Nursing robot imitation learning method and device and medium thereof
By utilizing neural network mutual information estimation and slice Wasserstein distance to select the optimal trajectory in the imitation learning of nursing robots, the problem of insufficient number of expert demonstration samples in existing technologies is solved, and efficient imitation learning effect is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUILIN UNIV OF ELECTRONIC TECH
- Filing Date
- 2026-01-19
- Publication Date
- 2026-04-10
AI Technical Summary
Existing robot imitation learning methods heavily rely on the number of expert demonstration samples. When there are insufficient expert demonstration samples, the success rate of imitation learning decreases, and it becomes impossible to learn and imitate trajectories from a single expert demonstration sample.
A nursing robot imitation learning method is adopted. By initializing environmental parameters, an imitation trajectory is generated. The mutual information term is calculated using a neural network mutual information estimation method and transformed into an internal reward term of the policy network. The network parameters are updated by combining a soft critic algorithm and the optimal trajectory is selected based on the slice Wasserstein distance, thereby reducing the dependence on expert demonstration samples.
It effectively reduces the robot's dependence on expert samples in imitation learning, reduces the computational load of trajectory matching, improves computational efficiency, and enables the learning of diverse imitation trajectories from a single expert demonstration.
Smart Images
Figure CN121821376A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of robotics, and in particular to a method, device and medium for imitation learning of a nursing robot. Background Technology
[0002] In recent years, China's aging population has continued to deepen, and the total number of professionals in elderly care and related fields is insufficient to meet societal needs. Nursing robots, as a crucial component of elderly care resources that uses technology to replace human labor, have been a focus of ongoing research efforts. However, traditional pre-programmed robots struggle to directly and smoothly perform nursing functions due to the unstructured and highly time-varying nature of nursing scenarios. To improve the intelligence level of nursing robots, teaching them the operational actions required for nursing tasks has become a key task. Imitation learning, as an important means of motion planning for nursing robots, allows them to learn operational tasks by observing expert demonstrations. This not only enables them to directly follow expert demonstrations to complete robot tasks but also allows them to generalize and apply this learning to similar robot tasks. It also demonstrates strong practicality for highly time-varying and unstructured nursing environments and has significant research value.
[0003] Existing robot imitation learning methods mainly fall into three categories: behavior cloning, inverse reinforcement learning, and generative adversarial learning. Behavior cloning primarily extracts effective information from expert demonstrations, enabling the agent to generate trajectories that match the expert demonstrations. Inverse reinforcement learning derives the reward function from expert demonstrations, then uses reinforcement learning to continuously explore and ultimately generate trajectories that approximate the expert demonstrations. Generative adversarial learning calculates the difference between the generated strategy and the expert demonstration using a discriminant function, reducing the difference through iterative processes until the generated strategy is indistinguishable from the expert demonstration in the discriminant function. All three methods heavily rely on the number of expert demonstration samples; insufficient sample size significantly reduces the success rate of imitation learning, making it impossible to learn and imitate trajectories from a single expert demonstration sample. Summary of the Invention
[0004] The technical problem to be solved by this invention is: in order to address the technical problem that existing technologies rely on the number of expert demonstration samples, which significantly reduces the success rate of imitation learning due to insufficient expert demonstration samples, and cannot learn and imitate trajectories from a single expert demonstration sample, this invention provides a nursing robot imitation learning method that enables nursing robots to learn from a single expert demonstration, effectively reducing the dependence of robot imitation learning on expert samples, greatly reducing the amount of computation for trajectory matching, and improving computational efficiency.
[0005] The technical solution adopted by this invention to solve its technical problem is: a nursing robot imitation learning method, the method comprising the following steps:
[0006] S1 initializes the parameters within the environment;
[0007] S2, sample and generate imitation trajectories from the policy network;
[0008] S3, set the maximum number of rounds or time steps, and collect and process the generated imitation trajectory based on experience;
[0009] S4, sample empirical tuples from the imitation trajectory, perturb the sampling to satisfy the action prior, ensure that the policy output does not deviate from the expert demonstration sample, and perform imitation constraints;
[0010] S5, based on the preset paradigm requirements, calculates the mutual information terms using a neural network mutual information estimation method;
[0011] S6. Based on the mutual information terms, the constraint signals are transformed into internal reward terms of the policy network, and pseudo-rewards are saved.
[0012] S7, add trajectory segments with pseudo-rewards to the experience feedback buffer;
[0013] S8, update the network parameters according to the soft critic algorithm and the preset paradigm requirements, and repeat steps S2 to S8 until the convergence condition is met;
[0014] S9 calculates the slice Wasserstein distance based on the imitation trajectory and expert demonstration samples, and selects the optimal trajectory from the imitation trajectories according to the slice Wasserstein distance.
[0015] Furthermore, specifically, in step S1, the parameters include the position and posture of the nursing robot, the position and posture of the target object, the position of obstacles in the environment, the physical posture information of the nursing robot, the strategy function parameter θ, and the Q-value function parameters ϕ and ψ.
[0016] Furthermore, specifically, in step S2, an initial policy z is sampled from expert demonstration samples, and the agent is guided to generate an imitation trajectory around a single expert demonstration based on the initial policy z.
[0017] Furthermore, specifically, in step S4, the disturbance sampling satisfies the following condition:
[0018]
[0019] in, Indicates the action at time t. The parameter is random strategy, Indicates the state as The action at that time, z represents the expert demonstration sample sampling information.
[0020] Furthermore, specifically, the internal reward term is defined as the sum of the results from two mutual information estimation methods, calculated using the following formula:
[0021]
[0022] in, MINE represents the mutual information estimation method in neural networks, where MINE stands for intrinsic reward. Indicates the initial policy. express The status of the nursing robot over time. This indicates the playback buffer.
[0023] Furthermore, specifically, the formula for calculating the slice Wasserstein distance is as follows:
[0024]
[0025] in, , Let X represent the probability distribution of random variables X and Y. Represents any direction vector. , This represents the projection of X and Y in that direction. Let pWortherstein distance represent the distance between one-dimensional distributions.
[0026] Furthermore, specifically, the pre-defined paradigm requirements include:
[0027] Paradigm 1 maximizes the mutual information between the random variable and the state information, and the calculation formula is as follows:
[0028] ;
[0029] Where I represents mutual information, S represents state, and Z represents random variable;
[0030] Paradigm 2 minimizes the mutual information between random variables and state-based action sequences, calculated as follows:
[0031] ;
[0032] Where A|S represents action A in state S;
[0033] Paradigm 3 maximizes the information entropy of a sequence of actions in a specific state, calculated using the following formula:
[0034] ;
[0035] Where H represents information entropy;
[0036] Paradigm 4 maximizes the mutual information between the imitation trajectory and the expert demonstration sample, calculated using the following formula:
[0037]
[0038] Where C represents the expert demonstration sample.
[0039] Furthermore, specifically, when the method is adapted to a real-world environment, it includes the following steps:
[0040] Extract the desired position of the joint angle corresponding to each position on the simulated trajectory in the simulation environment;
[0041] The robot is controlled to drive each joint to follow the corresponding desired position in sequence, thus enabling the movement of the nursing robot.
[0042] A computer device, comprising:
[0043] processor;
[0044] Memory, used to store executable instructions;
[0045] The processor is configured to read the executable instructions from the memory and execute the executable instructions to implement the nursing robot imitation learning method as described above.
[0046] A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, which, when executed by a processor, causes the processor to implement the nursing robot imitation learning method as described above.
[0047] The beneficial effects of this invention include:
[0048] 1. The nursing robot imitation learning method of the present invention utilizes a single expert demonstration sample to generate imitation trajectories with similar shapes and diversity around the single expert demonstration. The nursing robot learns from the single expert demonstration, which effectively reduces the dependence of robot imitation learning on expert samples.
[0049] 2. The nursing robot imitation learning method of the present invention uses the Wasserstein distance between the imitation trajectory and the slice of expert demonstration samples as the standard for evaluating the quality of the imitation trajectory. By decomposing the high-dimensional distribution estimate into a simpler estimate of multiple one-dimensional distributions, the computational amount of trajectory matching is greatly reduced and the computational efficiency is improved. Attached Figure Description
[0050] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0051] Figure 1 This is a schematic diagram of the method flow of Embodiment 1 of the present invention.
[0052] Figure 2 This is a schematic diagram of the computer device structure according to Embodiment 2 of the present invention.
[0053] In the diagram, 10 is a computer device; 1002 is a processor; 1004 is a memory; and 1006 is a transmission device. Detailed Implementation
[0054] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0055] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, features defined with "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.
[0056] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0057] Example 1
[0058] This application provides a method for imitation learning of a nursing robot, such as... Figure 1 As shown, the method includes the following steps:
[0059] S1, initialize the parameters in the environment; the parameters include the position and attitude of the nursing robot, the position and attitude of the target object, the position of obstacles in the environment, the physical attitude information of the nursing robot, the policy function parameter θ, and the Q-value function parameters ϕ and ψ.
[0060] S2, sample and generate imitation trajectories from the policy network; further, sample and obtain an initial policy z from expert demonstration samples, and guide the agent to generate imitation trajectories around a single expert demonstration according to the initial policy z. The generated imitation trajectories are similar to the single expert demonstration in form and have diversity.
[0061] S3 sets the maximum number of rounds or time steps to collect and process experience on the generated imitation trajectories. By setting the maximum number of rounds or time steps as training parameters, the data collection and constraint signal quantization on each imitation trajectory are limited, ensuring that sufficient experience samples are provided for the policy network.
[0062] S4, sample experience tuples from the imitation trajectory, perturb the sampling to satisfy the action prior, ensure that the policy output does not deviate from the expert demonstration sample, and apply imitation constraints; further, the perturbation sampling satisfies the following conditions:
[0063]
[0064] in, Indicates the action at time t. The parameter is random strategy, Indicates the state as The action at that time, z represents the expert demonstration sample sampling information.
[0065] S5, based on the preset paradigm requirements, calculates the mutual information terms using a neural network mutual information estimation method;
[0066] S6, based on the mutual information term, transform the constraint signal into the internal reward term of the policy network, preserving the spurious reward; further, the internal reward term is defined as the sum of the results of the two mutual information estimation methods, calculated using the following formula:
[0067]
[0068] in, MINE represents the mutual information estimation method in neural networks, where MINE stands for intrinsic reward. Indicates the initial policy. express The status of the nursing robot over time. This indicates the playback buffer.
[0069] S7, add trajectory segments with pseudo-rewards to the experience feedback buffer;
[0070] S8 updates the network parameters using the Soft Commentator Algorithm (SCA algorithm) and the preset paradigm requirements, and repeats steps S2 to S8 until the convergence condition is met. The convergence condition is that the similarity between pseudo-rewards remains unchanged, that is, no new effective imitation trajectories are generated.
[0071] S9. Calculate the slice Wasserstein distance based on the imitation trajectory and expert demonstration samples, and select the optimal trajectory from the imitation trajectories according to the slice Wasserstein distance. In other words, use the slice Wasserstein distance between the generated imitation trajectory and the expert demonstration samples as an evaluation metric to perform trajectory matching and select a suitable imitation trajectory from the generated imitation trajectories.
[0072] In this embodiment, the policy network refers to a deep neural network in a reinforcement learning framework, such as the SAC algorithm, used to determine what action an agent should take in a given state.
[0073] Furthermore, the slice Wasserstein distance (SWD) is derived from the Wasserstein distance, also known as the excavator distance, which measures the difference between two probability distributions and is defined as the minimum cost required to transform one distribution into another. However, because the Wasserstein distance employs a linear programming strategy, it becomes difficult to compute in linear applications, especially when dealing with high-dimensional data such as the trajectory data of nursing robots. The slice Wasserstein distance (SWD) builds upon the Wasserstein distance by decomposing the challenging estimation of high-dimensional distributions into simpler estimates of multiple one-dimensional distributions.
[0074] The SWD method first randomly selects a direction, i.e., a one-dimensional subspace, and orthogonally projects the high-dimensional information of the two trajectories onto this one-dimensional line. Next, it calculates the Wasserstein distance between the two projected one-dimensional distributions. For one-dimensional data, calculating the Wasserstein distance is very simple and efficient; it can be obtained by averaging the absolute values of the differences after sorting the samples. Finally, the projection and difference calculation process is repeated, and the expected value of the Wasserstein distances in these processes is taken as the final slice Wasserstein distance. The formula for calculating the slice Wasserstein distance is:
[0075]
[0076] in, , Let X represent the probability distribution of random variables X and Y. Represents any direction vector. , This represents the projection of X and Y in that direction. Let pWortherstein distance represent the distance between one-dimensional distributions.
[0077] In this embodiment, the trajectory generated by the minimum SWD value is selected as the imitation trajectory by calculating the slice Wasserstein distance between all imitation trajectories and expert demonstration samples.
[0078] In this embodiment, to ensure that the generated imitation trajectory is as close as possible to the trajectory in the expert demonstration sample, the preset paradigm requirements include:
[0079] Paradigm 1: The imitated trajectory is state-distinguishable. The imitated trajectory should be determined by the environmental state accessed by the nursing robot and be distinguishable at the state level, thus ensuring that the generated trajectory revolves around a specific state. The mutual information between random variables and state information is maximized, and the calculation formula is as follows:
[0080] ;
[0081] Where I represents mutual information, S represents state, and Z represents a random variable randomly sampled from the expert demonstration sample C.
[0082] Paradigm 2, where the imitation trajectory is distinguished based on state rather than action, ensures that the trajectory's distinguishability primarily stems from changes in the environmental state. This involves minimizing the mutual information between random variables and state-based action sequences, calculated using the following formula:
[0083] ;
[0084] A|S represents action A in state S.
[0085] Paradigm 3: The imitation trajectory should be explored and learned randomly as much as possible, and the imitation trajectory segments with discriminative characteristics should be learned. The nursing robot is encouraged to explore and learn discriminative imitation trajectory segments randomly under given environmental conditions, thereby ensuring that a sufficient number of imitation trajectories with high information entropy can be generated for subsequent matching. The information entropy of the action sequence in a specific state is maximized, and the calculation formula is as follows:
[0086] ;
[0087] H represents information entropy.
[0088] Paradigm 4 stipulates that the imitated trajectory should be similar to the imitated demonstration sample. The trajectory generated by the nursing robot should be as similar as possible to the expert demonstration sample, thereby imposing constraints on the sufficient number of trajectories generated in Paradigm 3. This ensures that the generated trajectory is geometrically similar to the expert demonstration sample, avoiding the generation of a large number of invalid or inefficient trajectories. The mutual information between the imitated trajectory and the expert demonstration sample is maximized, and the calculation formula is as follows:
[0089]
[0090] C represents the expert demonstration sample.
[0091] To enable the transfer from simulated to real-world environments, this embodiment uses the expected position of joint angles as intermediate general data to accurately control the nursing robot to achieve the same functional effects in different experimental environments.
[0092] When the method is transferred to a real-world environment, it includes: first, extracting the desired joint angle for each position on the simulated trajectory in the simulated environment; then, controlling the real robot to drive each joint to sequentially follow the desired position, thus realizing the movement of the nursing robot. During the movement of the nursing robot, the size of the time step is controlled to ensure that each joint can complete the movement within the corresponding time step. At the same time, the difference between the current joint angle and the desired joint angle is continuously adjusted and reduced, so that each joint of the robot can accurately fit the simulated state at the target time point.
[0093] The nursing robot imitation learning method of this invention utilizes a single expert demonstration sample to generate morphologically similar yet diverse imitation trajectories around that single expert demonstration. The nursing robot learns from this single expert demonstration, effectively reducing its dependence on expert samples in imitation learning. Furthermore, using the slice Wasserstein distance between the imitation trajectory and the expert demonstration sample as a criterion for evaluating the quality of the imitation trajectory, and decomposing the high-dimensional distribution estimation into simpler estimates of multiple one-dimensional distributions, greatly reduces the computational load of trajectory matching and improves computational efficiency.
[0094] Example 2
[0095] This application provides a computer device including a processor and a memory. The memory stores at least one instruction or at least one program, which is loaded and executed by the processor to implement a nursing robot imitation learning method as provided in the above method embodiments.
[0096] Figure 2 A schematic diagram of the hardware structure of a device for implementing a nursing robot imitation learning method provided in the embodiments of this application is shown. The device can participate in or include the apparatus or system provided in the embodiments of this application. Figure 2 As shown, the computer device 10 may include one or more processors 1002 (the processor may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 1004 for storing data, and a transmission device 1006 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 2 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer device 10 may also include... Figure 2 The more or fewer components shown, or having the same Figure 2 The different configurations shown.
[0097] It should be noted that the aforementioned one or more processors and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuit may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer device 10 (or mobile device). As involved in the embodiments of this application, the data processing circuit serves as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).
[0098] The memory 1004 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to a nursing robot imitation learning method in this embodiment of the application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 1004, thereby implementing the aforementioned method. The memory 1004 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1004 may further include memory remotely located relative to the processor, and these remote memories can be connected to the computer device 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
[0099] The transmission device 1006 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer device 10. In one example, the transmission device 1006 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 1006 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.
[0100] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer device 10 (or mobile device).
[0101] Example 3
[0102] This application embodiment also provides a computer-readable storage medium, which can be disposed in a server to store at least one instruction or at least one program related to implementing a nursing robot imitation learning method in the method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement the nursing robot imitation learning method provided in the above method embodiment.
[0103] Optionally, in this embodiment, the storage medium may be located at at least one of the multiple network servers in a computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.
[0104] Example 4
[0105] This invention also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform a nursing robot imitation learning method provided in the various optional embodiments described above.
[0106] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are also possible or may be advantageous.
[0107] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device, equipment, and storage medium embodiments are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.
[0108] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.
[0109] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.
Claims
1. A method for imitation learning of a nursing robot, characterized in that, The method includes the following steps: S1 initializes the parameters within the environment; S2, sample and generate imitation trajectories from the policy network; S3, set the maximum number of rounds or time steps, and collect and process the generated imitation trajectory based on experience; S4, sample empirical tuples from the imitation trajectory, perturb the sampling to satisfy the action prior, ensure that the policy output does not deviate from the expert demonstration sample, and perform imitation constraints; S5, based on the preset paradigm requirements, calculates the mutual information terms using a neural network mutual information estimation method; S6. Based on the mutual information terms, the constraint signals are transformed into internal reward terms of the policy network, and pseudo-rewards are saved. S7, add trajectory segments with pseudo-rewards to the experience feedback buffer; S8, update the network parameters according to the soft critic algorithm and the preset paradigm requirements, and repeat steps S2 to S8 until the convergence condition is met; S9 calculates the slice Wasserstein distance based on the imitation trajectory and expert demonstration samples, and selects the optimal trajectory from the imitation trajectories according to the slice Wasserstein distance.
2. The nursing robot imitation learning method as described in claim 1, characterized in that, In step S1, the parameters include the position and posture of the nursing robot, the position and posture of the target object, the position of obstacles in the environment, the physical posture information of the nursing robot, the strategy function parameter θ, and the Q-value function parameters ϕ and ψ.
3. The nursing robot imitation learning method as described in claim 1, characterized in that, In step S2, an initial policy z is obtained by sampling from expert demonstration samples, and the agent is guided to generate an imitation trajectory around a single expert demonstration based on the initial policy z.
4. The nursing robot imitation learning method as described in claim 1, characterized in that, In step S4, the disturbance sampling satisfies the following condition: ， in, Indicates the action at time t. The parameter is random strategy, Indicates the state is The action at that time, z represents the expert demonstration sample sampling information.
5. The nursing robot imitation learning method as described in claim 1, characterized in that, The internal reward is defined as the sum of the results from two mutual information estimation methods, calculated using the following formula: ， in, MINE represents the mutual information estimation method in neural networks, where MINE stands for intrinsic reward. Indicates the initial strategy. express The status of the nursing robot over time. This indicates the playback buffer.
6. The nursing robot imitation learning method as described in claim 1, characterized in that, The formula for calculating the slice Wasserstein distance is: ， in, , Let X represent the probability distribution of random variables X and Y. Represents any direction vector. , This represents the projection of X and Y in that direction. Let pWortherstein distance represent the distance between one-dimensional distributions.
7. The nursing robot imitation learning method as described in claim 1, characterized in that, The pre-defined paradigm requirements include: Paradigm 1 maximizes the mutual information between the random variable and the state information, and the calculation formula is as follows: ； Where I represents mutual information, S represents state, and Z represents random variable; Paradigm 2 minimizes the mutual information between random variables and state-based action sequences, calculated as follows: ； Where A|S represents action A in state S; Paradigm 3 maximizes the information entropy of a sequence of actions in a specific state, calculated using the following formula: ； Where H represents information entropy; Paradigm 4 maximizes the mutual information between the imitation trajectory and the expert demonstration sample, calculated using the following formula: ， Where C represents the expert demonstration sample.
8. The nursing robot imitation learning method as described in claim 1, characterized in that, When the method is adapted to a real-world environment, it includes the following steps: Extract the desired position of the joint angle corresponding to each position on the simulated trajectory in the simulation environment; The robot is controlled to drive each joint to follow the corresponding desired position in sequence, thus enabling the movement of the nursing robot.
9. A computer device, characterized in that, include: processor; Memory, used to store executable instructions; The processor is configured to read the executable instructions from the memory and execute the executable instructions to implement the nursing robot imitation learning method as described in any one of claims 1 to 8.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the nursing robot imitation learning method as described in any one of claims 1 to 8.