A tracking control method for a mobile robot guided by sound source information

By introducing channel domain attention mechanism and dual closed-loop PID controller in the sound source positioning model, the problems of high complexity of the sound source positioning model and low robot motion control accuracy are solved, and efficient sound positioning and robot tracking control are achieved, improving Positioning accuracy and data transmission efficiency.

CN118655767BActive Publication Date: 2024-11-12ANHUI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411133991.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-11-12
Estimated Expiration
2044-08-19

AI Technical Summary

Technical Problem

The existing sound source positioning model has high complexity and difficult to ensure the robot's motion control accuracy, especially in complex environments and multi-sound source scenarios, and the limited network transmission resources lead to data conflicts, which reduces efficiency.

Method used

采用融入通道域注意力机制的卷积循环神经网络进行声音定位,并结合双闭环PID控制器进行机器人跟踪控制,通过短时傅里叶变换提取音频特征,构建数学模型,设计角度和速度环PID控制器,使用SCP协议优化网络传输。

Benefits of technology

It improves sound positioning accuracy and robot trajectory tracking accuracy, enhances the robustness of the model and data transmission efficiency, reduces data conflicts and bandwidth waste, and ensures that the robot can track predetermined trajectories stably in different environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118655767B_ABST
    Figure CN118655767B_ABST
Patent Text Reader

Abstract

The invention relates to a method for tracking and controlling a mobile robot guided by sound source information. The method extracts amplitude and phase features of audio as input features, uses a convolutional recurrent neural network integrated with a channel domain attention mechanism as a sound localization network, and when applied, the sound source localization information is transmitted to a controlled robot through a network using a scheduling control protocol SCP, and the values ​​of the robot's input left and right wheel motors are obtained by modeling the controlled object robot and using a double closed-loop PID controller, so as to track and control the robot to reach a preset position. By applying the method of the invention, the model feature selection capability and the model robustness can be enhanced, and the robot can maintain accurate tracking of a path while achieving smooth and coherent motion of the robot. Finally, by introducing the scheduling control protocol SCP, the data transmission efficiency can be improved and the bandwidth waste caused by data collision and retransmission can be reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a tracking control method for a mobile robot guided by sound source information. Background Art

[0002] Sound source localization technology accurately determines the location of sound in space by analyzing the spatial features of sound signals. The method is usually divided into two stages: feature extraction and deep classification positioning. First, key features are extracted from the sound signal, and then the trained neural network model is used to learn and predict the exact location of the sound source. This method performs well in dealing with complex environments and multi-sound source scenes, and with the continuous advancement of technology, more advanced algorithms such as generative adversarial networks and reinforcement learning may be integrated in the future to further improve the performance and adaptability of the system. With the continuous advancement of sound localization technology research, in order to ensure higher sound source localization accuracy on large datasets containing a large number of samples, developers continue to push the model architecture to a deeper level. This increase in depth enables the model to capture more subtle sound features and complex spatial relationships, thereby significantly improving the accuracy of positioning. However, this also brings a significant increase in model complexity, which means that the model requires more parameters and computing resources to train and run. The training process of deep models becomes more time-consuming and requires more memory and processing power, which may lead to increased training costs. In addition, as the complexity of the model increases, how to design effective regularization strategies to prevent overfitting and ensure that the model has good generalization ability has also become a challenge that researchers need to face.

[0003] In terms of robot tracking control, the robot is a very complex multi-input and multi-output nonlinear system with time-varying, strongly coupled and nonlinear dynamic characteristics. In addition, due to the high complexity of the robot system itself and the continuous time-varying nature of the tracked trajectory, the robot trajectory tracking control is very difficult and the control accuracy is difficult to guarantee.

[0004] In terms of network transmission, network communication resources are often limited, and it is impossible to transmit data unlimitedly. Therefore, there is a problem of conflict caused by direct data competition during network transmission, which directly reduces the efficiency of network resource utilization. Summary of the invention

[0005] In view of this, the purpose of the present invention is to propose a tracking control based on improved sound source localization technology to solve the problems of high complexity of existing sound source localization models and robot motion control.

[0006] To solve the above technical problems, the present invention provides the following technical solution: using a convolutional recurrent neural network with a channel domain attention mechanism as a sound localization network, and using a tracking control method to enable the robot to reach a preset position. The method includes the following steps:

[0007] S1. Build a data set and perform preprocessing;

[0008] S2, extracting the amplitude and phase features of the audio through short-time Fourier transform as input features of the positioning network;

[0009] S3, locate the amplitude and phase characteristics of the audio to obtain the spatial position of the sound;

[0010] S4, modeling the controlled robot to obtain its mathematical model;

[0011] S5, transmitting the location information obtained by locating the sound source to the robot through the network;

[0012] S6, construct a double closed-loop PID controller to obtain the input values ​​of the left and right wheel motors;

[0013] S7, driving the motors according to the input of the two-wheel motors to move toward the target direction;

[0014] Furthermore, in step S4, the specific process includes the following steps:

[0015] S41. A mathematical model is established for the robot according to its position in the plane coordinate system. The mathematical model expression is:

[0016]

[0017] in, is the position vector of the robot; represents the left wheel angular velocity; represents the right wheel angular velocity; Indicates the radius of the robot's driving wheel; Indicates the distance between the centers of the two driving wheels.

[0018] S42, set the reference robot position as Figure 6 N points in the middle, whose positions are determined by the vector Determine and establish a local coordinate system based on the M point of the two-wheeled robot , from which we can get the error vector of the robot moving in two positions:

[0019]

[0020] In the above formula, In the local coordinate system Vector coordinates, , , All are the coordinate position information of the reference position. , , All are current location coordinate information;

[0021] S43. According to the conditions, the position error differential equation of the two-wheel drive robot is obtained:

[0022]

[0023] The positioning of a two-wheel drive robot can be defined by the following formula:

[0024]

[0025] In the formula, and They correspond to the angular displacement changes measured by the right and left wheel encoders of the robot in a single sampling period; represents the straight-line distance covered by the robot during the sampling period; is the angle that the robot rotates during the sampling period; A vector representing the position of the robot at a specific point in time; is the driving wheel radius; Indicates the distance between the centers of the two driving wheels.

[0026] Furthermore, in step S5, the specific process includes the following steps:

[0027] S51, control signals are transmitted through the network, and the SCP protocol is used to schedule the signals to avoid network congestion;

[0028] S52. Convert the signal into a usable input signal through a compensation mechanism at the signal input end.

[0029] Furthermore, in step S6, the specific process includes the following steps:

[0030] S61. Design an angle loop PID controller, the controller form is as follows:

[0031]

[0032]

[0033] In the above formula, Represents the proportional gain in the angle control loop; Represents the integral gain in the angle control loop; Represents the differential gain in the angle control loop; To represent the heading angle of the robot’s target; Indicates the current heading angle of the robot; Indicates the difference between the robot's current heading angle and the target heading angle;

[0034] S62. Design a speed loop PID controller, the controller form is as follows:

[0035]

[0036] In the above formula, Refers to the proportional coefficient of the speed control loop, is the integral coefficient of the speed control loop, is the differential coefficient of the speed control loop; represents the target linear velocity of the robot; Indicates the current linear speed of the robot; Represents the error between the target linear speed and the actual linear speed.

[0037] S63, calculate the control signal using the PID regulator of the speed control loop , which is used to adjust the basic rotation speed of the left and right wheel drive motors of the two-wheel drive robot; then, the control signal is calculated by the PID regulator of the angle control loop , to control the robot's orientation. The method is as follows:

[0038]

[0039] in, and are the input values ​​passed to the right and left drive motors respectively, and K is the coefficient that converts the control signal into a PWM (pulse width modulation) signal for the motor drive module.

[0040] The present invention has the following beneficial effects:

[0041] (1) The channel-domain attention mechanism of the present invention can adaptively adjust the weights of different channels in the feature map, so that the model pays more attention to the feature channels that are critical to the sound localization task and ignores unimportant features, thereby enhancing the feature selection capability; at the same time, the convolutional recurrent neural network integrated with the channel-domain attention mechanism is used as the network model for sound localization, which can suppress noise and other interference to a certain extent, so that the model can still maintain a high positioning accuracy when facing a variety of interference signals, thereby improving the robustness of the model.

[0042] (2) The present invention accurately adjusts the rotation speed of the left and right wheels of the two-wheel drive robot to ensure that the robot can stably and accurately track its desired path. Through the dual closed-loop structure, the controller can more carefully monitor and adjust the operating status of the robot. The inner loop is responsible for rapid response to maintain precise control of the wheel speed, while the outer loop focuses on the tracking performance of the overall path to ensure that the robot can still maintain the predetermined trajectory when facing different environmental conditions and possible interference.

[0043] (3) The present invention improves the efficiency of data transmission and reduces bandwidth waste caused by data collision and retransmission. By dynamically adjusting the priority and time schedule of data transmission, the conflict of data packets is effectively avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0045] Figure 1 It is a flow chart of the sound source localization and control algorithm of the present invention;

[0046] Figure 2 This is a structural diagram of the positioning network of the present invention;

[0047] Figure 3 The structure of the channel attention mechanism of the present invention;

[0048] Figure 4 It is the channel attention module of the present invention;

[0049] Figure 5 This is a flow chart of positioning network training of the present invention;

[0050] Figure 6 A schematic diagram of the robot motion modeling of the present invention;

[0051] Figure 7 Schematic diagram of the dual closed-loop controller of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] Please refer to Figure 1-Figure 7 ,This embodiment uses a convolutional recurrent neural network that incorporates a channel domain attention mechanism as the ,network model for sound localization, and uses tracking control as the ,control algorithm for robot motion to improve the positioning accuracy and ,robot control effect.

[0054] Please refer to Figure 1 A convolutional recurrent neural network with a channel-domain attention mechanism is used as a network model for sound localization, and tracking control is used as a control method for robot motion. The method includes the following steps:

[0055] S1. In order to construct a dataset for sound source localization, we used speech samples from the TIMIT public dataset and used the Roomsim toolbox to simulate the audio signals captured by microphone arrays in different room environments. Since the localization network mainly works based on the amplitude and phase difference of the sound signal, it does not depend on a specific type of sound, making these simulated audio signals very suitable for training the localization network. This method not only provides rich data, but also allows us to control the training conditions and enhance the generalization ability of the model.

[0056] S2, using short-time Fourier transform to extract the amplitude and phase characteristics of the sound signal as the input features of the positioning network; as a preferred implementation of step S2, the specific process includes the following steps:

[0057] S21, Frame division: The sound signal itself is non-stationary, so frame analysis is required. is divided into consecutive frames, each of which lasts for samples, and the frame overlap is Samples. The continuous time representation of the frame signal is denoted by , which can be described mathematically as:

[0058]

[0059] in, is the index of the frame number.

[0060] S22, windowing: The sound signal of the frame is windowed and the result is:

[0061]

[0062] in is the Hamming window function.

[0063] S23. Short-time Fourier transform (STFT): STFT is a method for analyzing the frequency characteristics of sound signals at different time points. It is achieved by dividing the signal into short frames, applying a window function to each frame, and then performing a Fourier transform. STFT can provide a time-frequency representation of sound signals and is often used in fields such as sound analysis and signal processing. However, there is a trade-off between time resolution and frequency resolution in STFT, and choosing the appropriate window function and frame length is crucial to the analysis results.

[0064]

[0065] in is the frequency of the sound signal.

[0066] S24, splicing: splicing the spectrogram of the sound signal involves splitting the signal into frames, performing STFT transform on each frame, and then stacking the resulting spectrograms in time order to form a complete time-frequency representation of the sound signal. This process provides rich time-frequency features for sound analysis, which helps in sound source localization and sound recognition.

[0067] S25, extracting amplitude and phase: after obtaining the calculated STFT spectrum, extracting the amplitude and phase components of the sound signal to obtain input features based on amplitude and phase features.

[0068]

[0069] in, is the sound signal amplitude, is the phase of the sound signal, To extract the amplitude, To extract the phase angle.

[0070] S3, using short-time Fourier transform to extract the amplitude and phase features of the sound signal, and input them into the positioning network as input features for training. The structure of the positioning network model is as follows: Figure 2 As shown in Figure 2, the positioning network consists of 4 convolutional layers, 2 attention layers composed of channel attention mechanisms and convolutional layers, a gated recurrent unit (GRU) layer and 3 fully connected layers (Fully), and the structure of the channel attention mechanism is as follows: Figure 3 As shown, Figure 4 is the channel attention module. The training process of the positioning network is as follows Figure 5 Finally, after continuous training, the expected recognition network model can be obtained.

[0071] S4, modeling the controlled robot to obtain its mathematical model;

[0072] As a preferred implementation of step S4, the specific process includes the following steps:

[0073] S41. A mathematical model is established for the robot according to its position in the plane coordinate system. The mathematical model expression is:

[0074]

[0075] in, is the position vector of the robot; represents the left wheel angular velocity; represents the right wheel angular velocity; Indicates the radius of the robot's driving wheel; Indicates the distance between the two driving wheels;

[0076] S42, set the reference robot position as Figure 6 N points in the middle, whose positions are determined by the vector Determine and establish a local coordinate system based on the M point of the two-wheeled robot , from which we can get the error vector of the robot moving in two positions:

[0077]

[0078] In the above formula, In the local coordinate system Vector coordinates, , , All are the coordinate position information of the reference position. , , All are current location coordinate information;

[0079] S43. According to the conditions, the position error differential equation of the two-wheel drive robot is obtained:

[0080]

[0081] The position information of the two-wheel drive robot can be expressed by the following formula:

[0082]

[0083] In the formula, and They correspond to the angular displacement changes measured by the right and left wheel encoders of the robot in a single sampling period; represents the straight-line distance covered by the robot during the sampling period; is the angle that the robot rotates during the sampling period; A vector representing the position of the robot at a specific point in time; is the driving wheel radius; Indicates the distance between the centers of the two driving wheels.

[0084] S5, transmitting the location information obtained by locating the sound source to the robot through the network;

[0085] As a preferred implementation of step S5, the specific process includes the following steps:

[0086] S51, control signals are transmitted through the network, and the SCP protocol is used to schedule the signals to avoid network congestion;

[0087] S52, at the signal input end, the signal is converted into an available input signal through a compensation mechanism, and the input signal is composed of the signal transmitted at the current moment and the available signal at the previous moment. A matrix with diagonal elements of 1 or 0 and the remaining elements of 0 is used as a compensator. The compensator is as follows:

[0088]

[0089] Among them, the Kronecker function The value is 1 when a is equal to zero, and 0 otherwise. Indicates the channels that are currently allowed to access the network by the Scheduling Control Protocol (SCP). Indicates the total number of channels. is an identity matrix with dimensions same.

[0090] S6, construct a double closed-loop PID controller to obtain the input values ​​of the left and right wheel motors;

[0091] As a preferred implementation of step S6, the specific process includes the following steps:

[0092] S61. Design an angle loop PID controller, the controller form is as follows:

[0093]

[0094]

[0095] In the above formula, Represents the proportional gain in the angle control loop; Represents the integral gain in the angle control loop; Represents the differential gain in the angle control loop; To represent the heading angle of the robot’s target; Indicates the current heading angle of the robot; Indicates the difference between the robot's current heading angle and the target heading angle;

[0096] S62. Design a speed loop PID controller, the controller form is as follows:

[0097]

[0098]

[0099] In the above formula, Refers to the proportional coefficient of the speed control loop, is the integral coefficient of the speed control loop, is the differential coefficient of the speed control loop; represents the target linear velocity of the robot; Indicates the current linear speed of the robot; Represents the error between the target linear speed and the actual linear speed.

[0100] S63, calculate the control signal using the PID regulator of the speed control loop , which is used to adjust the basic rotation speed of the left and right wheel drive motors of the two-wheel drive robot; then, the control signal is calculated by the PID regulator of the angle control loop , to control the robot's orientation. The method is as follows:

[0101]

[0102] in, and are the input values ​​passed to the right and left drive motors respectively, and K is the coefficient that converts the control signal into a PWM (pulse width modulation) signal for the motor drive module.

[0103] S7, the two-wheeled robot drives the left and right wheels through two independently controlled motors. The input signal calculated by the PID controller and , respectively control the rotation speed of the right and left wheels to adjust the robot's movement direction and speed to achieve precise movement to the target direction.

[0104] Figure 2 This is the structural diagram of the sound source localization network. The localization network consists of 4 convolutional layers, 2 attention layers composed of channel attention mechanisms and convolutional layers, one gated recurrent unit (GRU) and 3 fully connected layers (Fully).

[0105] Figure 3 This is the structural diagram of channel attention, where the channel attention module is as follows Figure 3 shown.

[0106] Figure 4 It is a channel attention module, which includes a maximum pooling layer, an average pooling layer, a multi-layer perception layer and an activation function.

[0107] Figure 5 To locate the network training flow chart, the process is as follows:

[0108] First, preprocess the data set and divide it into training set and test set;

[0109] Then use the training set to train the positioning network and save the model after training;

[0110] Furthermore, the test set is input into the model for testing and its loss is obtained;

[0111] Furthermore, the obtained loss is passed to the Adam optimizer to update the parameters of the positioning network, and then continue training;

[0112] When the loss is minimized, the positioning network model is saved and the training process ends.

[0113] Figure 6 The robot's navigation process in a plane coordinate system is shown, where point M represents the robot's current position and point N is its target location. Through the path planning algorithm, the robot calculates an optimal path from point M to point N, and detects and avoids obstacles in real time during movement, while constantly adjusting its direction of travel based on environmental feedback until it reaches the destination point N safely and accurately.

[0114] Figure 7 This is a schematic diagram of a double closed-loop PID control, showing the input and output of the controller and its feedback loop.

[0115] The present invention accurately adjusts the rotation speed of the left and right wheels of the two-wheel drive robot to ensure that the robot can stably and accurately track its desired path. Through the double closed-loop structure, the controller can monitor and adjust the robot's operating status more carefully. The inner loop is responsible for rapid response to maintain accurate control of the wheel speed, while the outer loop focuses on the tracking performance of the overall path to ensure that the robot can still maintain the predetermined trajectory when facing different environmental conditions and possible interference. In addition, the parameters of the PID controller are optimized and adjusted to adapt to the dynamic characteristics and path tracking requirements of the robot. Through this control algorithm, the robot shows higher flexibility and reliability when performing complex path tracking tasks. Whether on a smooth road or in a changing terrain, the robot can achieve smooth and coherent movement while maintaining accurate tracking of the path. The application of this control strategy significantly improves the robot's capabilities in automated navigation and path planning, and provides strong technical support for the application of robotics in multiple fields.

[0116] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for tracking and controlling a mobile robot guided by sound source information, characterized in that: The method comprises the following steps: S1, construct the data set and perform preprocessing; S2, extracting the amplitude and phase features of the audio through short-time Fourier transform as input features of the positioning network; S3, locate the amplitude and phase characteristics of the audio to obtain the spatial position of the sound; S4, modeling the controlled robot to obtain its mathematical model; S5, transmitting the location information obtained by locating the sound source to the robot through the network; S6, construct a double closed-loop PID controller to obtain the input values ​​of the left and right wheel motors; S7, driving the motors according to the input of the two-wheel motors to move toward the target direction; S4 specifically includes the following steps: S41. A mathematical model is established for the robot according to its position in the plane coordinate system. The mathematical model expression is: , in, is the position vector of the robot; represents the left wheel angular velocity; represents the right wheel angular velocity; Indicates the radius of the robot's driving wheel; Indicates the distance between the centers of the two driving wheels; S42, set the reference robot position, whose position is determined by vector Determine and establish a local coordinate system based on the M point of the two-wheeled robot , from which we can get the error vector of the robot moving in two positions: , In the above formula, In the local coordinate system Vector coordinates, , , All are the coordinate position information of the reference position. , , All are current location coordinate information; S43. According to the conditions, the position error differential equation of the two-wheel drive robot is obtained: , The positioning of a two-wheel drive robot is expressed by the following formula: , In the formula, and They correspond to the angular displacement changes measured by the right and left wheel encoders of the robot in a single sampling period; represents the straight-line distance covered by the robot during the sampling period; is the angle that the robot rotates during the sampling period; A vector representing the position of the robot at a specific point in time; is the driving wheel radius; Indicates the distance between the centers of the two driving wheels.

2. The method for tracking and controlling a mobile robot guided by sound source information according to claim 1, characterized in that: S5 specifically includes the following steps: S51, control signals are transmitted through the network, and the SCP protocol is used to schedule the signals to avoid network congestion; S52. Convert the signal into a usable input signal through a compensation mechanism at the signal input end.

3. The method for tracking and controlling a mobile robot guided by sound source information according to claim 2, characterized in that: S6 specifically includes the following steps: S61. Design an angle loop PID controller, the controller form is as follows: , , In the above formula, Represents the proportional gain in the angle control loop; Represents the integral gain in the angle control loop; Represents the differential gain in the angle control loop; To represent the heading angle of the robot’s target; Indicates the current heading angle of the robot; Indicates the difference between the robot's current heading angle and the target heading angle; is the angle control signal; S62. Design a speed loop PID controller, the controller form is as follows: , , In the above formula, Refers to the proportional coefficient of the speed control loop, is the integral coefficient of the speed control loop, is the differential coefficient of the speed control loop; Indicates the target linear velocity of the robot; Indicates the current linear speed of the robot; Represents the error between the target linear speed and the actual linear speed; It is the speed control signal; S63, calculate the control signal using the PID regulator of the speed control loop , used to adjust the basic rotation speed of the left and right wheel drive motors of the two-wheel drive robot; Calculation of the control signal via the PID controller of the angle control loop , to control the robot's orientation, as follows: , in, and are the input values ​​passed to the right and left drive motors respectively, and K is the coefficient that converts the control signal into a pulse width modulation signal for the motor drive module.