Fault-tolerant reconstruction method for sensing fault of spatial dexterous hand

Abstract

The invention discloses a fault-tolerant reconstruction method for sensing faults of a spatial dexterous hand, and relates to the field of robot control. In order to solve the defects that in the prior art, a dexterous hand on-orbit sensor is weak in fault-tolerant capability, insufficient in model generalization capability and incapable of being adaptive to complex space environment changes, the technical scheme provided by the invention comprises the following steps: constructing a conditional variation auto-encoder model, inputting a joint position and a rope length sequence, and constructing a model; time sequence features are extracted, and a reconstructed joint position mean value and a prediction variance are output; the reconstructed joint position mean value is compared with actual joint data, a dynamic confidence interval is generated, and whether the data breaks down or not is judged; for the data judged to be normal, information entropy is calculated based on the prediction variance, and data samples with the information entropy exceeding a set threshold value are screened and written into a dynamic buffer area; and when the buffer data is fully loaded, incremental learning is triggered, and model parameters are updated by utilizing the buffer data. The method is suitable for real-time detection and fault-tolerant reconstruction of sensor faults in the long-term on-orbit operation process of the space dexterous hand.

Description

Technical Field

[0001] The invention relates to the field of robot control and belongs to a fault-tolerant reconstruction method of an on-orbit dexterous space hand. Background Art

[0002] As the complexity of on-orbit space operations continues to increase, space robots are developing towards greater intelligence and flexibility. As an important form of robotic end effector, the dexterous space hand, with its multi-degree-of-freedom and multi-finger control capabilities, is capable of performing a variety of high-precision on-orbit operations such as grasping, assembly, and maintenance, and has become a key component in the development of space service robots.

[0003] However, the space environment differs significantly from the terrestrial environment, featuring complex characteristics such as extreme temperature differences, strong radiation interference, magnetic field fluctuations, and microgravity. These characteristics pose a significant challenge to the stability of dexterous hand sensors. Especially in a dexterous hand system with high degrees of freedom and multiple redundant structures, a failure in any joint or sensor could trigger a cascading failure, seriously impacting the robot's overall performance and mission execution capabilities.

[0004] To ensure mission safety and system robustness, domestic and foreign scholars have conducted extensive research on the detection and fault-tolerant control of robot sensor failures. Common research methods include: Model-based methods, such as sliding mode observers and Kalman filters, achieve fault detection and variable estimation by building a system dynamics model. These methods rely on model accuracy and are difficult to adapt to environmental changes or nonlinear systems.

[0005] Data-driven methods, such as neural networks and autoencoders, achieve fault prediction and variable reconstruction by learning from historical data patterns. For example, one study proposed using long-short-term memory (LSTM) networks to predict robot joint angles and detect abnormal sensor outputs. However, most of these methods are trained based on ground-based environmental data and fail to fully consider the non-stationary disturbances and data scarcity issues of the on-orbit environment.

[0006] Fault-tolerance method based on redundant hardware: backup redundancy is achieved by adding multiple sensors and comparing outputs to determine faults. However, this method will bring additional load and structural complexity in space missions, making it difficult to adapt in the long term.

[0007] In recent years, research has attempted to apply generative models such as variational autoencoders (VAEs) and conditional VAEs (CVAEs) to robot state reconstruction. However, these models still have shortcomings in high-degree-of-freedom systems, adaptability to dynamic environments, and online update capabilities. Furthermore, existing fault diagnosis techniques often use static or sliding thresholds to determine anomalies, making them difficult to address multi-source interference in spatial environments and system drift caused by long-term operation.

[0008] In summary, the existing technology has the defects of weak fault tolerance of on-orbit sensors of dexterous hands, insufficient model generalization ability, and inability to adapt to changes in complex space environment. Summary of the Invention

[0009] To address the shortcomings of the existing technology, such as weak fault tolerance of on-orbit sensors, insufficient model generalization, and inability to adapt to changes in the complex space environment, the present invention provides the following technical solutions: The fault-tolerant reconstruction method for sensing failures of the spatial dexterous hand includes: Construct a conditional variational autoencoder model, input joint position and rope length sequence, extract time series features and output the reconstructed joint position mean and prediction variance; Comparing the reconstructed joint position mean with the actual joint data, performing Bayesian inference based on the reconstruction error and the prediction variance, generating a dynamic confidence interval, and determining whether the data is faulty; For data determined to be normal, calculating information entropy based on the prediction variance, screening data samples whose information entropy exceeds a set threshold and writing them into the dynamic buffer; When the buffer data is full, incremental learning is triggered to update the model parameters using the buffer data.

[0010] Furthermore, a preferred embodiment is provided, in which the conditional variational autoencoder model includes a bidirectional temporal encoder and a conditional decoupling decoder, wherein the bidirectional temporal encoder is used to extract the spatiotemporal features of the input data, and the decoder outputs the mean and variance of the reconstructed joint positions respectively.

[0011] Furthermore, a preferred embodiment is provided in which Bayesian inference constructs a normal distribution model of the reconstruction error and the prediction variance, and generates a dynamic confidence interval based on the model to determine whether the actual sensor data is an outlier.

[0012] Furthermore, a preferred embodiment is provided, in which information entropy is calculated based on the prediction variance to measure the uncertainty of the model for the current sample, and data whose information entropy exceeds a preset threshold is considered representative and is written into the buffer.

[0013] Furthermore, a preferred embodiment is provided in which the capacity of the dynamic buffer is set by grid search.

[0014] Furthermore, a preferred implementation is provided, in which an elastic weight consolidation mechanism based on the Fisher information matrix is ​​adopted in the incremental learning process to constrain changes in key parameters.

[0015] A fault-tolerant reconstruction device for sensing faults of a spatial dexterous hand is also provided, including: Build a conditional variational autoencoder model that takes as input a sequence of joint positions and rope lengths, extracts temporal features, and outputs a module that reconstructs the joint position mean and prediction variance. A module that compares the mean of the reconstructed joint positions with the actual joint data, performs Bayesian inference based on the reconstruction error and the prediction variance, generates a dynamic confidence interval, and determines whether the data is faulty; For data that is determined to be normal, the module calculates information entropy based on the predicted variance, filters out data samples whose information entropy exceeds the set threshold, and writes them into the dynamic buffer; When the buffer data is full, incremental learning is triggered and the module uses the buffer data to update the model parameters.

[0016] A computer storage medium is also provided for storing a computer program, and when the computer program is read by a computer, the computer executes the method.

[0017] A computer is also provided, comprising a processor and a storage medium, wherein when the processor reads a computer program stored in the storage medium, the computer executes the method.

[0018] A computer program product is also provided, which is a computer program that implements the method when the computer program is executed.

[0019] Compared with the prior art, the technical solution provided by the present invention is beneficial in that: This paper constructs an improved CVAE architecture consisting of a bidirectional temporal encoder and a conditional decoupling decoder, enhancing the joint angle reconstruction capability of a high-degree-of-freedom spatial dexterous hand even when sensor data is corrupted or distorted. A bidirectional LSTM extracts temporal features, improving the accuracy of modeling the nonlinear relationship between joint position and cable length. Compared to traditional VAE methods using static encoders, this approach more accurately captures motion continuity and temporal logic, thereby improving the accuracy of joint reconstruction.

[0020] This invention introduces a dual-path output structure for position mean and variance at the decoder end, enabling the model to estimate prediction uncertainty while simultaneously outputting the prediction result. Unlike traditional neural networks that only output point estimates, this approach can dynamically adjust the system's confidence in prediction results even in the presence of impaired perception, providing a reliable basis for subsequent fault detection and fault-tolerant control, and enhancing the system's adaptability to unstable factors in the spatial environment.

[0021] The loss function designed in this paper comprehensively considers reconstruction error, KL divergence, and a regularization term for prediction variance, effectively avoiding the training speculation behavior of traditional CVAE models that masks errors by inflating prediction variance. While ensuring the accuracy of model output, it also enhances the physical interpretability of the prediction distribution. Compared to CVAE methods using a fixed regularization term, this design demonstrates stronger generalization capabilities in spatially dynamic scenarios.

[0022] This paper uses a Bayesian adaptive thresholding method (BAT) based on inference, combined with the predicted mean and variance of the CVAE model to dynamically generate confidence intervals, enabling real-time detection of sensor faults. Compared to common fixed threshold and sliding window strategies, this method dynamically adjusts detection sensitivity based on model uncertainty, significantly reducing false alarm and missed alarm rates, and demonstrating higher diagnostic accuracy when dealing with long-term changes in sensor characteristics in spatial environments.

[0023] This paper introduces a dynamic incremental buffer (DIB) management strategy based on information entropy. This strategy, combined with the information entropy of the data calculated from the model's prediction variance, selects only reliable and representative new data for model updates, reducing invalid data interference. Compared with traditional methods without a filtering mechanism or based on mean difference filtering, this strategy ensures the effectiveness of incremental model learning while controlling learning costs and memory usage.

[0024] This paper uses the Elastic Weight Consolidation (EWC) strategy for incremental learning, effectively avoiding catastrophic forgetting and ensuring that the model retains performance on previous tasks while learning new data. Compared to conventional SGD or Adam training methods, this approach maintains model stability and long-term memory in the face of slow but long-term changes in data distribution in spatial environments, improving the model's ability to continuously adapt to complex task environments.

[0025] This paper introduces a dynamic learning rate adjustment mechanism into incremental training, adjusting the parameter update rate based on sample variance. This improves the model's sensitivity to sample variations and accelerates convergence. Compared to a fixed learning rate mechanism, this approach is more suitable for space scenarios where sensor accuracy varies with mission cycles, enabling the model to more quickly respond to input perturbations caused by dynamic environments.

[0026] It is suitable for the real-time detection and fault-tolerant reconstruction of sensor failures during the long-term operation of the space dexterous hand in orbit. DETAILED DESCRIPTION

[0027] In order to make the advantages and benefits of the technical solution provided by the present invention more clearly reflected, the technical solution provided by the present invention is now described in further detail, specifically: Implementation 1: This implementation provides a fault-tolerant reconstruction method for sensing failures of a spatial dexterous hand, including: Construct a conditional variational autoencoder model, input joint position and rope length sequence, extract time series features and output the reconstructed joint position mean and prediction variance; Comparing the reconstructed joint position mean with the actual joint data, performing Bayesian inference based on the reconstruction error and the prediction variance, generating a dynamic confidence interval, and determining whether the data is faulty; For data determined to be normal, calculating information entropy based on the prediction variance, screening data samples whose information entropy exceeds a set threshold and writing them into the dynamic buffer; When the buffer data is full, incremental learning is triggered to update the model parameters using the buffer data.

[0028] The conditional variational autoencoder model includes a bidirectional temporal encoder and a conditional decoupling decoder. The bidirectional temporal encoder is used to extract the spatiotemporal features of the input data, and the decoder outputs the mean and variance of the reconstructed joint positions respectively.

[0029] Bayesian inference constructs a normal distribution model of the reconstruction error and prediction variance, and generates a dynamic confidence interval based on the model to determine whether the actual sensor data is an outlier.

[0030] Information entropy is calculated based on the prediction variance and is used to measure the uncertainty of the model for the current sample. Data whose information entropy exceeds the preset threshold is considered representative and is written to the buffer.

[0031] The capacity of the dynamic buffer is set via grid search.

[0032] In the incremental learning process, an elastic weight consolidation mechanism based on the Fisher information matrix is ​​adopted to constrain the changes of key parameters.

[0033] Implementation Method 2: This implementation method further describes the technical solution provided in Implementation Method 1 in detail. Specifically: This embodiment provides a sensor fault-tolerant reconstruction method suitable for on-orbit operation of a dexterous space hand. It combines an improved conditional variational autoencoder model (CVAE), a Bayesian adaptive threshold fault detection mechanism, and a dynamic incremental learning strategy to improve the robustness and adaptability of the dexterous hand system in extreme environments.

[0034] Step 1: Build an improved CVAE model and complete ground pre-training First, a CVAE network architecture consisting of a bidirectional temporal encoder and a conditional decoupled decoder was designed. The inputs to this architecture include joint position data as target variables and the associated rope length data as conditional variables, which are fed into a bidirectional LSTM encoder to extract temporal features. The extracted features are mapped to a latent space via a fully connected layer, which outputs the mean and variance of the latent variable. After sampling using a reparameterization technique, the sampled results are fed into a decoder. The decoder employs a dual-path architecture to output the predicted mean and variance of the joint positions, which are used to construct the reconstruction output and its uncertainty estimate. Based on this architecture, a loss function is designed that integrates the reconstruction loss, the KL divergence term, and the regularization term for the prediction variance to guide the model's pre-training on a ground simulation dataset.

[0035] Step 2: Use the model output to perform Bayesian adaptive threshold fault detection The CVAE model, pre-trained on the ground, is deployed in the on-orbit system, receiving real-time input of the rope length and outputting the corresponding reconstructed joint position and prediction variance. The mean of the model output is used as the reconstruction result, and the prediction variance is used as the error estimate. The reconstruction error is then compared with the actual sensor joint data. An error distribution model is constructed based on the reconstruction error and prediction uncertainty, and a dynamic confidence interval is calculated using Bayesian inference. An adjustable interval coefficient is introduced to construct a flexible detection boundary, and the threshold width is dynamically adjusted based on the model's confidence in the current prediction, enabling adaptive control of fault detection. When the actual joint data falls outside the confidence interval, it is considered abnormal; otherwise, it is considered normal.

[0036] Step 3: Perform information entropy filtering and build a dynamic incremental buffer For data samples determined to be normal, the information entropy of the current data is further calculated based on the predicted variance of the model output, which is used to measure the model's uncertainty about the current input. High-value samples with information entropy exceeding a set threshold are written to a dynamic incremental buffer. This buffer is structured as a first-in, first-out queue, storing a limited number of joint position and rope length sample pairs. To prevent the model from being corrupted by invalid data, the buffer capacity is set to an optimal value using a grid search method.

[0037] Step 4: Trigger online incremental learning and adopt elastic weight consolidation strategy When the dynamic buffer reaches the set upper limit, the incremental learning process is triggered. First, the importance of each parameter to the original task is calculated from the pre-trained model parameters to form a parameter importance evaluation matrix. A regularization term is introduced during incremental training to constrain the adjustment range of parameters and prevent the loss of old knowledge. The buffer data is divided into several mini-batches. A forward propagation is performed on each batch of data to calculate the model output. Gradients are calculated based on the aforementioned multiple loss functions, and the AdamW optimizer is used to update the parameters. The learning rate is dynamically adjusted based on the predicted variance to adapt to the changing sample distribution and ensure stable convergence of updates.

[0038] Step 5: Continuously iterate and update the CVAE model to adapt to changes in the orbital environment Through this online incremental learning process, the model continuously absorbs new data features, gradually adapting to the dynamic changes in sensor aging, performance drift, and environmental interference in the space environment, maintaining highly robust state perception capabilities. Furthermore, relying on a flexible weight consolidation mechanism, the model ensures that while incorporating new knowledge, it retains the stable features acquired during ground-based pre-training. Ultimately, this results in a fault-tolerant reconstruction system capable of long-term stable operation during on-orbit space missions.

[0039] Implementation Method 3: This implementation method further describes the above technical solution in detail through specific examples, specifically: This embodiment is intended to solve the problem that the fault diagnosis and fault-tolerant control of the existing high-degree-of-freedom dexterous hand cannot meet the on-orbit service requirements.

[0040] An Online Conditional Variational Autoencoder (Online-CVAE) model combined with Bayesian inference is used to achieve on-orbit fault-tolerant reconstruction of the dexterous hand. The implementation steps are as follows: Design an improved CVAE architecture, including a bidirectional temporal encoder, a sampling layer and a conditional decoupling decoder. The encoder converts the joint position and rope length As the input target variable and conditional variable, a bidirectional LSTM layer is used to extract spatiotemporal features, which are then mapped to the latent space through a fully connected layer, and the latent variable is finally output. The mean and variance Hidden variable sampling is achieved by introducing reparameterization techniques;

[0041] In order to simultaneously learn to reconstruct joint positions and their estimated uncertainties, a dual-path decoupling structure is adopted in the decoder network to predict the joint position mean and and variance Position mean The reconstructed joint position corresponding to the rope length can be output; variance This reflects the uncertainty in the model predictions.

[0042] To maximize the input target variable Given a condition variable The log-likelihood variational lower bound (ELBO) loss function is Designed to:

[0043] The reconstruction loss is used to measure the ability of the decoder to reconstruct the input data from the latent variables. Its core is to maximize the log-likelihood of the data to ensure that the generated result is as close to the original input as possible, assuming that the output of the decoder follows a Gaussian distribution. , reconstruction loss It can be defined as the negative log-likelihood:

[0044] KL divergence loss is used to regularize the latent space so that the latent variable distribution of the encoder output Approximate the preset prior distribution In order to prevent the KL divergence from decreasing rapidly in the early stage of training, which will lead to a decrease in the quality of model generation, the cosine annealing strategy is used to set The KL divergence is shown below:

[0045] The variance regularization term contains the variance amplitude and smoothness constraints on the reconstruction, preventing the model from reducing the reconstruction loss by arbitrarily increasing the variance:

[0046] A Bayesian Adaptive Threshold (BAT)-based fault detection method is proposed. Together with the CVAE model designed above, it forms an online conditional variational autoencoder model that combines Bayesian inference.

[0047] The Bayesian method uses the reconstruction error output by the CVAE model and variance Perform Bayesian inference and build a dynamic fault detection framework. Assume that the reconstruction error obey Normal distribution, and set the variance of the reconstruction error is the variance of the CVAE model output , the mean of the reconstruction error is the mean output of the CVAE model .

[0048] The dynamic confidence interval of the Bayesian method can be expressed as:

[0049] In the formula is a variable interval coefficient. The variance of the CVAE model output reflects the confidence in the reconstructed value, so the adaptive expansion factor is introduced Variable interval coefficient It can be expressed as

[0050] Posterior parameters They are designed as follows:

[0051]

[0052] Where, is the forgetting factor of the setting, is the confidence weight.

[0053] When the model prediction is reliable, the confidence weight Larger, The closer , the threshold boundary is kept narrow to ensure sensitivity to small anomalies. On the contrary, when the model prediction is unreliable, Increasing ⋅ expands the threshold boundary, thereby reducing the false alarm rate. The hyperbolic tangent function tanh(⋅) ensures the smoothness of threshold adjustment and avoids the influence of sudden changes on the detection results.

[0054] The Bayesian fault detection method uses the uncertainty of the CVAE model output to perform Bayesian inference and adaptively adjust the error detection interval and data distribution. Therefore, the angle input signal of the fault-tolerant impedance control system is involved. It can be expressed as:

[0055] When the system uses the Bayesian adaptive threshold method to detect anomalies in real-time data, if the data is judged to be normal and the information entropy of the data is Greater than threshold , then the motor position and joint position data are stored in the dynamic incremental buffer (DIB). This embodiment uses the prediction variance output by the decoder network of the CVAE model Calculate information entropy to measure the uncertainty of the model's understanding of the current data. The specific calculation formula for information entropy is:

[0056] The information entropy threshold is set to ,in and is the mean and standard deviation of the entropy of normal samples in the pre-training dataset. Let the capacity of the dynamic incremental buffer DIB be When the accumulated data reaches the upper limit, the incremental learning process is triggered. The data in DIB is ,in is the rope length data, The data is the joint angle data. Since a too small DIB capacity may lead to slow model convergence, and a too large capacity will cause challenges in storage space and iteration time, this embodiment uses a grid search method to select the most appropriate DIB capacity.

[0057] For the incremental learning part, this implementation introduces the Elastic Weight Consolidation (EWC) strategy to avoid the catastrophic forgetting problem that occurs during the incremental learning process. The EWC strategy constrains the changes in model parameters by adding a regularization term to the loss function, so that it can retain the memory of old knowledge while adapting to new data. First, M batches are sampled from the pre-training data distribution. , then for each parameter Calculate the Fisher matrix :

[0058] Where L is the original loss function, Characterize the importance of parameters to old tasks. In the incremental update process, define the EWC regularization term for:

[0059] in represents the optimal parameters obtained in the pre-trained network; is the weight coefficient of the EWC regularization term, which controls the retention strength of old knowledge. In order to balance the “stability-plasticity trade-off” in incremental learning, this implementation method Perform grid search within the range.

[0060] Online incremental updates use a batch gradient accumulation strategy. The data in the dynamic incremental buffer is divided into several sub-batches according to the preset mini-batch size. For each sub-batch, the network output is calculated using forward propagation. and variance , and calculate the gradient according to the loss function. AdamW optimizer is used for parameter update, and the momentum parameter is set to and The learning rate is dynamically adjusted according to the data variance. Through this online incremental learning mechanism, the model can continuously receive the latest data while maintaining its memory of the original data distribution, thereby achieving adaptive updates to the dynamically changing space environment.

[0061] This implementation is suitable for long-term on-orbit service and has strong anti-interference capabilities against complex characteristics such as extreme temperature differences, strong radiation interference, magnetic field fluctuations and microgravity.

[0062] This implementation can reduce learning costs through ground pre-training and perform incremental learning by actually collecting on-orbit operation data, thereby adjusting the control strategy to adapt to the dynamic deviations caused by long-term on-orbit operation of the high-degree-of-freedom dexterous hand.

[0063] This embodiment uses a Bayesian adaptive threshold algorithm to determine sensor fault conditions, thereby improving the accuracy of system status acquisition and ensuring the effectiveness of online incremental learning. During the ground pre-training process, the loss of each fold in this implementation method continued to decrease with the increase of training rounds. The training loss of all folds converged to the range of 1.09±0.01 within 40 epochs. The convergence trend verified the stability and generalization ability of the model. The final average loss of the five-fold cross-validation was 1.07, and its relative difference was less than 2.78%, indicating that the model has good robustness to changes in data distribution, indicating that the Online-CVAE model constructed in this implementation method can accurately predict joint angles. In space environment simulation experiments, the Bayesian adaptive threshold diagnosis used in this embodiment significantly improved the accuracy of CVAE model fault diagnosis compared to other fault diagnosis algorithms, such as fixed threshold, sliding threshold, and LSTM prediction. This is because on-orbit servicing is a long-term task. Due to factors such as equipment aging and damage, the dynamic performance of the dexterous hand and the detection accuracy of the sensors can deviate. Furthermore, special interference such as cosmic rays and high-energy particles in the space environment can also change over time. These factors pose challenges to the immediacy and robustness of on-orbit servicing prediction systems. Methods such as fixed threshold and LSTM prediction both pre-set detection mechanisms on the ground and cannot adaptively adjust detection parameters to changes in the surrounding environment during on-orbit servicing. Dynamic threshold methods, on the other hand, are significantly less adaptable to environmental changes than the Bayesian adaptive threshold diagnosis proposed in this embodiment, making it difficult to adapt to complex and changing space environments. Therefore, the Bayesian adaptive threshold diagnosis used in this embodiment is more suitable for on-orbit servicing scenarios and achieves higher accuracy.

[0064] The above further describes the technical solution provided by the present invention in detail through several specific embodiments in order to highlight the advantages and benefits of the technical solution provided by the present invention. However, the several specific embodiments described above are not intended to limit the present invention. Any reasonable modification and improvement of the present invention, combination of embodiments and equivalent replacement based on the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A fault-tolerant reconstruction method for sensing failures of a spatial dexterous hand, characterized by: include: Construct a conditional variational autoencoder model, input joint position and rope length sequence, extract time series features and output the reconstructed joint position mean and prediction variance; Comparing the reconstructed joint position mean with the actual joint data, performing Bayesian inference based on the reconstruction error and the prediction variance, generating a dynamic confidence interval, and determining whether the data is faulty; For data determined to be normal, calculating information entropy based on the prediction variance, screening data samples whose information entropy exceeds a set threshold and writing them into the dynamic buffer; When the buffer data is full, incremental learning is triggered to update the model parameters using the buffer data.

2. The fault-tolerant reconstruction method for sensing failure of the spatial dexterous hand according to claim 1 is characterized in that: The conditional variational autoencoder model includes a bidirectional temporal encoder and a conditional decoupling decoder. The bidirectional temporal encoder is used to extract the spatiotemporal features of the input data, and the decoder outputs the mean and variance of the reconstructed joint positions respectively.

3. The fault-tolerant reconstruction method for sensing failure of the spatial dexterous hand according to claim 1 is characterized in that: Bayesian inference constructs a normal distribution model of the reconstruction error and prediction variance, and generates a dynamic confidence interval based on the model to determine whether the actual sensor data is an outlier.

4. The fault-tolerant reconstruction method for sensing failure of the spatial dexterous hand according to claim 1 is characterized in that: Information entropy is calculated based on the prediction variance and is used to measure the uncertainty of the model for the current sample. Data whose information entropy exceeds the preset threshold is considered representative and is written to the buffer.

5. The fault-tolerant reconstruction method for sensing failure of the spatial dexterous hand according to claim 1 is characterized in that: The capacity of the dynamic buffer is set via grid search.

6. The fault-tolerant reconstruction method for sensing failure of the spatial dexterous hand according to claim 1 is characterized in that: In the incremental learning process, an elastic weight consolidation mechanism based on the Fisher information matrix is ​​adopted to constrain the changes of key parameters.

7. A fault-tolerant reconstruction device for sensing faults of a spatial dexterous hand, characterized in that: include: Build a conditional variational autoencoder model that takes as input a sequence of joint positions and rope lengths, extracts temporal features, and outputs a module that reconstructs the joint position mean and prediction variance. A module that compares the mean of the reconstructed joint positions with the actual joint data, performs Bayesian inference based on the reconstruction error and the prediction variance, generates a dynamic confidence interval, and determines whether the data is faulty; For data that is determined to be normal, the module calculates information entropy based on the predicted variance, filters out data samples whose information entropy exceeds the set threshold, and writes them into the dynamic buffer; When the buffer data is full, incremental learning is triggered and the module uses the buffer data to update the model parameters.

8. A computer storage medium for storing a computer program, characterized in that When the computer program is read by a computer, the computer executes the method according to claim 1 .

9. A computer comprising a processor and a storage medium, characterized in that When the processor reads the computer program stored in the storage medium, the computer executes the method according to claim 1 .

10. A computer program product, being a computer program, characterized in that When the computer program is executed, the method according to claim 1 is implemented.

Images