[0030] The above-mentioned drawings are part of the specification of the embodiments of the present invention, which illustrate exemplary embodiments of the present invention. The attached drawings and the description of the specification are used together to illustrate the principles of the embodiments of the present invention. It should be understood that the above general description of the drawings and the following specific implementations are only exemplary and illustrative, and they cannot limit the scope of the present invention. DETAILED DESCRIPTION

[0031] In order to make the objectives, technical solutions and advantages of the embodiments of the present invention more clearly understood, the spirit of the contents disclosed by the present invention will be clearly explained with the accompanying drawings and detailed descriptions below. After understanding the embodiments of the contents of the present invention, any technician in the relevant technical field can change and modify the techniques taught by the contents of the present invention without departing from the spirit and scope of the contents of the present invention.

[0032] The exemplary embodiments and descriptions of the present invention are used to explain the present invention, but are not intended to limit the present invention. In addition, elements / components with the same or similar reference numerals used in the drawings and embodiments are used to represent the same or similar parts.

[0033] The terms “first”, “second”, etc. used in the present invention do not particularly refer to an order or sequence, nor are they used to limit the present invention. They are only used to distinguish elements or operations described with the same technical terms.

[0034] The directional terms used in the present invention, such as up, down, left, right, front or back, etc., are only used to refer to the directions of the drawings. Therefore, the directional terms used are used to illustrate and not to limit the present invention.

[0035] The words “include,” “including,” “have,” “contain,” etc. used in the present invention are open-ended terms, meaning including but not limited to.

[0036] The term "and / or" used in the present invention includes any or all combinations of the items mentioned.

[0037] Regarding the present invention, "plurality" includes "two" and "more than two"; regarding the present invention, "plurality of groups" includes "two groups" and "more than two groups".

[0038] The terms "substantially" and "approximately" used in the present invention are used to modify any quantity or error that may vary slightly, but these slight changes or errors do not change their essence. Generally speaking, the range of slight changes or errors modified by such terms may be 20% in some embodiments, 10% in some embodiments, 5% in some embodiments, or other values. Those skilled in the art should understand that the aforementioned values ​​can be adjusted according to actual needs and are not limited thereto.

[0039] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0040] In the case of using expressions such as "at least one of A, B, and C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but not be limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or a system having A, B, C, etc.). In the case of using expressions such as "at least one of A, B, or C, etc.", it should generally be interpreted in accordance with the meaning of the expression generally understood by those skilled in the art (for example, "a system having at least one of A, B, or C" should include but not be limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or a system having A, B, C, etc.). Those skilled in the art should also understand that any transitional conjunctions and / or phrases that substantially represent two or more optional items, whether in the specification, claims, or drawings, should be understood to give the possibility of including one of these items, either of these items, or both of these items. For example, the phrase "A or B" should be understood to include the possibilities of "A" or "B", or "A and B".

[0041] To promote the development of embodied intelligence and quickly obtain large-scale, accurate and human-level operation sequences, the action training of embodied intelligent agents has become an urgent need. Existing methods mainly include exploration and operation generation based on reinforcement learning (RL), and data collection of human-machine alignment through teleoperation. However, both methods have limitations: traditional reinforcement learning requires carefully designed task-specific reward functions. For example, recent studies have proposed customized solutions for complex two-handed tasks (such as twisting bottle caps). These methods rely on task-specific reward functions, which usually limit the complexity of the task and may cause unnatural robot movements and high computational overhead; while teleoperation methods are costly and inefficient, and real-time data collection and correction are performed in a "human-involved" mode. In addition, due to the lack of tactile feedback, the movements are often stiff and unnatural, which limits the support for fine-grained operations, and the generated data are usually targeted at specific intelligent agents and are not universal.

[0042] Compared with using a large amount of data training to enable the intelligent agent to have the ability to perform corresponding actions, a more promising solution is to transfer human operation actions to the dexterous hand of the intelligent agent in the real environment and achieve this goal through imitation learning. The dexterous hand is the end tool for robot operation and action execution, and belongs to the category of end effector in the field of robotics. For humanoid robots, the dexterous hand is generally installed at the end of the robot's wrist and is a device that directly performs tasks. As the final link and execution component of the interaction between the robot and the environment, the end effector plays an extremely important role in improving the flexibility and ease of use of the robot. The quality of its performance largely determines the working performance of the entire robot.

[0043] Dexterous hands imitate the structure and function of human hands and play a key role in the interaction between robots and the environment. The dexterous hands of robots refer to human hands in structure and function, and can flexibly manipulate objects and grasp objects flexibly to meet various work needs. For example, the Tesla robot "dexterous hand" uses a more classic six-motor drive method. It uses 5 fingers like human hands and has 11 degrees of freedom. The thumb uses dual motors to drive bending and side swing, and the other four fingers are driven by one motor each. It can complete actions such as carrying, watering, assembly tasks, and grasping power tools. However, in real situations, existing dexterous hands are unable to perform more complex and delicate actions.

[0044] For example, in a real space environment, a real human and a target agent stand in front of an operating table. The human initiates a conversation with the target agent: "Can you reproduce my next action?" After receiving the conversation, the target agent will immediately observe the hand movements of the human on the opposite side of the operating table and answer the human: "I can imitate the action, please do it." In this scenario, the agent only needs to imitate and reproduce the target action according to the acquired human action execution process. Although the two-handed dexterous hand has a high degree of freedom (DoF), its morphological characteristics are still significantly different from those of the human hand, and the direct action redirection effect is often poor; in addition, although the motion capture data is usually more accurate, the error accumulation may lead to critical failures in high-precision tasks, thereby hindering the accurate execution of the dexterous hand action imitation; at the same time, the traditional motion generation method is difficult to capture fine finger movements due to the problem of smooth transition. Therefore, the above factors directly lead to many cutting-edge explorations still being limited to simple grasping and lifting action tasks, while the research on complex two-handed tasks such as unscrewing bottle caps and capping pens is still in its infancy.

[0045] One aspect of an embodiment of the present invention provides a method for optimizing motion imitation of a dexterous hand of an intelligent agent, which includes: acquiring motion image data of a target execution motion; extracting corresponding motion trajectory data based on the motion image data; and optimizing the dexterous hand's imitation execution of the target execution motion through preset motion trajectory optimization conditions based on the current execution state and motion trajectory data of the dexterous hand.

[0046] Figure 1 The application scenario diagram of the method, device, equipment, medium and program product for optimizing the motion imitation of the intelligent agent's dexterous hand according to an embodiment of the present invention is schematically shown.

[0047] like Figure 1 As shown, the application scenario 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for a communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0048] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only examples).

[0049] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0050] The server 105 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0051] It should be noted that the action imitation optimization method of the intelligent dexterous hand provided in the embodiment of the present invention can generally be executed by the server 105. Accordingly, the action imitation optimization device of the intelligent dexterous hand provided in the embodiment of the present invention can generally be set in the server 105. The action imitation optimization method of the intelligent dexterous hand provided in the embodiment of the present invention can also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Correspondingly, the action imitation optimization device of the intelligent dexterous hand provided in the embodiment of the present invention can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0052] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0053] The following will be based on Figure 1 The scene described by Figure 2~Figure 4B The motion imitation optimization method of the intelligent agent's dexterous hands in the disclosed embodiment is described in detail.

[0054] like Figure 2-3C As shown, one aspect of an embodiment of the present invention provides a method for optimizing motion imitation of a dexterous hand of an intelligent agent, which includes operations S201 to S203.

[0055] In operation S201, obtaining action image data of a target performing an action;

[0056] In operation S202, corresponding motion trajectory data is extracted according to the motion image data; and

[0057] In operation S203, according to the current execution state and motion trajectory data of the dexterous hand, the imitation execution of the target action by the dexterous hand is optimized by using preset motion trajectory optimization conditions.

[0058] The intelligent agent can be the executor of the above-mentioned action imitation optimization in the embodiment of the present invention, or it can be the executor controlled by the action imitation optimization, specifically, it can be a humanoid intelligent robot with dexterous hands or other embodied intelligence.

[0059] The target execution action is the object of action imitation of the dexterous hand in the embodiment of the present invention. For example, in a real space environment, a real human and a target agent stand in front of an operating table, and the human initiates a dialogue with the target agent: "Can you reproduce my next action?" After receiving the dialogue, the target agent will immediately observe the hand movements of the human opposite the operating table and answer the human: "I can imitate the action, please do it." At this time, the human begins to perform the following "putting a cap on the pen" operation: pick up the pen on the table with the left hand, pick up the pen cap with the right hand, and then put the pen cap on the pen with the left and right hands. Correspondingly, the target agent needs to reproduce the above operation based on the "observation" of the above human operation to achieve action imitation. Among them, the target execution action can be the above-mentioned human-performed "putting a cap on the pen" action.

[0060] Figure 3A An application scenario diagram of motion image data processing of a motion imitation optimization method for a dexterous hand of an intelligent agent according to an embodiment of the present invention is schematically shown.

[0061] Action image data is the image data of the target execution action acquired by the intelligent agent through a visual sensor (such as a camera) based on the "observation" of humans performing the above target execution action. It can be video stream data, which can define the image of human operation in the target execution action at each moment, such as Figure 3A The left side operation video image is shown.

[0062] Figure 3B An application scenario diagram of motion trajectory data extraction of a motion imitation optimization method for a dexterous hand of an intelligent agent according to an embodiment of the present invention is schematically shown.

[0063] The action trajectory data is the trajectory change data of the human hand (left hand and right hand) and objects (such as pen caps and pens) generated based on the spatial coordinates, extracted from the images of human operations in the target execution action at each moment defined in the action image data. Figure 3B shown.

[0064] The agent can detect its own environment and its own state information in real time, such as the ambient temperature, weather, street, house or even specific functional room (such as bedroom, living room, workshop, etc.). In addition, it can also detect the current dexterous hands, supporting legs, expressions, etc., and obtain the execution status and expression status of these actions. Specifically, the agent can judge the spatial environment and its own state in each round of iteration. Therefore, the current execution state can be at least one of the state information of the self-executed action and the state of the environment detected by the agent at the current moment, wherein the state information of the self-executed action can include the name of the currently executed action and the execution information corresponding to the action name (such as the torque, angle of rotation, estimated execution time of the wrist and the spatial position of each key point of the finger, etc.), and the state of the environment can include the information of the surrounding people and objects in the space where the agent is currently located, such as the size, position (distance) and even shape and color of obstacles on the route, and can also be the dynamic change information of humans or other agents in the surrounding environment, such as the speed of the person's movement, the emotional information of the face, etc.

[0065] By using the above-mentioned current execution state as the starting state for the intelligent body to execute the target execution action, the dexterous hand is further controlled to imitate the target execution action according to the extracted action trajectory data. The action execution in the imitation process can be optimized by the preset action trajectory optimization conditions of the embodiment of the present invention. Among them, the preset action trajectory optimization conditions can be reward conditions for correcting the action execution of the dexterous hand during the imitation execution of the target execution action. When the dexterous hand is in the process of imitating the execution, as long as the action at each moment deviates from the reference action at the corresponding moment of the action trajectory data, the deviation can be continuously rewarded and corrected until the deviation becomes smaller and smaller. In other words, the imitation action is more consistent with the target action.

[0066] Specifically, the intelligent agent obtains the action image data of the target action through visual "observation" of the whole process of human target execution action, and extracts the corresponding action trajectory data from the action image data. At this time, the intelligent agent can determine its current execution state and the environment information (such as whether there are obstacles blocking the intelligent agent from imitating the subsequent target execution action), and use the current execution action determined by the current execution state as the starting action (such as action name, joint position, fingertip position, corresponding angle and speed, etc.), and start to control the dexterous hand to perform action imitation. At this time, for the action execution process of the intelligent agent's dexterous hand, the execution optimization is performed using the introduced preset action trajectory optimization conditions, and the deviation of the action execution of the dexterous hand is continuously rewarded and optimized to minimize the deviation, so that the action execution trajectory formed by the action execution process of the dexterous hand can be more closely matched with the action trajectory of the action trajectory data, so as to achieve a more accurate fit between the action imitation and the real human target execution action, a smoother action execution, avoid the frustration of action execution, and achieve a more perfect action execution reproduction, so as to efficiently realize the direct transfer of human hand action operation skills to the dexterous hand.

[0067] Therefore, after real humans or even other intelligent agents have performed the target action, the intelligent agent can accurately and efficiently reproduce the above target action in its entirety, achieve high-precision real-time action imitation, and avoid the sense of frustration in action execution, ensuring that the imitated action is overall smooth, realistic and smooth. The action execution process is more accurately aligned with the real action execution reaction of humans, and can efficiently and accurately realize the imitation and reproduction of human hand actions by the intelligent agent, with a higher level of intelligence.

[0068] like Figure 2-3C As shown, according to an embodiment of the present invention, in operation S201, obtaining action image data of a target performing an action includes:

[0069] In response to the visual detection instruction, the detection target performs an action;

[0070] The action image data is obtained according to the execution detection process of the target execution action.

[0071] The visual detection instruction is a control instruction generated by the agent itself, which can be a command code, and is mainly used as a response instruction for performing detection on the target. For example, in the above-mentioned dialogue scene between a real human and an agent:

[0072] Real human: "Can you reproduce my next move?"

[0073] Agent: "I can imitate the movements, please do it."

[0074] At this time, the intelligent agent can generate corresponding visual detection instructions based on the conversation content of the real human being, "Can you reproduce my next action?", and perform visual detection (i.e., "observation") of the entire action execution process while the human being performs the target action, thereby obtaining the full amount of video data (i.e., action image data) of the entire target action execution process. The entire process can be understood as motion capture (Mocap). In addition, the intelligent agent can also generate the visual detection instruction based on the autonomous recognition of the surrounding environment. For example, when the intelligent agent walks out of the bedroom and sees the real human or other intelligent agent in the living room, the intelligent agent can directly generate the corresponding visual detection instruction, and directly detect the current execution action of the real human or other intelligent agent according to the visual detection instruction, and obtain the action image data of the current execution action, that is, complete the detection of the target execution action.

[0075] The intelligent agent continuously "observes" the execution process of the target action and obtains the action execution video stream of the entire execution process based on its own visual sensor (such as camera). The video stream can be used as the action image data that matches the target action. Figure 3A Shown in the left picture.

[0076] Therefore, by responding to the above-mentioned visual detection instructions, the intelligent agent can autonomously acquire motion image data, making it more autonomous and more intelligent.

[0077] like Figure 2-3C As shown, according to an embodiment of the present invention, in operation S202, extracting corresponding motion trajectory data according to the motion image data includes:

[0078] Identify the action motion data of the target action object in each frame of image data in the action image data;

[0079] Generate motion trajectory data based on the motion data.

[0080] For the action image data acquired through detection, the target action object in each frame of the action image data can be identified by reading each frame of the action image data frame by frame. The target action object is the executor of the target action detected by the intelligent agent, such as the real hands of a human being and the objects involved in the process of the hands performing the action. In the above process of identifying the target action object, the frame image with only the target action object can be obtained by performing background segmentation and replacement on each frame of the acquired image, such as Figure 3A shown.

[0081] According to the recognition and extraction of the target action object, frame-by-frame image data having only the target action object can be generated, wherein each frame of the frame-by-frame image data can identify the position, shape, and specific wrist, finger, and joint motion state (such as the angle and speed of the corresponding degree of freedom of the finger) of the corresponding human hand and object. The position of the target action object for each frame of the image can be defined with reference to the spatial coordinate system established in the process of the intelligent body detecting the spatial environment. Therefore, the action motion data is the above-mentioned identified data including the position, shape, and specific hand motion state of the target action object corresponding to each frame of the image.

[0082] By combining the action motion data of all frame images in all action image data with the generation time of each frame image, an action execution trajectory based on the generation time can be formed, as shown in the following example: Figure 3B The hand trajectory and object trajectory shown in the figure. Among them, the action trajectory data is the trajectory change data of the hand (left hand and right hand) and the object (such as pen cap and pen) generated based on the spatial coordinates, extracted from the image of the human operation in the target execution action at each moment defined in the action image data.

[0083] Therefore, motion capture (MoCap) can be used to more accurately obtain the salient data of human action demonstrations, and thus, through the extraction of the above-mentioned action trajectory data, it is possible to provide a trajectory basis for the intelligent agent to subsequently accurately imitate the target execution actions of the above-mentioned human demonstrations, thereby ensuring that the imitation actions of the intelligent agent are more realistic and consistent with the real human demonstration actions.

[0084] like Figure 2-3C As shown, according to one embodiment of the present invention, before operation S203 optimizes the imitation execution of the target action by the dexterous hand according to the current execution state and action trajectory data of the dexterous hand through the preset action trajectory optimization condition, it also includes:

[0085] Retrieve product type information of Dexterous Hands;

[0086] According to the product type information, the motion trajectory data is mapped to the dexterous hand.

[0087] For intelligent agents, different intelligent agents usually have different targeted functions, so the dexterous hands as their end effectors will also have differences, which can be reflected in the difference in the number of fingers, the number of joints, the degree of freedom of the joints, and the speed of joint rotation. Therefore, even if the mechanical dexterous hands look the same, there may be differences in the specific function implementation. Each dexterous hand will be defined as a different product type according to its different function implementation. These product types correspond to the basic mechanical design of the dexterous hand and the parameter information of the corresponding execution function, that is, the product type information. For example, the basic mechanical design of the dexterous hand is four fingers without a little finger, the index finger, middle finger, and ring finger have 4 knuckles, and the thumb has 3 knuckles. The rotating motor and gear design corresponding to each knuckle have different mechanical design indicators; at the same time, the corresponding execution function parameters are defined for each mechanical design indicator. For example, the degree of freedom of the first knuckle connecting the index finger to the palm is 2, and the knuckle can achieve a bending angle of 0-60 degrees, which can be used as its corresponding function parameter information.

[0088] There are also significant differences between the dexterous hands of different product types and the real hands of humans. In order to enable the dexterous hands to accurately imitate the target execution actions, a correspondence between the real hands of humans and the dexterous hands can be established to reduce the impact of the differences in structure and function between the two.

[0089] A primitive human hand model (such as the MANO model) can be built based on real human hands, where the primitive human hand model can have 45 degrees of freedom (DoFs), far exceeding the usual 10 to 20 degrees of freedom of dexterous hands. This difference causes the strategy based on reinforcement learning (RL) to face the challenge of too large an exploration space. Therefore, by restricting certain degrees of freedom and fixing the hand collision grid, the primitive human hand MANO model is simplified to a 22-degree-of-freedom articulated model (such as the articulated MANO model). For dexterous hands of specific product types (such as Inspire Hand and Shadow Hand), further degrees of freedom restrictions are made according to their respective structures.

[0090] In addition, in the selection of fingertip key points, the midpoint of the three fingertip anchor points can be selected as the key point. For dexterous hands, due to their different shapes, the fingertip key point can be defined as the maximum curvature point along the central axis of the fingertip.

[0091] Based on the above, we can define dexterous hands , need to cooperate to imitate human hands The behavior of two objects Interact (such as Figure 3AIn the target action task of pen cap removal shown in FIG, one hand holds the pen cap and the other hand holds the pen body). The reference trajectory (i.e., the action trajectory data) of the human demonstration of the above target action is defined as and , where T is the total number of frames and t is the time step. The trajectory of each hand Six degrees of freedom (6-DoF) pose including wrist , linear velocity and angular velocity corresponding to the posture , and the finger joint positions defined based on the original human hand model and its corresponding joint velocity , where F represents the number of key points of the hand including the fingertips. Similarly, the trajectory of each object Including its 6-DoF pose As well as its linear velocity and angular velocity At the same time, to reduce the spatial complexity, all translations can be normalized relative to the dexterous wrist position, and the rotation information can be kept to maintain the correctness of the gravity direction.

[0092] Therefore, through the original human hand model established above and the product type information of the dexterous hand, the mapping relationship between the key points of the human hand and the dexterous hand can be established, so that the key points and corresponding motion parameters of the human hands in the process of executing the target action in the action trajectory data can be mapped to the corresponding positions of the dexterous hand, and the action trajectory data can be mapped to the dexterous hand. In other words, based on the mapping relationship between the human hand model and the mechanical structure design of the dexterous hand, the action trajectory information generated by the human hand in the process of executing the target action can be transferred to the dexterous hand of the intelligent body.

[0093] In other words, it is possible to decouple the dexterous hand's imitation of human hands from the physics-based interaction constraints, significantly reducing the complexity of the dexterous hand's motion space in each action stage during the subsequent motion imitation process, thereby improving the dexterous hand's learning efficiency. For most hand-object interaction tasks (including one-handed operation of a single object, two-handed operation of a hinged object, two-handed operation of two objects, etc.), the dexterous hand can achieve smooth and natural imitation movements in a shorter time, while ensuring the realism and fluency of the dexterous hand's imitation movements.

[0094] Figure 3C A diagram schematically shows an application scenario of an object grasping action of a dexterous hand of a specific product type according to the action imitation optimization method of the dexterous hand of an intelligent agent according to an embodiment of the present invention.

[0095] like Figure 3CAs shown in the figure, a hand model of a dexterous hand (such as the AllegroHand model) can have K=16 degrees of freedom, only four fingers, and a size about twice that of a human hand, which is completely different from real human hands. To this end, the fingertip mapping relationship can be adaptively adjusted to map the little finger and ring finger of the human hand to the same fingertip of the dexterous hand model. In addition, in order to adapt to the larger size of the dexterous hand, the offset threshold of the fingertip key point can be adjusted. Relax to 8 cm. Figure 3C As shown, after testing, it can be seen that although the dexterous hand has only four fingers and is significantly larger than a human hand, its movements remain stable and natural.

[0096] Therefore, according to the mapping of the motion trajectory data and the dexterous hand, the dexterous hand can promote a more natural interaction between the dexterous hand and the object by imitating the human operation trajectory, thereby achieving more smooth and realistic movements.

[0097] like Figure 2-4B As shown, according to one embodiment of the present invention, in operation S203, according to the current execution state and action trajectory data of the dexterous hand, the imitation execution of the dexterous hand on the target action is optimized by preset action trajectory optimization conditions, including:

[0098] According to the current wrist state of the current execution state of the dexterous hand and the human hand trajectory data of the motion trajectory data, the wrist posture and wrist motion speed difference of the dexterous hand are minimized by presetting the wrist tracking reward condition, so as to optimize the imitation execution of the dexterous hand on the target execution action;

[0099] The preset motion trajectory optimization conditions include preset wrist tracking reward conditions.

[0100] For the dexterous hand that has completed the mapping of the motion trajectory data, it can perform the corresponding target execution action according to the mapping data, such as imitating Figure 3A The human hands in the video shown complete the pen cap removal task.

[0101] Specifically, we can establish a hand trajectory imitation model , in order to accurately reproduce the detailed movements of human fingers. To this end, the state of each dexterous hand can be defined as , which contains the target hand trajectory in the motion trajectory data (i.e., reference hand trajectory) and current proprioceptive information .here, and They represent the joint angles and wrist postures of the dexterous hand, and They represent the joint velocity and wrist velocity of the dexterous hand, respectively.

[0102] Furthermore, the reward function can be imitated by defining the hand trajectory , used to encourage the dexterous hand to follow the reference hand trajectory in the motion trajectory data , while ensuring the stability and smoothness of the movement.

[0103] Among them, the hand trajectory imitates the reward function Includes the Wrist Following Reward function defined by the preset wrist following reward conditions The reward is obtained by minimizing and , to encourage the wrist movement of the dexterous hand to be closer to the wrist movement of the human hand reference trajectory in the action trajectory data when performing the imitation action. express The difference in posture in space.

[0104] Therefore, through the above-mentioned preset wrist tracking reward conditions, the wrist movement can be continuously adjusted during the imitation process of the dexterous hand performing the target action, so that the difference in wrist posture and wrist movement speed is reduced to a minimum, so that the wrist movement can eventually be infinitely close to the wrist movement of the action trajectory data, ensuring that the action imitation is more realistic and more consistent with the real human hand movement.

[0105] like Figure 2-4B As shown, according to one embodiment of the present invention, in operation S203, according to the current execution state and action trajectory data of the dexterous hand, optimizing the imitation execution of the dexterous hand on the target action by preset action trajectory optimization conditions, further includes:

[0106] According to the current finger state of the current execution state of the dexterous hand and the finger trajectory data of the action trajectory data, the position difference of the key points of the fingers of the dexterous hand is minimized by presetting the finger imitation reward condition, so as to optimize the imitation execution of the dexterous hand on the target execution action;

[0107] Among them, the preset motion trajectory optimization conditions include preset finger imitation reward conditions.

[0108] Hand Trajectory Imitation Reward Function It can also include a Finger Imitation Reward function defined by the preset Finger Imitation Reward condition This reward encourages the finger joint positions of the dexterous hand to be consistent with the reference trajectory. The F key points on the dexterous hand mesh corresponding to the original human hand model (such as the MANO model) can be selected , and set the weight and attenuation coefficient To highlight the fingertips, especially the thumb, index finger and middle finger. The hyper parameters of the finger reward are shown in Table 1 below, where the weight and decay rate Used to balance the importance of each finger. Each cell contains four values, representing the parameters of the proximal, mid-segment, distal, and fingertip joints.

[0109] Table 1 - Hyperparameters of finger rewards

[0110]

[0111] Table 1 shows the reward parameters of the finger imitation reward function of the embodiment of the present invention. It can be seen that the thumb, index finger and middle finger play a key role in grasping and manipulation tasks, and their frequency of interaction with objects is significantly higher than that of other fingers. Therefore, weights can be assigned based on the contact frequency. In reality, if a dexterous hand lacks specific fingers or joints (for example, the Inspire Hand lacks distal joints), the corresponding parameters are directly set to 0.

[0112] Therefore, the preset finger imitation reward condition can greatly help alleviate the impact of the morphological difference between human hands and dexterous hands on imitation. Specifically, the finger imitation reward function corresponding to the preset finger imitation reward condition can be defined as: .

[0113] Therefore, when the dexterous hand is imitating the target action, by continuously minimizing the differences between the finger key points of the dexterous hand and the finger key points of the human hand, the movement of the finger key points can be continuously adjusted so that the difference in finger posture can be reduced to a minimum, so that it can eventually be infinitely close to the finger movement of the action trajectory data, ensuring that the detailed operation of the finger imitation is more realistic and more consistent with the real human hand movement.

[0114] Furthermore, the method does not rely on manually designed reward functions specific to action tasks, but rather on action imitation reward functions designed based on the correspondence between human fingertips and mechanical dexterous fingertips, so that the method can be effectively extended to dexterous hands of various structures, thereby making dexterous hands of different product types effectively applicable to the above-mentioned method of the embodiment of the present invention.

[0115] like Figure 2-4B As shown, according to one embodiment of the present invention, in operation S203, according to the current execution state and action trajectory data of the dexterous hand, optimizing the imitation execution of the dexterous hand on the target action by preset action trajectory optimization conditions, further includes:

[0116] According to the current joint state of the current execution state of the dexterous hand and the joint trajectory data of the action trajectory data, the element-by-element product of the joint velocity and torque of the dexterous hand is penalized by a preset smoothness reward condition, so as to optimize the imitation execution of the dexterous hand on the target execution action;

[0117] Among them, the preset action trajectory optimization conditions include preset smoothness reward conditions.

[0118] Hand Trajectory Imitation Reward Function It can also include a smoothness reward function defined by a preset smoothness reward condition By introducing smoothness rewards and penalizing the element-by-element product of joint velocity and torque, the abruptness of the movements during the operation of the dexterous hand can be effectively reduced, greatly reducing the sense of movement frustration. It is the inverse of the product of joint velocity and torque.

[0119] Furthermore, when the key point of the dexterous hand The deviation from the reference motion trajectory exceeds a preset threshold When , it can be terminated early and reinitialized to a randomly selected motion capture state. At the same time, curriculum learning is introduced to gradually reduce the preset threshold The value is adjusted to continuously reduce the sense of frustration of hand movements during the process of dexterous hands performing action imitation, and achieve concentrated and precise finger control.

[0120] Therefore, by integrating the above three reward conditions, the total reward function (hand trajectory imitation reward function) can be defined as: .

[0121] Figure 4A An application scenario diagram of dexterous hand motion imitation of a method for optimizing motion imitation of dexterous hands of an intelligent agent according to an embodiment of the present invention is schematically shown.

[0122] like Figure 4A The operation trajectory of the two-handed imitation action of the dexterous hand shown in the figure can be continuously constrained and adjusted by the above-mentioned hand trajectory imitation reward function during the imitation process, and finally high-precision and efficient imitation of hand actions such as wrists and fingers can be achieved.

[0123] Therefore, by focusing on imitating wrist and finger movements and dealing with the physical constraints of the hand, the smooth transition problem can be effectively alleviated, the fine movements of the fingers can be captured, and more precise and natural two-handed imitation operations of dexterous hands can be achieved.

[0124] like Figure 2-4BAs shown, according to one embodiment of the present invention, in operation S203, according to the current execution state and action trajectory data of the dexterous hand, optimizing the imitation execution of the dexterous hand on the target action by preset action trajectory optimization conditions, further includes:

[0125] According to the current execution state of the dexterous hand and the object trajectory data of the action trajectory data, the displacement and speed difference between the simulated object and the object trajectory data is minimized by presetting the object tracking reward condition, so as to optimize the imitation execution of the dexterous hand on the target execution action; and / or

[0126] According to the current execution state of the dexterous hand and the hand-object distance data of the action trajectory data, the contact force between the fingertips of the dexterous hand and the simulated object is rewarded through a preset contact force reward condition, so as to optimize the imitation execution of the dexterous hand on the target action;

[0127] Among them, the preset motion trajectory optimization conditions include preset object tracking reward conditions and preset contact force reward conditions.

[0128] In order to fully capture the interactive information between the dexterous hand and the object, the hand-related state of the dexterous hand Based on the state space, we can add additional information related to the interaction. First, we can obtain the object mesh based on the motion trajectory data of the motion capture data. The convex hull of the dexterous hand is used to generate collidable objects in the simulated environment. To make the object's motion trajectory Approach the reference object trajectory in the motion trajectory data, defining the object's position relative to the wrist Location and speed , its centroid , and the gravity vector , to enhance the perception of force and torque. Among them, by analyzing the object trajectory data of the action trajectory data, the dexterous hand can use the object trajectory data as a reference to establish object data in a virtual environment, that is, simulated object.

[0129] Furthermore, in order to better encode the shape of the object, the BPS (spherical point set, BPS) representation can be used to encode the object. In addition, the distance metric formula can also be used Represents the spatial relationship between the key points of the hand and the position of the object. In order to capture the tactile feedback in the interaction, the contact force generated by the simulated environment is further introduced. , which reflects the interaction between the dexterous fingertips and the surface of the object. This feedback is crucial for achieving stable grasping and manipulation, especially ensuring precise control in complex tasks.

[0130] In addition to the hand imitation reward during the imitation action execution phase , two additional reward conditions can be further introduced, namely the preset object tracking reward condition and the preset contact force reward condition.

[0131] Among them, the object following reward function defined by the preset object tracking reward condition This reward function can encourage precise manipulation by minimizing the difference in displacement and velocity between the simulated object and its reference object trajectory: and .

[0132] At the same time, the contact force reward function defined by the preset contact force reward condition This reward function can encourage appropriate contact force. When the hand-object distance in the motion trajectory data is lower than the set threshold When , the reward function is defined as follows:

[0133] .

[0134] in, Indicates fingertips The minimum distance to the object surface in the current coordinate system, is the indicator function, The contact force of the dexterous fingertips, and are weight parameters and decay coefficients, respectively, used to balance the reward function. and Both can be set to 1.

[0135] In addition, the contact termination condition can be introduced as follows: when the motion capture data (MoCap) corresponding to the action trajectory data indicates that the human hand is firmly grasping (i.e., in is the threshold value), the contact force Must be non-zero. Otherwise, it means that the dexterous hand cannot achieve a firm grasp of the object. This mechanism ensures that the agent learns to control the contact force and promotes more stable object manipulation.

[0136] Therefore, for the imitation process of the dexterous hand performing the target execution action, the total reward function of the corresponding execution trajectory can be defined as: At the same time, according to the contact termination condition, that is, when the distance between the human hand and the manipulated object is less than a certain threshold, the dexterous hand must have a more stable and powerful contact with the surface of the object, so as to obtain a more stable action imitation effect.

[0137] In summary, according to the above-mentioned imitation process of the dexterous hand executing the target action, the imitation process is continuously optimized through the above-mentioned preset action trajectory optimization conditions, so that the dexterous hand can continuously correct the accuracy of the action execution and continuously reduce the deviation of the action execution, and finally enable the dexterous hand to achieve the precise and stable placement, movement and even flipping of objects and other complex and sophisticated actions. Figure 4B Another application scenario diagram of dexterous hand motion imitation of the motion imitation optimization method of the intelligent agent dexterous hand according to an embodiment of the present invention is schematically shown. Finally, the dexterous hand can complete the accurate and efficient imitation of the target execution action, such as the precise execution of the pen cap removal action by the dexterous hand.

[0138] It can be seen that after real humans or even other intelligent agents have performed the target action, the intelligent agent can accurately and efficiently reproduce the above target action in its entirety, achieve high-precision real-time action imitation, and avoid the sense of frustration in action execution, ensuring the overall fluency, realism and smoothness of the imitated action. The action execution process is more accurately aligned with the real action execution reaction of humans, and can efficiently and accurately realize the imitation and reproduction of the actions performed by human hands by the intelligent agent, enabling the intelligent agent to autonomously learn human operating actions and have a higher level of intelligence.

[0139] Based on the above-mentioned intelligent agent dexterous hand motion imitation optimization method, the present invention also provides an intelligent agent dexterous hand motion imitation optimization device. Figure 5 The device is described in detail.

[0140] Figure 5 The structural block diagram of the motion imitation optimization device of the intelligent agent dexterous hand according to an embodiment of the present invention is schematically shown.

[0141] like Figure 5 As shown, the motion imitation optimization device 500 of the intelligent dexterous hand of this embodiment includes an image acquisition module 510, a trajectory extraction module 520 and a motion optimization module 530.

[0142] The image acquisition module 510 is used to acquire the action image data of the target performing the action. In one embodiment, the image acquisition module 510 can be used to perform the operation S201 described above, which will not be described in detail here.

[0143] The trajectory extraction module 520 is used to extract corresponding motion trajectory data according to the motion image data. In one embodiment, the trajectory extraction module 520 can be used to perform the operation S202 described above, which will not be described in detail here.

[0144] The action optimization module 530 is used to optimize the imitation execution of the target action by the dexterous hand according to the current execution state and action trajectory data of the dexterous hand by presetting the action trajectory optimization conditions. In one embodiment, the action optimization module 530 can be used to perform the operation S203 described above, which will not be repeated here.

[0145] According to an embodiment of the present invention, any multiple modules of the image acquisition module 510, the trajectory extraction module 520 and the motion optimization module 530 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present invention, at least one of the image acquisition module 510, the trajectory extraction module 520 and the motion optimization module 530 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in any appropriate combination of any of them. Alternatively, at least one of the image acquisition module 510, the trajectory extraction module 520 and the motion optimization module 530 can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be executed.

[0146] Figure 6 A block diagram of an electronic device suitable for implementing a method for optimizing motion imitation of a dexterous hand of an intelligent agent according to an embodiment of the present invention is schematically shown.

[0147] The electronic device provided by an embodiment of the present invention includes one or more processors and a memory, and the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above-mentioned intelligent agent dexterous hand motion imitation optimization method.

[0148] like Figure 6As shown, the electronic device 600 according to an embodiment of the present invention includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage part 608 to a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include an onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.

[0149] In RAM 603, various programs and data required for the operation of electronic device 600 are stored. Processor 601, ROM 602 and RAM 603 are connected to each other via bus 604. Processor 601 performs various operations of the method flow according to the embodiment of the present invention by executing the program in ROM 602 and / or RAM 603. It should be noted that the program can also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 can also perform various operations of the method flow according to the embodiment of the present invention by executing the program stored in the one or more memories.

[0150] According to an embodiment of the present invention, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to the bus 604. The electronic device 600 may further include one or more of the following components connected to the I / O interface 605: an input portion 606 including a keyboard, a mouse, etc.; an output portion 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 608 including a hard disk, etc.; and a communication portion 609 including a network interface card such as a LAN card, a modem, etc. The communication portion 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed, so that a computer program read therefrom is installed into the storage portion 608 as needed.

[0151] The present invention also provides a computer-readable storage medium on which executable instructions are stored. When the instructions are executed by a processor, the processor executes the above-mentioned intelligent agent's dexterous hand motion imitation optimization method.

[0152] The computer-readable storage medium may be included in the device / apparatus / system described in the above embodiment; or it may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present invention is implemented.

[0153] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present invention, the computer-readable storage medium may include the ROM 602 and / or RAM 603 described above and / or one or more memories other than ROM 602 and RAM 603.

[0154] An embodiment of the present invention also includes a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the above-mentioned intelligent agent dexterous hand action imitation optimization method.

[0155] The computer program includes program codes for executing the method shown in the flow chart. When the computer program product is run in a computer system, the program codes are used to enable the computer system to implement the method provided by the embodiment of the present invention.

[0156] The computer program executes the above functions defined in the system / device of the embodiment of the present invention when it is executed by the processor 601. According to the embodiment of the present invention, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0157] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 609, and / or installed from a removable medium 611. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0158] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, the above functions defined in the system of the embodiment of the present invention are performed. According to the embodiment of the present invention, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0159] According to an embodiment of the present invention, the program code for executing the computer program provided by the embodiment of the present invention can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0160] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0161] It will be appreciated by those skilled in the art that the features described in the various embodiments and / or claims of the present invention may be combined and / or combined in various ways, even if such combinations and / or combinations are not explicitly described in the present invention. In particular, the features described in the various embodiments and / or claims of the present invention may be combined and / or combined in various ways without departing from the spirit and teachings of the present invention. All of these combinations and / or combinations fall within the scope of the present invention.

[0162] The embodiments of the present invention are described above. However, these embodiments are only for the purpose of illustration, and are not intended to limit the scope of the present invention. Although the embodiments are described above, this does not mean that the measures in the various embodiments cannot be used in combination. The scope of the present invention is defined by the appended claims and their equivalents. Without departing from the scope of the present invention, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present invention.

Claims

1. A method for optimizing the imitation of the action of an intelligent agent's dexterous hand, characterized in that: include: Acquire action image data of the target performing an action; Extracting corresponding motion trajectory data according to the motion image data; as well as According to the current execution state of the dexterous hand and the motion trajectory data, the imitation execution of the target action by the dexterous hand is optimized by preset motion trajectory optimization conditions, including: According to the current wrist state of the current execution state of the dexterous hand and the human hand trajectory data of the motion trajectory data, the wrist posture and wrist motion speed difference of the dexterous hand are minimized by presetting the wrist tracking reward condition, so as to optimize the imitation execution of the target execution action by the dexterous hand; According to the current finger state of the current execution state of the dexterous hand and the finger trajectory data of the action trajectory data, the position difference of the finger key points of the dexterous hand is minimized by presetting the finger imitation reward condition, so as to optimize the imitation execution of the dexterous hand on the target execution action; According to the current joint state of the current execution state of the dexterous hand and the joint trajectory data of the action trajectory data, the element-by-element product of the joint velocity and torque of the dexterous hand is penalized by a preset smoothness reward condition, so as to optimize the imitation execution of the dexterous hand on the target execution action; According to the current execution state of the dexterous hand and the object trajectory data of the action trajectory data, the displacement and speed difference between the simulated object and the object trajectory data is minimized by a preset object tracking reward condition, thereby optimizing the imitation execution of the target action by the dexterous hand; and / or according to the current execution state of the dexterous hand and the hand-object distance data of the action trajectory data, the contact force between the fingertips of the dexterous hand and the simulated object is rewarded by a preset contact force reward condition, thereby optimizing the imitation execution of the target action by the dexterous hand; Among them, the preset action trajectory optimization condition is a reward condition for correcting the dexterous hand action execution during the imitation execution process of the dexterous hand executing the target action; the preset action trajectory optimization condition includes the preset wrist tracking reward condition, the preset finger imitation reward condition, the preset smoothness reward condition, the preset object tracking reward condition and the preset contact force reward condition.

2. The method according to claim 1, characterized in that The step of acquiring the action image data of the target performing the action includes: In response to the visual detection instruction, detecting that the target performs an action; The action image data is acquired according to an execution detection process of the target execution action.

3. The method according to claim 1, characterized in that In the step of extracting corresponding motion trajectory data according to the motion image data, the following steps are included: Identifying action motion data of a target action object in each frame of image data in the action image data; The motion trajectory data is generated according to the motion data.

4. The method according to claim 1, characterized in that: Before optimizing the imitation execution of the target action by the dexterous hand through preset action trajectory optimization conditions according to the current execution state of the dexterous hand and the action trajectory data, the method further includes: Retrieving product type information of the dexterous hand; The motion trajectory data is mapped to the dexterous hand according to the product type information.

5. A motion imitation optimization device for an intelligent dexterous hand, characterized in that: include: An image acquisition module, used to acquire action image data of the target performing an action; A trajectory extraction module, used to extract corresponding action trajectory data according to the action image data; as well as The action optimization module is used to optimize the imitation execution of the target action by the dexterous hand according to the current execution state of the dexterous hand and the action trajectory data by preset action trajectory optimization conditions, including: According to the current wrist state of the current execution state of the dexterous hand and the human hand trajectory data of the motion trajectory data, the wrist posture and wrist motion speed difference of the dexterous hand are minimized by presetting the wrist tracking reward condition, so as to optimize the imitation execution of the target execution action by the dexterous hand; According to the current finger state of the current execution state of the dexterous hand and the finger trajectory data of the action trajectory data, the position difference of the finger key points of the dexterous hand is minimized by presetting the finger imitation reward condition, so as to optimize the imitation execution of the dexterous hand on the target execution action; According to the current joint state of the current execution state of the dexterous hand and the joint trajectory data of the action trajectory data, the element-by-element product of the joint velocity and torque of the dexterous hand is penalized by a preset smoothness reward condition, so as to optimize the imitation execution of the dexterous hand on the target execution action; According to the current execution state of the dexterous hand and the object trajectory data of the action trajectory data, the displacement and speed difference between the simulated object and the object trajectory data is minimized by a preset object tracking reward condition, thereby optimizing the imitation execution of the target action by the dexterous hand; and / or according to the current execution state of the dexterous hand and the hand-object distance data of the action trajectory data, the contact force between the fingertips of the dexterous hand and the simulated object is rewarded by a preset contact force reward condition, thereby optimizing the imitation execution of the target action by the dexterous hand; Among them, the preset action trajectory optimization condition is a reward condition for correcting the dexterous hand action execution during the imitation execution process of the dexterous hand executing the target action; the preset action trajectory optimization condition includes the preset wrist tracking reward condition, the preset finger imitation reward condition, the preset smoothness reward condition, the preset object tracking reward condition and the preset contact force reward condition.

6. An electronic device comprising: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors execute the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the method according to any one of claims 1 to 4.

8. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

Images