Interference-based control method, device and medium for robot dexterous hand

Abstract

This invention relates to the field of dexterous hands in humanoid robots, and more particularly to a method, device, and medium for controlling a robot dexterous hand based on interferometric identification. The method includes: Step S1: Obtaining a hand shape sequence based on task information parsing; Step S2: Generating corresponding joint states for each hand shape in the hand shape sequence, obtaining a joint state vector for each hand shape; Step S3: Forming a hand shape group by grouping two adjacent hand shapes in the hand shape sequence; Step S4: Using the joint state vectors of the two hand shapes in each hand shape group as boundary constraints, and filtering them using an interferometric identification model, generating multiple joint transition state vectors; Step S5: Sequentially concatenating all joint state vectors and joint transition state vectors to obtain a control vector for the dexterous hand; Step S6: Controlling the joint movements of the dexterous hand based on the obtained control vector. Compared with existing technologies, this invention has advantages such as eliminating the need for complex spatial state calculations.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of dexterous hands of humanoid robots, and in particular to a robot dexterous hand control method and device based on interference identification and a medium. BACKGROUND

[0002] Traditional robot control relies on pre-written rules, which greatly restricts the development of robots. In recent years, the development of artificial intelligence has provided more flexible possibilities for robot motion control. Based on artificial intelligence, the control method does not need to pre-write rules for robot control, but can automatically generate control parameters for each joint, so that the number of joints is not constrained in a very low range during robot development. Based on this, some robots increase the number of joints to form new robot shapes.

[0003] Such new shape robots can perform more flexible tasks due to the larger number of joints, and are also called dexterous hands. For example, Chinese patent CN107081777A discloses a humanoid dexterous hand based on a shape memory alloy flexible intelligent digital composite structure. The dexterous hand provided by the application can be used in the structure of sign language display. The number of joints of some dexterous hands even exceeds 20.

[0004] The motion control of existing dexterous hands relies on visual demonstration and spatial calculation. For example, Chinese patent CN119388470A discloses a dexterous hand control method, a dexterous hand control device, a dexterous hand and a robot. According to the first mapping data and the remote operation data, the target angle of the driving motor is determined, and the target angle of the driving motor is sent to the driving motor to make the driving motor rotate to the target angle, so that the driving motor drives the joint to move, so that the finger moves to the target position or the joint rotates to the target angle, realizing the control of the joint, and then controlling the dexterous hand to realize the action of remote operation. When controlling the dexterous hand, complex spatial calculation is needed to realize the rotation angle of each joint. In addition, Chinese patent CN120816500A discloses a dexterous hand general grasping method and device based on single-view visual and demonstration optimization. The object grasping control of the dexterous hand is realized by combining artificial teaching with spatial constraint.

[0005] In summary, the existing control method of the dexterous hand relies on the space constraint successive approximation method to control the rotation angle of the joint. On the one hand, the spatial constraint calculation requires extremely strong computing power, and it is difficult to realize autonomous motion control of the dexterous hand on a low-power platform. On the other hand, the motion control of the entire dexterous hand is relatively rigid and not smooth. SUMMARY

[0006] The application aims to provide a robot dexterous hand control method, device and medium based on interference identification, to solve the problem of low-power platform-based dexterous hand self-control without demonstration learning.

[0007] The application aims to provide a robot dexterous hand control method, device and medium based on interference identification, to solve the problem of low-power platform-based dexterous hand self-control without demonstration learning.

[0008] A robot dexterous hand control method based on interference identification, comprising:

[0009] Step S1: obtaining task information and obtaining hand shape sequence based on task information analysis;

[0010] Step S2: generating corresponding joint state for each hand shape in the hand shape sequence, to obtain joint state vector corresponding to each hand shape;

[0011] Step S3: grouping all two adjacent hand shapes in the hand shape sequence into a hand shape group;

[0012] Step S4: taking joint state vectors of two hand shapes in each hand shape group as boundary constraint conditions, and combining interference identification model filtering to generate multiple joint transition state vectors;

[0013] Step S5: sequentially splicing all joint state vectors and joint transition state vectors to obtain control vector of the dexterous hand;

[0014] Step S6: controlling joint action of the dexterous hand based on the obtained control vector.

[0015] The joint state includes angle and timestamp, and the joint state vector is composed of angles and timestamps of each joint.

[0016] The step S4 comprises:

[0017] Step S4-1: taking angles of joint state vectors of two hand shapes in each hand shape group as boundary constraint conditions;

[0018] Step S4-2: taking extreme value of difference between boundary angles of each joint in the boundary constraint conditions as first time allocation weight of each hand shape group;

[0019] Step S4-3: generating allocation time of each hand shape group based on the first time allocation weight of each hand shape group and combining task total time consumption constraint;

[0020] Step S4-4: generating multiple joint transition state vectors based on allocation time of each hand shape group and boundary constraint conditions, and combining interference identification model filtering.

[0021] The process of generating joint transition state vectors for a single hand shape group in step S4-4 comprises:

[0022] Step S4-4-1: Determine the number of intermediate state vectors based on the allocation time of the hand group;

[0023] Step S4-4-2: Using the angles in the boundary constraints of the hand group as end values, generate multiple intermediate state vectors with equal angle differences between the end values. The intermediate state vectors are composed of the angles of each joint.

[0024] Step S4-4-3: Concatenate all two adjacent intermediate state vectors to obtain the first feature vector, input it into the trained interference identification model, and obtain the interference identification result. The interference identification model is a binary classification result, and the interference identification result is either the presence of mechanical interference or the absence of mechanical interference.

[0025] Step S4-4-4: Determine whether there is interference between any two adjacent intermediate state vectors. If the determination result is that mechanical interference exists, proceed to step S4-4-5; otherwise, proceed to step S4-4-6.

[0026] Step S4-4-5: Regenerate the intermediate state vector and return to step S4-4-3;

[0027] Step S4-4-6: Generate joint transition state vectors based on all current intermediate state vectors.

[0028] Step S4-4-5 includes:

[0029] Step S4-4-5-1: Take the two intermediate state vectors corresponding to the first feature vector that indicates the presence of mechanical interference and all subsequent intermediate state vectors as intermediate state vectors to be optimized.

[0030] Step S4-4-5-2: Regenerate all intermediate state vectors to be optimized using the optimization algorithm, and return to step S4-4-3.

[0031] The optimization algorithm is either a genetic algorithm or a particle swarm optimization algorithm.

[0032] Step S4-4-6 includes:

[0033] Step S4-4-6-1: Calculate the sum angle difference of each adjacent intermediate state vector:

[0034]

[0035] in: For the first i The intermediate state vector and the first intermediate state vector i +1 sum of angle differences of intermediate state vectors N The number of joints in a dexterous hand. For the firsti the angle value of the jth joint in the ith intermediate state vector, k the angle value of the jth joint in the ith intermediate state vector, i k the angle value of the jth joint in the ith intermediate state vector, the angle value of the jth joint in the joint state vector of the last hand shape in the hand shape group, k the angle value of the jth joint in the joint state vector of the first hand shape in the hand shape group, the angle value of the jth joint in the joint state vector of the first hand shape in the hand shape group; k

[0036] Step S4-4-6-2: Based on the calculated sum angle difference of each adjacent intermediate state vector, the time length corresponding to each adjacent intermediate state vector is generated in combination with the allocation time of the hand shape group:

[0037]

[0038] wherein: the time length between the ith intermediate state vector and the (i+1)th intermediate state vector, when i=1, the time length between the joint state vector of the last hand shape in the hand shape group and the 1th intermediate state vector, when i= i -1, the time length between the last intermediate state vector and the joint state vector of the last hand shape in the hand shape group, i the number of sum angle differences participating in the calculation; M M

[0039] Step S4-4-6-3: Based on the time length of each adjacent intermediate state vector, the time stamp of each intermediate state vector is generated, and the obtained time stamp and each intermediate state vector are combined to obtain a joint transition state vector, wherein the joint transition state vector is composed of the angle of each joint and the time stamp.

[0040] The positive samples in the interference identification model training include:

[0041] The first positive sample is obtained based on the 3D camera collecting human hand actions;

[0042] The second positive sample is obtained by simulation editing;

[0043] The negative sample in the interference identification model training is obtained by simulation editing.

[0044] A robot dexterous hand control device based on interference identification, comprising a memory, a processor, and a program stored in the memory, wherein the processor executes the program to realize the method as described above.

[0045] ​​​​​A storage medium having stored thereon a program which, when executed, implements the method as described above.

[0046] Compared with the prior art, the present application has the following beneficial effects:

[0047] 1. By designing to generate a plurality of transition state vectors between each two hand shape joint state vectors in batches, and performing mechanical interference verification based on the joint transition state vectors, the dexterous hand can be controlled autonomously without demonstration learning based on a low computing power platform without complex spatial calculation.

[0048] 2. The use of the first time allocation weight to obtain the allocation time of each hand shape group can improve the coordination of the overall motion.

[0049] 3. The intermediate state vector does not contain a timestamp, and the two intermediate state vectors are spliced and used as the input of the interference identification model. On the one hand, the absence of a timestamp allows the interference identification model to focus on the joint angle itself, avoiding interference from useless information. On the other hand, compared to directly performing interference identification on a single intermediate state vector, the time interval between adjacent intermediate state vectors can be improved, thereby reducing the number of intermediate state vectors and greatly reducing the computing power requirement. On the other hand, since the model input is unified, the difference between adjacent hand shapes in the hand shape sequence can not be required, and the number of generated joint transition state vectors between any two hand shape groups can not be constrained, thereby greatly improving the degree of freedom and adaptability, and also ensuring that the time interval between adjacent transition vectors is within a suitable range, thereby improving the classification accuracy of the interference identification model.

[0050] 4. The use of the normalized total angle difference ratio to allocate the time length between the joint transition state vectors can improve the coordination and smoothness of the overall motion. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 The figure is a schematic diagram of the main steps of the method of the present application.

[0052] Figure 2 The figure is a schematic diagram of the control vector.

[0053] In the figure: 101, joint state vector; 102, joint transition state vector. DETAILED DESCRIPTION

[0054] The present application will be described in detail below in conjunction with the drawings and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present application, and detailed implementation methods and specific operation processes are given, but the protection scope of the present application is not limited to the following embodiments.

[0055] A robot dexterous hand control method based on interference identification is proposed. By designing a method to generate multiple transition state vectors in batches between the joint state vectors of every two hand shapes, and performing mechanical interference verification based on the joint transition state vectors, a dexterous hand can achieve teaching-free learning and autonomous control on a low-computing-power platform without the need for complex spatial calculations.

[0056] like Figure 1 As shown, it includes:

[0057] Step S1: Obtain task information and parse the hand shape sequence based on the task information;

[0058] Specific task information can be directly input or generated through some coarse control algorithms. In this embodiment, taking sign language teaching as an example, after the user inputs text, the sign language translation software automatically translates it into multiple gestures, including hand shape and position. This application focuses on the hand shape, while the position is completed by the arm joint.

[0059] All the hand shapes in a gesture, arranged in order, form a hand shape sequence.

[0060] Step S2: For each hand shape in the hand shape sequence, generate the corresponding joint states to obtain the joint state vector for each hand shape;

[0061] In this embodiment, the joint state includes angle and timestamp. The joint state vector is composed of the angle and timestamp of each joint. In the initial state, the timestamps are all null values ​​and are assigned by the subsequent process. In addition, under normal circumstances, the task information also includes a total time constraint, which represents the total time required to complete the entire hand shape sequence.

[0062] Step S3: Combine any two adjacent hand shapes in the hand shape sequence into a hand shape group;

[0063] Step S4: Using the joint state vectors of the two hand shapes in each hand shape group as boundary constraints, and combining them with the interference identification model for filtering, multiple joint transition state vectors are generated, specifically including:

[0064] Step S4-1: Use the angle of the joint state vector of the two hand shapes in each hand shape group as the boundary constraint condition;

[0065] Step S4-2: Use the extreme value of the difference between the boundary angles of each joint in the boundary constraints as the first time weight for each hand shape group;

[0066] Step S4-3: Based on the first time allocation weight of each hand shape group and combined with the total task time constraint, generate the allocation time of each hand shape group. At this time, after obtaining the allocation time, the timestamp information of the joint state of each hand shape in the hand shape sequence can be improved.

[0067] Step S4-4: Based on the allocation time and boundary constraints of each hand shape group, and combined with the interference identification model filtering, multiple joint transition state vectors are generated. The process of generating a joint transition state vector for a single hand shape group includes:

[0068] Step S4-4-1: Determine the number of intermediate state vectors based on the allocation time of the hand group;

[0069] Step S4-4-2: Using the angles in the boundary constraints of the hand group as end values, generate multiple intermediate state vectors with equal angle differences between the end values. The intermediate state vectors are composed of the angles of each joint.

[0070] Step S4-4-3: Concatenate all two adjacent intermediate state vectors to obtain the first feature vector, input it into the trained interference identification model, and obtain the interference identification result. The interference identification model is a binary classification result, and the interference identification result is either the presence of mechanical interference or the absence of mechanical interference.

[0071] Positive samples used in training the interference identification model include:

[0072] First positive sample: obtained based on human hand movements captured by a 3D camera;

[0073] The second positive sample: obtained through simulation editing;

[0074] The negative samples used in training the interference identification model were obtained through simulation editing.

[0075] Furthermore, the time interval between the joint state vectors of positive and negative samples during the training of the interference identification model should be similar to the time interval between the pre-configured joint transition state vectors.

[0076] Step S4-4-4: Determine whether there is interference between any two adjacent intermediate state vectors. If the determination result is that mechanical interference exists, proceed to step S4-4-5; otherwise, proceed to step S4-4-6.

[0077] Step S4-4-5: Regenerate the intermediate state vector and return to step S4-4-3;

[0078] In this embodiment, step S4-4-5 includes:

[0079] Step S4-4-5-1: Take the two intermediate state vectors corresponding to the first feature vector that indicates the presence of mechanical interference and all subsequent intermediate state vectors as intermediate state vectors to be optimized.

[0080] Step S4-4-5-2: Regenerate all intermediate state vectors to be optimized using the optimization algorithm, and return to step S4-4-3.

[0081] Generally, optimization algorithms can employ genetic algorithms or particle swarm optimization, and a certain degree of diversity should be ensured to avoid getting trapped in local optima.

[0082] Step S4-4-6: Generate joint transition state vectors based on all current intermediate state vectors, specifically including:

[0083] Step S4-4-6-1: Calculate the sum angle difference of each adjacent intermediate state vector:

[0084]

[0085] in: For the first i The intermediate state vector and the first intermediate state vector i +1 sum of angle differences of intermediate state vectors N The number of joints in a dexterous hand. For the first i +1 intermediate state vectors k The angle values ​​of each joint. For the first i The th intermediate state vector k The angle values ​​of each joint. For the joint state vector of the next hand shape in the corresponding hand shape group, the first... k The angle values ​​of each joint. For the joint state vector of the previous hand shape in the corresponding hand shape group, the first... k The angle values ​​of each joint;

[0086] Step S4-4-6-2: Based on the calculated sum of angle differences between adjacent intermediate state vectors, and combined with the allocation time of the hand shape group, generate the duration corresponding to each adjacent intermediate state vector:

[0087]

[0088] in: For the first i The intermediate state vector and the first intermediate state vector i The duration between +1 intermediate state vectors, when i=1, is taken as the duration between the joint state vector of the next hand shape in the corresponding hand shape group and the first intermediate state vector; when i= M When -1, the time interval between the last intermediate state vector and the joint state vector of the next hand shape in the corresponding hand shape group is taken. M This represents the number of angle differences involved in the calculation.

[0089] Setting the duration using this method allows for adaptive time allocation based on the amount of movement, ensuring smooth and coordinated movements. It uses the proportion of the angle change in the current step relative to the total angle change in the entire transition process; transitions with large joint angle changes are automatically allocated more time, while those with smaller changes are allocated less. This avoids the problems that can arise from mechanically distributing time evenly: movements that are too fast in the small amplitude phases requiring fine adjustments, and too slow in the large amplitude phases. The result is that the speed of the entire hand transition movement is adaptive, non-uniform but highly coordinated, closer to the natural laws of human hand movement, greatly improving the smoothness and anthropomorphism of the movement.

[0090] Furthermore, it ensures the precise satisfaction of the total time constraint. Although each duration is allocated proportionally, normalization based on the total change ensures that the sum of the durations of all adjacent vectors is strictly equal to the allocated time for the entire hand gesture group. This guarantees that the macroscopic time planning of the entire action sequence is precisely controllable and can strictly meet the total time requirements of the task. The algorithm achieves local smoothness without violating global time constraints.

[0091] Finally, normalization achieves a fair trade-off across joints. On one hand, it eliminates the influence of dimensions; different joints may have different ranges of motion and units of measurement. Normalization unifies the angular changes of all joints to the same scale, allowing the contributions of each joint to be added fairly. A joint that rotates 10 degrees but has a total stroke of only 15 degrees will contribute far more than another joint that rotates 10 degrees but has a total stroke of 180 degrees. On the other hand, it focuses on the importance of relative motion, paying attention to the relative magnitude of change of each joint within its own range of motion, rather than absolute angle values. This aligns more with physical intuition; a tiny tremor in one joint may be just as important as a large movement in a less sensitive joint, and normalization captures this difference.

[0092] Step S4-4-6-3: Generate the timestamp of each intermediate state vector based on the duration of each adjacent intermediate state vector, and combine the obtained timestamps and each intermediate state vector to obtain the joint transition state vector, wherein the joint transition state vector is composed of the angle and timestamp of each joint.

[0093] In the above scheme, "model classification and judgment" replaces "complex numerical calculation," and "intelligent time allocation" and "dynamic path optimization" ensure smooth and interference-free movements, ultimately achieving the goal of dexterous, natural, and reliable autonomous control of a robot hand on a low-computing-power platform. Specifically, this is reflected in:

[0094] 1. Significantly reduce computing power requirements and achieve efficient control on low-computing-power platforms.

[0095] Traditional methods rely on complex spatial geometric calculations to avoid mechanical interference, resulting in a heavy computational burden. This solution introduces an interference identification model, transforming the complex spatial computation problem into an efficient binary classification problem. The model only needs to determine whether interference is present or absent in the spliced ​​joint state vectors, greatly simplifying the computation process and making it possible to achieve real-time, autonomous dexterous hand control on embedded platforms or mobile robot bodies with limited computing power.

[0096] 2. Improve the smoothness and coordination of motion planning.

[0097] Intelligent time allocation: By using the extreme values ​​of the changes in the boundary angles of each joint as the weights for time allocation, it ensures that the joints with the largest changes have enough time to complete the movement. This avoids abrupt or uncoordinated overall movements caused by a single joint requiring a large range of rapid motion, making the transitions between hand shapes more natural and smooth.

[0098] Dynamic timestamp allocation: When generating the final joint transition state vector, the time is not simply allocated evenly, but rather dynamically allocated based on the normalized sum angle difference between adjacent intermediate state vectors. This means that transition phases with large joint angle changes are allocated more time, while phases with small changes are allocated less time. This "variable speed" motion planning further enhances the overall smoothness and anthropomorphic coordination of the movement.

[0099] 3. Optimize the efficiency and accuracy of interferometric detection.

[0100] Focusing on key information: The intermediate state vector contains only joint angles and no timestamps. This allows the interference identification model to focus on the joint spatial configuration itself, eliminating the interference of time information and helping to improve the accuracy and generalization ability of the model's judgment.

[0101] Detecting "motion process" rather than "static state": A key innovation of this method is that it concatenates the vectors of two adjacent intermediate states and uses them as model input. Instead of detecting interference from a single static hand shape, it detects whether interference will occur during the brief process of moving from one state to the next. This is more consistent with actual physical processes and can effectively identify situations where there is no interference in the static state but collisions will occur along the motion path, greatly improving the practicality and accuracy of the detection.

[0102] 4. Enhance the robustness and adaptability of the algorithm.

[0103] Local optimization strategy: When interference is detected, instead of completely rejecting the entire algorithm and starting over, the algorithm optimizes only the intermediate state vectors from the first node where interference occurs. This strategy is computationally efficient, avoids the massive computational cost of global search, and allows the algorithm to quickly find interference-free feasible paths, making it more robust.

[0104] Process standardization: Step S4 breaks down the complex hand shape transition problem into a series of standardized sub-steps. This modular design enables the algorithm to handle the transition between any two hand shapes, regardless of specific hand shape differences, and provides strong versatility and adaptability.

[0105] Step S5: Concatenate all joint state vectors and joint transition state vectors in sequence to obtain the control vector of the dexterous hand, such as... Figure 2 As shown, when there are 3 hand shapes in a hand shape sequence, the joint transition state vector 102 and the joint state vector 101 are arranged in order, and all joint transition state vectors 102 located between a certain hand shape group are located between the two joint state vectors 101 of that hand shape group.

[0106] Step S6: Control the joint movements of the dexterous hand based on the obtained control vectors.

[0107] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A robot dexterous hand control method based on interference identification, characterized by, The method comprises the following steps: Step S1: obtaining task information and parsing hand shape sequences based on the task information; Step S2: generating corresponding joint state vectors for each hand shape in the hand shape sequences; Step S3: grouping two adjacent hand shapes in the hand shape sequences into a hand shape group; Step S4: taking the joint state vectors of the two hand shapes in each hand shape group as boundary constraint conditions, and combining interference identification model filtering to generate a plurality of joint transition state vectors; Step S5: sequentially splicing all the joint state vectors and the joint transition state vectors to obtain a control vector of the dexterous hand; Step S6: controlling the joint actions of the dexterous hand based on the obtained control vector. The joint state comprises an angle and a timestamp, and the joint state vector is composed of the angles and the timestamps of the joints. The step S4 comprises the following steps: Step S4-1: taking the angles of the joint state vectors of the two hand shapes in each hand shape group as boundary constraint conditions; Step S4-2: taking the extreme value of the difference between the boundary angles of the joints in the boundary constraint conditions as a first time allocation weight of each hand shape group; Step S4-3: generating an allocation time of each hand shape group based on the first time allocation weight of each hand shape group and combining a total time consumption constraint of the task; Step S4-4: generating a plurality of joint transition state vectors based on the allocation time of each hand shape group and the boundary constraint conditions and combining interference identification model filtering; The process of generating the joint transition state vectors for a single hand shape group in the step S4-4 comprises the following steps: Step S4-4-1: determining the number of intermediate state vectors based on the allocation time of the hand shape group; Step S4-4-2: generating a plurality of intermediate state vectors in an equal angle difference manner between the end values of the angles in the boundary constraint conditions of the hand shape group, wherein the intermediate state vectors are composed of the angles of the joints; Step S4-4-3: splicing the first feature vector obtained by splicing all adjacent two intermediate state vectors, inputting the first feature vector into the trained interference identification model, and obtaining an interference identification result, wherein the interference identification model is a binary classification result, and the interference identification result is mechanical interference or no mechanical interference; Step S4-4-4: determining whether the interference identification result of any adjacent two intermediate state vectors is mechanical interference, if yes, executing step S4-4-5, otherwise, executing step S4-4-6; Step S4-4-5: regenerating the intermediate state vectors and returning to step S4-4-3; Step S4-4-6: generating the joint transition state vectors based on the current all intermediate state vectors.

2. The robot dexterous hand control method based on interference identification according to claim 1, wherein, The step S4-4-5 comprises the following steps: Step S4-4-5-1: taking the two intermediate state vectors corresponding to the first feature vector with the mechanical interference as the to-be-optimized intermediate state vectors and all the intermediate state vectors after the first feature vector; Step S4-4-5-2: regenerating all the to-be-optimized intermediate state vectors by using an optimization algorithm, and returning to step S4-4-3.

3. The robot dexterous hand control method based on interference identification according to claim 2, characterized in that, The optimization algorithm is a genetic algorithm or a particle swarm algorithm.

4. The robot dexterous hand control method based on interference identification according to claim 2, wherein, The step S4-4-6 comprises the following steps: Step S4-4-6-1: Calculate the sum angle difference of each adjacent intermediate state vector respectively: ； wherein: is the sum angle difference of the first i intermediate state vector and the first i +1 intermediate state vector, N is the number of joints of the dexterous hand, is the angle value of the first i joint in the first k +1 intermediate state vector, is the angle value of the first i joint in the first k intermediate state vector, is the angle value of the first k joint in the joint state vector of the latter hand shape in the corresponding hand shape group, is the angle value of the first k joint in the joint state vector of the former hand shape in the corresponding hand shape group; Step S4-4-6-2: Based on the calculated sum angle difference of each adjacent intermediate state vector, generate the time length corresponding to each adjacent intermediate state vector in combination with the allocation time of the hand shape group: ; in: For the first i The intermediate state vector and the first intermediate state vector i The duration between +1 intermediate state vectors, when i=1, is taken as the duration between the joint state vector of the next hand shape in the corresponding hand shape group and the first intermediate state vector; when i= M When -1, the time interval between the last intermediate state vector and the joint state vector of the next hand shape in the corresponding hand shape group is taken. M This represents the number of angle differences involved in the calculation. Step S4-4-6-3: Based on the time length of each adjacent intermediate state vector, generate the time stamp of each intermediate state vector, and combine the obtained time stamp and each intermediate state vector to obtain a joint transition state vector, wherein the joint transition state vector is composed of the angle of each joint and the time stamp.

5. The robot dexterous hand control method based on interference identification according to claim 1, wherein, The positive samples during training of the interference identification model include: The first positive sample is obtained based on a 3D camera collecting human hand actions; The second positive sample is obtained by simulation editing; The negative sample during training of the interference identification model is obtained by simulation editing.

6. A robot dexterous hand control device based on interference identification, comprising a memory, a processor, and a program stored in the memory, wherein the program comprises the following steps of: The processor implements the method of any one of claims 1-5 when executing the program.

7. A storage medium having stored thereon a program, characterized by The program is executed to implement the method of any one of claims 1-5.

Images