A wheel mobile robot slip ratio prediction method based on a deep learning algorithm

Abstract

The application belongs to the field of wheeled mobile robot state estimation, and discloses a wheel slip rate prediction method for a wheeled mobile robot based on deep learning, which solves the problems of insufficient prediction accuracy caused by the inability of real-time sensor measurement to predict in advance and the complexity of wheel-ground interaction. The specific method is as follows: firstly, a wheel slip rate data set of related motion data of the wheeled mobile robot is constructed; then, the network layers and structure of the CNN-LSTM are optimized and improved to efficiently extract information and capture features; in addition, a hidden state enhancement module is designed to strengthen the update of the hidden state, and an improved attention mechanism module is designed to dynamically adjust the input feature weight and highlight the most important features; finally, a compound loss function considering mechanical constraints is designed to train the model, and the trained model is verified and performance evaluated. The application enables the wheeled mobile robot to have good adaptability on different terrains and improves the accuracy and stability of the wheel slip rate prediction.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of robot application, and particularly relates to a wheel type mobile robot slip ratio prediction method based on a deep learning algorithm. BACKGROUND

[0002] Wheel type mobile robots have become an ideal choice in forestry, agriculture, industry and other fields due to their high reliability, simple design and inherent balance. With the progress of science and technology, wheel type robots are deployed in more challenging environments to reduce risks or replace humans to complete specific tasks. However, in outdoor environments and rugged terrains, the navigation and control of wheel type mobile robots still face great challenges. When dealing with geometric hazards in various environments, special attention needs to be paid to the potential risk of wheel slip in outdoor environments, which can significantly affect the mobility performance of wheel type mobile robots in off-road conditions. If this problem is not solved, wheel type robots may be trapped or lose control in soft or rugged terrains. Therefore, there is an urgent need for a reliable method to predict the slip ratio of the wheels in advance, to compensate for the slip by controlling the speed or torque, so as to avoid the negative effects caused by the terrain and achieve more precise control of the wheel type mobile robot.

[0003] In the perspective of prediction methods, there are approximately three categories: proprioceptive sensor-based, exteroceptive sensor-based, and model-based prediction methods. In the context of autonomous vehicles, proprioceptive sensors are on-board sensors such as inertial measurement units (IMUs), global positioning systems (GPS), and wheel encoders that measure the state of the vehicle. The combination of inertial measurement units (IMUs) and wheel encoders will provide sufficient information to represent the behavior of a wheeled robot for slip ratio prediction. Although this method can directly measure the physical interaction between the wheels and the terrain surface, it cannot predict the slip ratio in advance. On the other hand, exteroceptive sensor-based methods can predict the potential slip of a wheeled robot at a target location in real time. However, existing exteroceptive sensor-based methods usually rely on vision sensors and are based only on the characteristics of the terrain surface, which may not be sufficient for accurate slip ratio prediction. The wheel-terrain interaction is inherently nonlinear and highly complex, so deep learning algorithms are potentially a more suitable solution. In addition, wheeled mobile robots are dynamic systems, and the slip of the robot depends on the current state and past state of the robot, and the estimation problem of time series data also needs to be considered. Therefore, taking into account the above factors, the present application proposes an improved CNN-LSTM network model, the core of which is to design and add a hidden state enhancement module and an improved attention mechanism module, as well as a composite loss function with mechanical constraints during model training, which can more efficiently and accurately extract spatio-temporal feature information and capture time series features, fully consider the influence of different input variables on prediction, speed up the model error convergence speed, and improve the generalization ability and robustness of the model. SUMMARY

[0004] To solve the problems in the prior art, the present application provides a wheeled mobile robot slip ratio prediction method based on a deep learning algorithm, mainly comprising the following steps:

[0005] Step 1: Construct a wheeled mobile robot wheel slip ratio dataset containing mobile robot driving data under four different terrains (ice, road, mud, and sand). The input variables of the dataset are: terrain category, motor torque value T, linear speed v in the x direction (forward) and z direction (longitudinal), angular speed ω, and slope value θ, and the output variable is the slip ratio s;

[0006] Step 2: An improved CNN-LSTM network model is proposed, and the core lies in designing and adding a hidden state enhancement module and an improved attention mechanism module, and the improved attention mechanism module is a dynamic weighting SENet (DW-SENet). The model predicts the slip ratio of the wheeled mobile robot by extracting spatiotemporal feature information and capturing time series features. The combination and connection structure of the convolution layer and the pooling layer are optimized to efficiently extract feature information. The consistency of the input and output dimensions is maintained in each layer to ensure that the model fully utilizes the input features at different levels, avoids information loss caused by dimension changes, and improves the utilization efficiency of features. The designed and added hidden state enhancement module further optimizes the update of the hidden state by adding a linear layer to learn the information in the hidden state. The designed and added dynamic weighting channel attention mechanism module (DW-SENet) generates a dynamic weight matrix for each channel by adding a convolution layer based on SENet, fully considers the influence of different input variables on prediction, and adjusts and allocates weights;

[0007] Step 3: A composite loss function considering mechanical constraints is designed, and by controlling the weight of the mechanical constraint in the composite loss function, the error convergence speed during model training is faster, the required training rounds are fewer, and the generalization ability and robustness of the model are enhanced. According to the data set prepared in step 1, 70% of the training set is used to train the model, and 30% of the validation set is input into the trained network model for verification. After verification, the performance indicators are evaluated;

[0008] The beneficial effects of the application are as follows:

[0009] The wheeled mobile robot slip ratio prediction method based on the deep learning algorithm proposes an improved CNN-LSTM network model, learns the information in the hidden state by designing and adding a hidden state enhancement module, can effectively strengthen the capture of dependency relationships, and this improvement optimizes the update mechanism of the hidden state, can improve the accuracy and stability of prediction, and enhance the sensitivity to slip ratio changes under different terrains. By designing and adding an improved attention mechanism module, a dynamic weight matrix is generated for each channel, fully considering the influence of different input variables on prediction, adjusting and allocating weights, highlighting the most important features for slip ratio prediction, and suppressing secondary or noise features. In this way, the feature representation capability of the network is enhanced. By designing and adding a composite loss function considering mechanical constraints, the error convergence speed during model training is faster, the required training rounds are fewer, and the generalization ability and robustness of the model are enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 Flow chart of the technical route implemented by the application

[0011] Figure 2 Improved CNN-LSTM network model structure diagram

[0012] Figure 3 Improved attention mechanism module structure diagram

[0013] Figure 4 Designed hidden state enhancement module structure diagram DETAILED DESCRIPTION

[0014] The embodiments of the present application provide a wheeled mobile robot slip ratio prediction method based on a deep learning algorithm. The content shown includes the following steps:

[0015] Step 1: Construct a wheeled mobile robot wheel slip ratio dataset containing four different terrains (ice, road, mud and sand) under the mobile robot driving data. The input variables of the dataset are: terrain category, motor torque value T, x direction (forward) and z direction (longitudinal) linear speed v, angular speed ω and slope value θ, and the output variable is slip ratio.

[0016] The calculation method of the parameters contained in the dataset is as follows: the slope angle is calculated using the difference between the vertical position of the front wheel (Fz) and the vertical position of the rear wheel (Rz) and the length (L) between the front and rear wheels. The specific formula is: The longitudinal slip ratio is calculated, and the specific formula is: The slip value is between 0 and 1, where s represents the slip value, r is the wheel radius, v is the actual speed, and ω is the angular speed. The calculated longitudinal slip ratio is used as the true value of the dataset.

[0017] Step 2: An improved CNN-LSTM network model is proposed, which is characterized by the design and addition of a hidden state enhancement module and an improved attention mechanism module. The improved attention mechanism module is a dynamic weighting SENet (DW-SENet). The model extracts spatio-temporal feature information and captures time series features to predict the slip ratio of the wheeled mobile robot.

[0018] The improved CNN-LSTM network model efficiently extracts feature information by optimizing the combination and connection structure of convolution layers and pooling layers. The consistency of input and output dimensions is maintained in each layer to ensure that the model fully utilizes input features at different levels, avoids information loss due to dimension changes, and improves the utilization efficiency of features.

[0019] Extract spatio-temporal feature information, feature information F d (i) = {fi 1 ,f i 2 ,f i 3 ,f i 4 ,f i 5}, F g (i)={f i 6}, i = 1, 2, 3...N, where Fd represents the five features extracted from the motion data, Fg represents the terrain category, and N is the total number of instances in the dataset.

[0020] The input data is fed into a convolutional layer with four layers of 3×1 kernels. These layers are stacked to form a deep feature map representation. After each convolution operation, the ReLU activation function is applied to the activation layer to introduce nonlinear features, normalize the data, and filter out negative values, enhancing the network's expressive power.

[0021] The feature information after convolution and activation operations is input into a 4-layer pooling layer with a pooling window size and a step size of 2. The maximum value of every 2 elements is taken to halve the length of the feature map layer by layer, reducing the amount of data and retaining the main features.

[0022] To capture time series features, the model employs a three-layer stacked module structure. The input dimension of the first layer is directly derived from the number of output features of the feature extraction component, ensuring that convolutional features are fully utilized. The input dimension of each layer depends on the hidden state dimension of the previous layer, enabling layer-by-layer feature extraction and abstraction. In the forward propagation of each layer, the dropout parameter is set to 0.3 to prevent overfitting of the model.

[0023] The designed and added hidden state enhancement module further optimizes the update of the hidden state by adding a linear layer to learn the information in the hidden state.

[0024] The hidden state enhancement module is deployed in the improved LSTM network. The specific steps are: adding a linear layer to extract enhanced information on the hidden state, and controlling the inflow of enhanced information through a sigmoid function. The expression of the sigmoid function is: The specific formula design of the hidden state enhancement module is:

[0025] output t =sigmoid(W m ·h t +b m )

[0026] Among them, W m is the weight matrix, ht is the output hidden state, b m is the bias term, output t is the additional information extracted from the hidden state.

[0027] The extracted additional information is combined with the output hidden state h t of the model to form a reinforced hidden state h t ', the specific formula is designed as:

[0028] h t ' = output t + h t

[0029] This reinforced hidden state can better retain and reinforce important memory information for subsequent time steps.

[0030] The designed and added dynamic weight channel attention mechanism module (DW-SENet) generates a dynamic weight matrix for each channel on the basis of SENet by adding a convolution layer, fully considers the influence of different input variables on prediction, and adjusts and allocates weights;

[0031] The dynamic weight channel attention mechanism module (DW-SENet) has the following specific steps: a lightweight 1×1 convolution layer is designed to process the input features of each channel, thereby generating the weight of the channel. The specific formula is expressed as:

[0032] W c = σ(Conv 1×1 (x c ))

[0033] Where, Conv 1×1 is a lightweight 1×1 convolution layer, σ is a sigmoid function, x c is the input feature of different channels, and W c is a dynamic weight matrix.

[0034] The generated dynamic weight matrix is multiplied by the input feature map, and then summed according to the time step to obtain the dynamic feature description of each channel. The specific formula is expressed as:

[0035]

[0036] Where, is the dynamic feature description, W(x c )(t) is the dynamic weight of channel c at the t-th time step, and x c (t) is the feature of the time step.

[0037] Global Average Pooling (GAP) is used to generate global feature description for each channel, which is expressed as: The dynamic feature description and the global feature description are combined by a balance parameter a to generate the global description vector of each channel, which is designed as:

[0038]

[0039] where a is a learnable parameter, and the initial value of a is set to 0.4. During the training process, the model will automatically adjust the value of a according to the loss function and gradient descent algorithm to adapt to different input features.

[0040] The output of the weighted pooling is passed to the subsequent fully connected layer to calculate the final weight S c of each channel, which is expressed as:

[0041]

[0042] where W1 and W2 are the weight matrices of the two fully connected layers.

[0043] The generated final weight of each channel is multiplied by the feature map after convolution, and then used as the input feature information of the improved LSTM network.

[0044] Step 3: Design a composite loss function considering mechanical constraints, control the weight of mechanical constraints in the composite loss function to make the error convergence faster during model training, reduce the required training rounds, and enhance the generalization ability and robustness of the model.

[0045] A composite loss function considering mechanical constraints is designed, and the specific steps are as follows: the mechanical constraint equation is designed as:

[0046]

[0047] where N is the number of samples, T is the torque value, R is the wheel radius, μ max is the maximum friction coefficient, s critical is the critical slip rate, s i is the true value of the slip rate, m is the mass of the wheeled mobile robot, and θ is the slope angle. μ max is set to 0.2, and s critical is set to 0.9.

[0048] The specific expression of the standard MSE loss function is: where is the predicted value of the slip rate. Considering the weight of the mechanical constraint in the composite loss function, the above mechanical constraint equation is added to the MSE loss function to obtain the composite loss function as:

[0049] L = L m + λL p ,

[0050] where L is the total loss function, L m is the MSE loss function, λ is the weight that controls the mechanics constraint in the composite loss function, the initial value of λ is set to 0.1, and a more appropriate value is gradually adjusted and learned during the training process. Regularization constraints are imposed to change within a reasonable range. L p is the loss function considering the mechanical constraint.

[0051] Therefore, the specific expression of the composite loss function considering the mechanical constraint is:

[0052]

[0053] According to the data set division ratio of 70% of step 1, the training set is used to train the model, and the validation set with a division ratio of 30% is input into the trained network model for verification. After verification, the performance indicators are evaluated.

[0054] The optimizer is selected as Adam, and the model parameters are adjusted by iteratively training the data. In each iteration, the model's predicted output is compared with the actual label, the loss function value is calculated, and then the model weights are updated to reduce the loss. The training round epoch is set to 100, and the learning rate is set to 0.0003. The final version of the model and the version that performs best on the test set can be saved.

[0055] The random seed is set to make the experimental results reproducible, and the mean square error Root mean square error Mean absolute error These three errors are used as indicators to evaluate the performance of the model.

Claims

1. A method for predicting slip rate of a wheeled mobile robot based on a deep learning algorithm, characterized in that: The following steps are involved: Step 1: Construct a wheel slip rate dataset for a wheeled mobile robot. This dataset contains driving data on four different terrains: ice, road, mud, and sand. The input variables of the dataset are: terrain type, motor torque value T, linear velocity v in the x-direction (forward) and z-direction (longitudinal) direction, angular velocity ω, and slope value θ. The output variable is the slip rate s. Step 2: An improved CNN-LSTM network model is proposed, which adds a hidden state enhancement module and an improved attention mechanism module. The improved attention mechanism module is the dynamic weight channel attention mechanism Dynamic Weighting SENet (DW-SENet). The model predicts the slip rate of wheeled mobile robots by extracting spatiotemporal feature information and capturing time series features. By optimizing the combination and connection structure of convolutional layers and pooling layers, feature information is efficiently extracted, and the consistency of input and output dimensions is maintained in each layer, ensuring that the model fully utilizes input features at different levels, avoiding information loss due to dimensionality changes, and improving feature utilization efficiency. The added hidden state enhancement module learns the information in the hidden state by adding a linear layer, further optimizing the update of the hidden state. The added dynamic weight channel attention mechanism module DW-SENet generates a dynamic weight matrix for each channel by adding a convolutional layer on the basis of SENet, fully considering the impact of different input variables on the prediction to adjust and distribute weights; Step 3: Design a composite loss function that takes mechanical constraints into account. By controlling the weight of the mechanical constraints in the composite loss function, the error convergence speed during model training is faster, fewer rounds of training are required, and the generalization ability and robustness of the model are enhanced. According to the data set in step 1, a training set with a ratio of 70% is used to train the model, and a validation set with a ratio of 30% is input into the network model described after training for verification. After the verification is completed, its performance indicators are evaluated.

2. The method for predicting slip rate of a wheeled mobile robot based on a deep learning algorithm according to claim 1, characterized in that: The calculation method of the parameters contained in the data set in step 1 is as follows: Slope angle calculation, using the difference between the vertical position of the front wheel (Fz) and the vertical position of the rear wheel (Rz) and the length between the front and rear wheels (L) to calculate the slope value at each moment, the specific formula is: The specific formula for calculating the longitudinal slip rate is: The slip value is between 0 and 1, where s represents the slip value, r is the wheel radius, v is the actual speed, ω is the angular velocity, and the calculated longitudinal slip rate is used as the true value of the dataset.

3. The method for predicting slip rate of a wheeled mobile robot based on a deep learning algorithm according to claim 1, characterized in that: The specific steps in step 2 are: Step 3.1, extract spatiotemporal feature information, the feature information is: F d (i)={f i 1 ,f i 2 ,f i 3 ,f i 4 ,f i 5 }, F g (i)={f i 6 }, i = 1, 2, 3... N, where Fd represents the five features extracted from the motion data, Fg represents the terrain category, and N is the total number of instances in the dataset. The feature information is input into a convolutional layer with 4 layers of 3×1 kernels. After stacking the convolution layers, a deep feature representation is formed. After each convolution operation, the ReLU activation function is applied in the activation layer to introduce nonlinear features, normalize and filter out negative values ​​to enhance the network's expressive power. The feature information after convolution and activation operations is input into a 4-layer pooling layer with a pooling window size and a stride of 2. The maximum value of every 2 elements is taken to halve the length of the feature map, reducing the amount of data while retaining the main features. Step 3.2: Capturing time series features. A three-layer stacked module structure is used. The input dimension of the first layer is directly taken from the number of output features of the feature extraction part, ensuring that the convolutional features can be fully utilized. The input dimension of each layer depends on the hidden state dimension of the previous layer, which can extract and abstract features layer by layer. In the forward propagation of each layer, the packet loss rate (dropout) parameter is set to prevent overfitting of the model. Step 3.3, a method for predicting slip rate of a wheeled mobile robot based on a deep learning algorithm according to claim 1, is characterized in that the hidden state enhancement module in step 2 specifically comprises the following steps: adding a linear layer on the hidden state to enhance information extraction, and controlling the inflow of enhanced information through a sigmoid function. The specific formula is designed as follows: output t =sigmoid(W m ·h t +b m ) Among them, W m is the weight matrix, h t is the hidden state of the output, b m is the bias term, output t It is the additional information extracted from the hidden state, and the extracted additional information is compared with the output hidden state h of the model. t Combined to form a reinforced hidden state h t ', this enhanced hidden state can better retain and strengthen important memory information for use in subsequent time steps; h t '=output t +h t Step 3.4, a wheeled mobile robot slip rate prediction method based on a deep learning algorithm according to claim 1, characterized in that the improved attention mechanism module in step 2, specifically comprises the following steps: designing a lightweight 1×1 convolutional layer to generate a dynamic weight matrix for each channel, multiplying the generated dynamic weight matrix by the input feature map, and then summing them by time steps to obtain a dynamic feature description of each channel, combining the dynamic feature description with the global feature description through a balance parameter α to generate a global description vector for each channel, and the specific formula is expressed as: in, It is a dynamic feature description. It is a global feature description. Finally, the output of the weighted pooling is passed to the subsequent fully connected layer to calculate the weight of each channel.

4. The method for predicting slip rate of a wheeled mobile robot based on a deep learning algorithm according to claim 1, characterized in that: In step 3, a composite loss function considering mechanical constraints is designed. The specific steps are as follows: the mechanical constraint equation is designed as: Where N is the number of samples, T is the torque value, R is the wheel radius, μ max is the maximum friction coefficient, s critical is the critical slip rate, m is the mass of the wheeled mobile robot, and θ is the slope angle; By adding the above mechanical constraint equation to the MSE loss function, we can get the complete composite loss function L = L m +λL p , its specific expression is: Among them, s i is the true value of the slip rate, is the predicted value of slip rate, λ is the weight of the control mechanical constraint in the composite loss function, The optimizer is Adam. The model parameters are continuously adjusted by iterating the training data. In each iteration, the output predicted by the model is compared with the actual label, the loss function value is calculated, and then the model weights are updated to reduce the loss. The training round epoch is set to 100 and the learning rate is 0.0003. The final version of the model and the version with the best performance on the test set can be saved. The random seed is set to make the experimental results reproducible. The mean square error is defined. Root mean square error Mean absolute error These three errors are used as indicators to evaluate the performance of the model.

Images