A method and system for generating motion trajectories of humanoid robot arms

Abstract

The present application discloses a method and system for generating motion trajectories of both arms of a humanoid robot, the method comprising: obtaining motion data and interaction data of both arms of a humanoid robot and performing data preprocessing to obtain preprocessed feature data of both arms of the humanoid robot; performing causal data analysis processing on the preprocessed feature data of both arms of the humanoid robot based on a causal reasoning engine model to obtain causal relationship data; performing data interaction training processing on the preprocessed feature data of both arms of the humanoid robot based on a multi-agent interaction model to obtain multi-agent dynamic interaction data; combining causal relationship data with multi-agent dynamic interaction data to control the motion of both arms of the humanoid robot and generate motion trajectories of both arms of the humanoid robot. The embodiments of the present application can improve the generation efficiency and rationality of the motion trajectories of both arms of a humanoid robot, so that both arms of a humanoid robot can better adapt to complex and changing environments. The present application can be widely used in the field of humanoid robot control technology.

Description

Technical Field

[0001] The present application relates to the technical field of humanoid robot control, and in particular to a method and system for generating motion trajectories of both arms of a humanoid robot. Background Art

[0002] With the rapid development of robotics technology, humanoid robots are increasingly used in industrial manufacturing, services, medical and other fields. Dual-arm coordinated motion is one of the key capabilities of humanoid robots to complete complex tasks. However, related dual-arm motion trajectory generation methods often lack adaptability to complex environmental changes and understanding of deep causal relationships, making it difficult to achieve efficient, safe and ergonomic dual-arm coordinated motion, and difficult to directly meet people's production needs and applications.

[0003] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a method and system for generating motion trajectories of the dual arms of a humanoid robot, which can improve the generation efficiency and rationality of the motion trajectories of the dual arms of a humanoid robot, thereby making the dual arms of the humanoid robot more adaptable to complex and changing environments.

[0005] To achieve the above purpose, one aspect of an embodiment of the present application provides a method for generating a motion trajectory of a humanoid robot's two arms, the method comprising:

[0006] Acquire motion data of the humanoid robot's arms and interaction data of the humanoid robot's arms and perform data preprocessing to obtain preprocessed feature data of the humanoid robot's arms;

[0007] Based on the causal reasoning engine model, causal data analysis and processing are performed on the preprocessed humanoid robot dual-arm feature data to obtain causal relationship data;

[0008] Based on the multi-agent interaction model, data interaction training is performed on the pre-processed humanoid robot dual-arm feature data to obtain multi-agent dynamic interaction data;

[0009] The causal relationship data is combined with the multi-agent dynamic interaction data to control the movement of the humanoid robot's arms and generate the movement trajectory of the humanoid robot's arms.

[0010] In some embodiments, the acquiring of the motion data of the humanoid robot's arms and the interaction data of the humanoid robot's arms and performing data preprocessing to obtain the preprocessed feature data of the humanoid robot's arms includes:

[0011] Acquire motion data of both arms of the humanoid robot through sensors, wherein the motion data includes joint angle sensor data, velocity sensor data, and acceleration sensor data;

[0012] Acquire interaction data between humanoid robots and interaction data between humanoid robots and intelligent devices, and construct interaction data of both arms of the humanoid robots;

[0013] Integrating the motion data of the two arms of the humanoid robot and the interaction data of the two arms of the humanoid robot to obtain feature data of the two arms of the humanoid robot;

[0014] The feature data of both arms of the humanoid robot are standardized and subjected to data feature extraction and fusion processing to obtain the preprocessed feature data of both arms of the humanoid robot.

[0015] In some embodiments, the step of normalizing the feature data of the humanoid robot's arms and extracting and fusing the feature data to obtain the preprocessed feature data of the humanoid robot's arms includes:

[0016] Performing standardization processing on the feature data of both arms of the humanoid robot to obtain standardized feature data of both arms of the humanoid robot, wherein the standardized feature data of both arms of the humanoid robot includes structured data and unstructured data;

[0017] Performing data fusion processing on the structured data by using a feature hash algorithm to obtain fused structured data;

[0018] Performing feature extraction and fusion processing on the unstructured data through a deep learning model to obtain fused unstructured data;

[0019] The fused structured data and the fused unstructured data are integrated to obtain the preprocessed humanoid robot dual-arm feature data.

[0020] In some embodiments, the causal data analysis and processing is performed on the pre-processed humanoid robot dual-arm feature data based on the causal reasoning engine model to obtain causal relationship data, including:

[0021] Inputting the preprocessed humanoid robot dual-arm feature data into the causal reasoning engine model, wherein the causal reasoning engine model includes a graph neural network, a convolutional neural network, a recurrent neural network, and a causal inference unit;

[0022] Based on the graph neural network of the causal reasoning engine model, the structured data in the preprocessed feature data of the humanoid robot arms are processed to capture the dynamic change relationship to obtain structured dynamic change data;

[0023] Based on the convolutional neural network and the recurrent neural network of the causal reasoning engine model, the unstructured data in the preprocessed feature data of the humanoid robot arms are subjected to spatial feature extraction and temporal feature extraction to obtain unstructured spatial feature data and unstructured temporal feature data;

[0024] Based on the causal inference unit of the causal inference engine model, causal variable inference is performed on the structured dynamically changing data, the unstructured spatial feature data and the unstructured temporal feature data to obtain the causal relationship data.

[0025] In some embodiments, based on the multi-agent interaction model, the pre-processed humanoid robot dual-arm feature data is subjected to data interaction training to obtain multi-agent dynamic interaction data, including:

[0026] Inputting the preprocessed humanoid robot dual-arm feature data into the multi-agent interaction model;

[0027] Performing feature vector conversion processing on the preprocessed humanoid robot dual-arm feature data to obtain a humanoid robot dual-arm feature data vector;

[0028] Determining the number of neighbors of the humanoid robots according to the relationship between the humanoid robots;

[0029] Combined with the number of neighbors of the humanoid robot, interactive calculation is performed on the feature data vectors of both arms of the humanoid robot to obtain the multi-agent dynamic interaction data.

[0030] In some embodiments, the interactive calculation of the feature data vectors of the two arms of the humanoid robot in combination with the number of neighbors of the humanoid robot to obtain the multi-agent dynamic interaction data includes:

[0031] Determining the trainable weights of the multi-agent interaction model and the bias term of the multi-agent interaction model;

[0032] The humanoid robot dual-arm feature data vector is multiplied by the trainable weight and then divided by the number of neighbors of the humanoid robot, and then normalized and summed to obtain a weighted feature data vector of the humanoid robot dual-arm;

[0033] The bias term is introduced, and the weighted feature data vectors of the humanoid robot's arms are interactively calculated through a nonlinear activation function to obtain the multi-agent dynamic interaction data.

[0034] In some embodiments, the expression of the multi-agent interaction model is specifically as follows:

[0035] ;

[0036] In the above formula, Indicates Layer Agent The characteristic vector of Indicates Layer Agent The characteristic vector of Indicates Layer Agent The characteristic vector of Representing an Agent Neighbors, Representing an Agent With Agent In the The trainable weights of the layer, Indicates Layer trainable bias, and represents the adjustable coefficient, represents a non-linear activation function.

[0037] In some embodiments, the combining of the causal relationship data and the multi-agent dynamic interaction data to control the movement of the humanoid robot's arms and generate the movement trajectory of the humanoid robot's arms includes:

[0038] Training the adaptive behavior synthesis component on the causal relationship data and the multi-agent dynamic interaction data through reinforcement learning to obtain behavior strategy data;

[0039] The movement of the two arms of the humanoid robot is controlled according to the behavior strategy data to generate the movement trajectory of the two arms of the humanoid robot.

[0040] In some embodiments, the adaptive behavior synthesis component represents a dynamic reward function, and the expression of the dynamic reward function is specifically as follows:

[0041] ;

[0042] In the above formula, represents the efficiency parameter, Represents the security parameter, represents the multi-agent interaction parameters, Indicates the task completion parameters, represents the adaptability parameter, , , , , represents the adjustable coefficient, represents the dynamic reward function, represents the state of the agent, Represents the action of the agent.

[0043] To achieve the above-mentioned purpose, another aspect of the embodiment of the present application provides a system for generating a motion trajectory of a humanoid robot with two arms, the system comprising:

[0044] The first module is used to obtain motion data of the humanoid robot's arms and interaction data of the humanoid robot's arms and perform data preprocessing to obtain preprocessed feature data of the humanoid robot's arms;

[0045] The second module is used to perform causal data analysis on the pre-processed humanoid robot dual-arm feature data based on a causal reasoning engine model to obtain causal relationship data;

[0046] The third module is used to perform data interaction training processing on the pre-processed humanoid robot dual-arm feature data based on a multi-agent interaction model to obtain multi-agent dynamic interaction data;

[0047] The fourth module is used to combine the causal relationship data with the multi-agent dynamic interaction data to control the movement of the humanoid robot's arms and generate the movement trajectory of the humanoid robot's arms.

[0048] The embodiments of the present application include at least the following beneficial effects: The present application provides a method and system for generating the motion trajectory of both arms of a humanoid robot. The scheme obtains the motion data of both arms of the humanoid robot and the interaction data of both arms of the humanoid robot and performs data preprocessing, taking into account the motion environment data of the individual robot and the environmental data of the interaction between robots, and then based on the causal reasoning engine model, performs causal data analysis processing on the preprocessed feature data of both arms of the humanoid robot, and based on the multi-agent interaction model, performs data interaction training processing on the preprocessed feature data of both arms of the humanoid robot, and then learns the spatial relationship and temporal coordination between the two arms of the humanoid robot, and finally controls the movement of the two arms of the humanoid robot, which can improve the generation efficiency and rationality of the motion trajectory of the two arms of the humanoid robot, and make the two arms of the humanoid robot more adaptable to complex and changing environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flow chart of a method for generating a motion trajectory of a humanoid robot's two arms provided in an embodiment of the present application;

[0050] Figure 2 It is a structural schematic diagram of a humanoid robot dual-arm motion trajectory generation system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present application. They are only examples of systems and methods consistent with some aspects of the embodiments of the present application as detailed in the attached claims.

[0052] It is understood that the terms "first", "second", etc. used in this application can be used to describe various concepts in this article, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another concept. For example, without departing from the scope of the embodiment of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein can be interpreted as "at the time of" or "when" or "in response to determination".

[0053] The terms "at least one", "multiple", "each", "any", etc. used in this application, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0055] Reference Figure 1 , Figure 1 A flowchart of a method for generating a motion trajectory of a humanoid robot dual arms provided in an embodiment of the present invention, referring to Figure 1 , the method comprises the following steps:

[0056] S100, acquiring motion data of both arms of the humanoid robot and interaction data of both arms of the humanoid robot and performing data preprocessing to obtain feature data of both arms of the humanoid robot after preprocessing;

[0057] It should be noted that, in some embodiments, step S100 may include:

[0058] S110, obtaining motion data of both arms of the humanoid robot through sensors, where the motion data includes joint angle sensor data, velocity sensor data, and acceleration sensor data;

[0059] S120, acquiring interaction data between humanoid robots and interaction data between the humanoid robots and intelligent devices, and constructing interaction data of both arms of the humanoid robots;

[0060] In this embodiment, motion data is collected by means of joint motor encoders and current loops on the arms of the humanoid robot. The motion data includes: angle, speed, acceleration, and current value. At the same time, relevant data of multiple intelligent bodies such as humans, other robots, and other intelligent devices in the work scene are obtained.

[0061] S130, integrating the motion data of the two arms of the humanoid robot and the interaction data of the two arms of the humanoid robot to obtain feature data of the two arms of the humanoid robot;

[0062] S140, performing standardization and data feature extraction fusion processing on the feature data of both arms of the humanoid robot to obtain preprocessed feature data of both arms of the humanoid robot.

[0063] Specifically, these data are standardized and normalized according to the mean and standard deviation of different data. For structured data, feature hashing is used for data fusion; for unstructured data, feature extraction is performed through deep learning models, ultimately achieving effective data fusion.

[0064] In some specific embodiments, step S140 also includes: S141, standardizing the feature data of the humanoid robot's arms to obtain standardized feature data of the humanoid robot's arms, wherein the standardized feature data of the humanoid robot's arms include structured data and unstructured data; S142, performing data fusion processing on the structured data through a feature hashing algorithm to obtain fused structured data; S143, performing feature extraction and fusion processing on the unstructured data through a deep learning model to obtain fused unstructured data; S144, integrating the fused structured data and the fused unstructured data to obtain the pre-processed feature data of the humanoid robot's arms.

[0065] In this embodiment, for structured data, feature hashing is used to perform data fusion, firstly data grouping is performed, then hash calculation and fusion are performed, and finally mapping processing is performed.

[0066] Specifically;

[0067] 1) Data grouping: structured data is grouped according to certain rules, and each group of data contains several features;

[0068] 2) Hash calculation: For each set of data, use the SHA-1 hash function to calculate it and get a fixed-length hash value. Initialize five 32-bit registers A, B, C, D, and E. The initial values ​​are A = 0x67452301, B = 0xEFCDAB89, C = 0x98BADCFE, D = 0x10325476, and E = 0xC3D2E1F0. Divide each set of data into several 512-bit blocks and process each block. First, expand the current block into 80 32-bit words, and then perform loop processing, that is, perform four rounds of loops, each with 20 steps. In each step, use nonlinear functions and registers to calculate, update the register value, and finally update, add the calculated result to the previous register value, and update the register value. Finally, after all blocks are processed, combine the values ​​of registers A, B, C, D, and E to get the final hash value.

[0069] 3) Data fusion: Group or map the obtained hash values ​​into a specific space. The location of the grouping or mapping can be determined based on certain bits of the hash value;

[0070] 4) Data mapping: At each group or mapping position, the feature data belonging to the group is fused. Fusion operations include summing, averaging, and taking the maximum value.

[0071] S200, based on the causal reasoning engine model, performing causal data analysis and processing on the pre-processed feature data of the humanoid robot arms to obtain causal relationship data;

[0072] It should be noted that, in some embodiments, step S200 may include: S210, inputting the preprocessed humanoid robot dual-arm feature data into the causal reasoning engine model, the causal reasoning engine model includes a graph neural network, a convolutional neural network, a recurrent neural network and a causal inference unit; S220, based on the graph neural network of the causal reasoning engine model, performing dynamic change relationship capture processing on the structured data in the preprocessed humanoid robot dual-arm feature data to obtain structured dynamic change data; S230, based on the convolutional neural network and the recurrent neural network of the causal reasoning engine model, performing spatial feature and temporal feature extraction processing on the unstructured data in the preprocessed humanoid robot dual-arm feature data to obtain unstructured spatial feature data and unstructured temporal feature data; S240, based on the causal inference unit of the causal reasoning engine model, performing causal variable inference on the structured dynamic change data, unstructured spatial feature data and unstructured temporal feature data to obtain causal relationship data.

[0073] In some specific embodiments, the causal inference engine includes a causal inference unit and a deep learning unit. The deep learning unit includes a graph neural network, a convolutional neural network, and a recurrent neural network. Structured data is input into the graph neural network to obtain dynamically changing data, and unstructured data is input into the convolutional neural network and the recurrent neural network to obtain spatial feature information and temporal feature information. The causal inference unit processes these data to obtain causal relationship data between the humanoid robot and the multi-agent.

[0074] Furthermore, it should be noted that:

[0075] 1) Graph neural network processes structured data: After the structured data is input into the graph neural network, the graph neural network captures the dynamic changing relationships in the data through the structure of nodes and edges. Nodes represent entities or features in the data, and edges represent the relationships between them. Through the training of the graph neural network, the model learns the dynamic change patterns between different elements in the structured data, thereby obtaining dynamically changing data.

[0076] 2) Convolutional neural networks and recurrent neural networks process unstructured data: Unstructured data is input into convolutional neural networks and recurrent neural networks. Convolutional neural networks are mainly used to extract spatial features of data. They filter and extract features from data through convolution operations to capture spatial patterns and local relationships in the data. Recurrent neural networks are used to process the temporal dependency of data. They can capture causal effects that change over time and learn the changing patterns of data in the temporal dimension by processing time series data.

[0077] 3) Causal inference unit integration processing: The causal inference unit receives the dynamically changing data output by the graph neural network, the spatial feature information and the temporal feature information output by the convolutional neural network and the recurrent neural network. Then, it uses the causal inference algorithm to comprehensively process this information and infer the causal relationship between the variables in the data. Specifically, it may use methods such as counterfactual reasoning to determine which factors are the cause of certain results, as well as the strength and direction of these causal relationships. Finally, the causal inference unit outputs causal relationship data.

[0078] S300, based on the multi-agent interaction model, performing data interaction training processing on the pre-processed humanoid robot dual-arm feature data to obtain multi-agent dynamic interaction data;

[0079] It should be noted that, in some embodiments, step S300 may include: S310, inputting the preprocessed humanoid robot dual-arm feature data into the multi-agent interaction model; S320, performing feature vector conversion processing on the preprocessed humanoid robot dual-arm feature data to obtain the humanoid robot dual-arm feature data vector; S330, determining the number of humanoid robot neighbors based on the relationship between the humanoid robots; S340, performing interactive calculations on the humanoid robot dual-arm feature data vector in combination with the number of humanoid robot neighbors to obtain multi-agent dynamic interaction data.

[0080] In some specific embodiments, step S340 also includes: S341, determining the trainable weights of the multi-agent interaction model and the bias term of the multi-agent interaction model; S342, multiplying the feature data vector of the humanoid robot's two arms by the trainable weights and then dividing by the number of neighbors of the humanoid robot, normalizing and summing the result to obtain the weighted feature data vector of the humanoid robot's two arms; S343, introducing the bias term, and performing interactive calculation on the weighted feature data vector of the humanoid robot's two arms through a nonlinear activation function to obtain multi-agent dynamic interaction data.

[0081] Specifically, the humanoid robot arms and the people, other robots, other intelligent devices, etc. in the work scene are regarded as multiple interactive intelligent agents, and a multi-agent interaction model is established. Through this model, the dynamic interaction between multiple intelligent agents is simulated, the spatial relationship and temporal coordination are captured, and the impact of the behavioral changes of one intelligent agent on other humanoid robots and other intelligent agents is predicted.

[0082] Furthermore, it should be noted that:

[0083] 1) Feature vector representation: The feature vector of each agent Indicates that the agent is in The state information of the layer may include the position, speed, posture and other related features of the intelligent body.

[0084] 2) Neighbor Relationship Definition: Agent Neighbors It is defined based on the actual interaction situation. For example, in a work scenario, if there is physical proximity, information exchange, or task association between two agents, they can be considered neighbors.

[0085] 3) Interactive computing: In computing intelligent bodies In the When calculating the feature vector of the layer, the feature vectors of neighboring agents are first weighted summed. Specifically, for neighboring agents , whose eigenvector Multiply The interaction weight is divided by the number of neighbors. Normalize, and then sum the weighted eigenvectors of all neighbors. Next, add the bias term ,in is a trainable bias, is an adjustable coefficient. Finally, this result is input into the nonlinear activation function In the In the The feature vector of the layer Through such calculations, the model can simulate the dynamic interactions between multiple agents, that is, the state change of one agent will affect the states of other agents through such interactive calculations.

[0086] It should be further explained that the multi-agent interaction model includes multiple interactive agents such as the humanoid robot arms and people in the work scene, other robots, and other intelligent devices. The formula of the multi-agent interaction model is expressed as:

[0087] ;

[0088] In the above formula, Indicates Layer Agent The characteristic vector of Indicates Layer Agent The characteristic vector of Indicates Layer Agent The characteristic vector of Representing an Agent Neighbors, Representing an Agent With Agent In the The trainable weights of the layer, Indicates Layer trainable bias, and represents the adjustable coefficient, represents a non-linear activation function.

[0089] S400, combining the causal relationship data and the multi-agent dynamic interaction data to control the movement of the humanoid robot's arms and generate movement trajectories of the humanoid robot's arms;

[0090] It should be noted that, in some embodiments, step S400 may include: S410, training the adaptive behavior synthesis component of causal relationship data and multi-agent dynamic interaction data through reinforcement learning to obtain behavior strategy data; S420, controlling the movement of the humanoid robot's arms according to the behavior strategy data to generate the movement trajectory of the humanoid robot's arms.

[0091] In some specific embodiments, an adaptive behavior synthesis component is trained through reinforcement learning, causal data, and dynamic interaction data to obtain behavior strategy data, wherein the strategy data includes a strategy network, and the strategy network outputs an expected reward value based on the current state and possible behavior choices; the motion trajectory of the humanoid robot's arms is obtained based on the behavior strategy data.

[0092] Among them, the adaptive behavior synthesis component represents the dynamic reward function, and the expression of the dynamic reward function is as follows:

[0093] ;

[0094] In the above formula, Represents efficiency parameters, such as the shorter the time to complete the task, the higher the reward, Indicates safety parameters, such as avoiding collisions with obstacles, maintaining body balance, etc. Rewards are given if no unsafe behavior occurs. Represents the multi-agent interaction parameters, for example, the better the collaboration with humans, other robots, or intelligent devices, the higher the reward. Indicates the task completion parameters, for example, the higher the degree of accurate completion of the specified task, the higher the reward, Represents an adaptability parameter. For example, the stronger the robot's ability to dynamically adjust its own trajectory according to environmental changes and the behavior of the multi-agent, the higher the reward. , , , , Represents an adjustable coefficient, which is used to adjust the weight of each factor in the reward. represents the dynamic reward function, represents the state of the agent, Represents the actions of the agent.

[0095] Furthermore, it should be noted that the behavior strategy data is obtained by training the adaptive behavior synthesis component through reinforcement learning, causal relationship data, and dynamic interaction data. These strategy data contain a policy network that can output the expected reward value based on the current state and possible behavior choices.

[0096] In practical applications, when a humanoid robot is in a specific state, the policy network predicts the possible reward value of each action based on the state and the feasible actions. Then, based on these predicted reward values, the humanoid robot chooses to perform the action that can obtain the maximum reward. As the robot continues to perform actions and interact with the environment, it collects more actual motion feedback information. This feedback information will be used to further optimize and adjust the policy network so that it can better adapt to different task requirements and environmental changes. Through continuous iteration and optimization of this process, the policy network gradually learns how to make the best decisions in various situations, thereby generating the optimal dual-arm motion trajectory. These motion trajectories take into account the influence of causality and dynamic interaction, and can meet the requirements of safety, efficiency and other aspects, making the dual-arm movement of the humanoid robot more natural, coordinated and efficient.

[0097] Finally, it also includes safety and efficiency evaluation, feedback and iterative optimization, as well as user interaction and customization of the obtained humanoid robot's dual-arm motion trajectory. Among them, safety and efficiency evaluation means that the specific indicators of safety evaluation include avoiding collision with obstacles, maintaining body balance, etc., and the standards of efficiency evaluation include the time taken to complete the task, etc., to evaluate the safety and efficiency of the motion trajectory in real time to ensure that potential risks in the motion process are effectively controlled; feedback and iterative optimization means that during the actual motion execution process, feedback information is collected through sensors and other devices, and the feedback information is used for iterative optimization of the model, adjusting model parameters, and updating the motion trajectory generation strategy to improve the adaptability and robustness of dual-arm motion; user interaction and customization means providing a user interaction interface and voice commands, allowing users to input relevant parameters and constraints according to specific task requirements. The system customizes the motion trajectory of the dual arms according to the user input to achieve personalized motion control.

[0098] In summary, the embodiments of the present invention have the following beneficial effects:

[0099] 1) A dual-arm motion trajectory generation method that can adapt to complex environmental changes is provided, which improves the adaptability and robustness of the humanoid robot.

[0100] 2) Through causal analysis and multi-agent interactive simulation, a deep understanding of the movement of both arms is achieved, making the movement trajectory more reasonable and natural.

[0101] 3) The dynamic reward mechanism and reinforcement learning algorithm were introduced to improve the efficiency, safety and multi-agent interactivity of motion trajectory generation.

[0102] 4) Through user interaction and customized design, personalized motion trajectory generation to meet different tasks and user needs is achieved.

[0103] See also Figure 2The embodiment of the present application also provides a humanoid robot dual-arm motion trajectory generation system, which can implement the above-mentioned humanoid robot dual-arm motion trajectory generation method, and the system includes:

[0104] The first module 201 is used to obtain motion data of the humanoid robot's arms and interaction data of the humanoid robot's arms and perform data preprocessing to obtain preprocessed feature data of the humanoid robot's arms;

[0105] The second module 202 is used to perform causal data analysis on the pre-processed humanoid robot dual-arm feature data based on the causal reasoning engine model to obtain causal relationship data;

[0106] The third module 203 is used to perform data interaction training processing on the pre-processed humanoid robot dual-arm feature data based on the multi-agent interaction model to obtain multi-agent dynamic interaction data;

[0107] The fourth module 204 is used to combine the causal relationship data with the multi-agent dynamic interaction data to control the movement of the humanoid robot's arms and generate the movement trajectory of the humanoid robot's arms.

[0108] It can be understood that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0109] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for generating a motion trajectory of a humanoid robot's arms, characterized in that: The method comprises the following steps: Acquire motion data of the humanoid robot's arms and interaction data of the humanoid robot's arms and perform data preprocessing to obtain preprocessed feature data of the humanoid robot's arms; Based on the causal reasoning engine model, causal data analysis and processing are performed on the preprocessed humanoid robot dual-arm feature data to obtain causal relationship data; Based on the multi-agent interaction model, data interaction training is performed on the pre-processed humanoid robot dual-arm feature data to obtain multi-agent dynamic interaction data, which includes: Inputting the preprocessed humanoid robot dual-arm feature data into the multi-agent interaction model; The expression of the multi-agent interaction model is specifically as follows: ； In the above formula, Indicates Layer Agent The characteristic vector of Indicates Layer Agent The characteristic vector of Indicates Layer Agent The characteristic vector of Representing an Agent Neighbors, Representing an Agent With Agent In the The trainable weights of the layer, Indicates Layer trainable bias, and represents the adjustable coefficient, represents a nonlinear activation function; Performing feature vector conversion processing on the preprocessed humanoid robot dual-arm feature data to obtain a humanoid robot dual-arm feature data vector; Determining the number of neighbors of the humanoid robots according to the relationship between the humanoid robots; Combined with the number of neighbors of the humanoid robot, interactive calculation is performed on the feature data vectors of the two arms of the humanoid robot to obtain the multi-agent dynamic interaction data, which includes: Determining the trainable weights of the multi-agent interaction model and the bias term of the multi-agent interaction model; The humanoid robot dual-arm feature data vector is multiplied by the trainable weight and then divided by the number of neighbors of the humanoid robot, and then normalized and summed to obtain a weighted feature data vector of the humanoid robot dual-arm; The bias term is introduced, and the weighted feature data vectors of the humanoid robot arms are interactively calculated through a nonlinear activation function to obtain multi-agent dynamic interaction data; The causal relationship data is combined with the multi-agent dynamic interaction data to control the movement of the humanoid robot's arms and generate the movement trajectory of the humanoid robot's arms.

2. The method according to claim 1, characterized in that The step of acquiring the motion data of the humanoid robot's arms and the interaction data of the humanoid robot's arms and performing data preprocessing to obtain the preprocessed feature data of the humanoid robot's arms includes: Acquire motion data of both arms of the humanoid robot through sensors, wherein the motion data includes joint angle sensor data, velocity sensor data, and acceleration sensor data; Acquire interaction data between humanoid robots and interaction data between humanoid robots and intelligent devices, and construct interaction data of both arms of the humanoid robots; Integrating the motion data of the two arms of the humanoid robot and the interaction data of the two arms of the humanoid robot to obtain feature data of the two arms of the humanoid robot; The feature data of both arms of the humanoid robot are standardized and subjected to data feature extraction and fusion processing to obtain the preprocessed feature data of both arms of the humanoid robot.

3. The method according to claim 2, characterized in that The step of normalizing the feature data of both arms of the humanoid robot and extracting and fusing the feature data to obtain the preprocessed feature data of both arms of the humanoid robot includes: Performing standardization processing on the feature data of both arms of the humanoid robot to obtain standardized feature data of both arms of the humanoid robot, wherein the standardized feature data of both arms of the humanoid robot includes structured data and unstructured data; Performing data fusion processing on the structured data by using a feature hash algorithm to obtain fused structured data; Performing feature extraction and fusion processing on the unstructured data through a deep learning model to obtain fused unstructured data; The fused structured data and the fused unstructured data are integrated to obtain the preprocessed humanoid robot dual-arm feature data.

4. The method according to claim 1, characterized in that: The causal data analysis and processing is performed on the pre-processed humanoid robot dual-arm feature data based on the causal reasoning engine model to obtain causal relationship data, including: Inputting the preprocessed humanoid robot dual-arm feature data into the causal reasoning engine model, wherein the causal reasoning engine model includes a graph neural network, a convolutional neural network, a recurrent neural network, and a causal inference unit; Based on the graph neural network of the causal reasoning engine model, the structured data in the preprocessed feature data of the humanoid robot arms are processed to capture the dynamic change relationship to obtain structured dynamic change data; Based on the convolutional neural network and the recurrent neural network of the causal reasoning engine model, the unstructured data in the preprocessed feature data of the humanoid robot arms are subjected to spatial feature extraction and temporal feature extraction to obtain unstructured spatial feature data and unstructured temporal feature data; Based on the causal inference unit of the causal inference engine model, causal variable inference is performed on the structured dynamically changing data, the unstructured spatial feature data and the unstructured temporal feature data to obtain the causal relationship data.

5. The method according to claim 1, characterized in that The step of combining the causal relationship data with the multi-agent dynamic interaction data to control the movement of the two arms of the humanoid robot and generate the movement trajectory of the two arms of the humanoid robot includes: Training the adaptive behavior synthesis component on the causal relationship data and the multi-agent dynamic interaction data through reinforcement learning to obtain behavior strategy data; The movement of the two arms of the humanoid robot is controlled according to the behavior strategy data to generate the movement trajectory of the two arms of the humanoid robot.

6. The method according to claim 5, characterized in that The adaptive behavior synthesis component represents a dynamic reward function, and the expression of the dynamic reward function is specifically as follows: ； In the above formula, represents the efficiency parameter, Represents the security parameter, represents the multi-agent interaction parameter, Indicates the task completion parameters, represents the adaptability parameter, , , , , represents the adjustable coefficient, represents the dynamic reward function, represents the state of the agent, Represents the actions of the agent.

7. A system for generating a motion trajectory of a humanoid robot's two arms as claimed in claim 1, characterized in that: The system comprises: The first module is used to obtain motion data of the humanoid robot's arms and interaction data of the humanoid robot's arms and perform data preprocessing to obtain preprocessed feature data of the humanoid robot's arms; The second module is used to perform causal data analysis on the pre-processed humanoid robot dual-arm feature data based on a causal reasoning engine model to obtain causal relationship data; The third module is used to perform data interaction training processing on the pre-processed humanoid robot dual-arm feature data based on a multi-agent interaction model to obtain multi-agent dynamic interaction data; The fourth module is used to combine the causal relationship data with the multi-agent dynamic interaction data to control the movement of the humanoid robot's arms and generate the movement trajectory of the humanoid robot's arms.

Images