An intelligent voice-controlled navigation system based on fog computing technology

Abstract

The present invention relates to an intelligent voice-controlled navigation system based on fog computing technology. The system is arranged on a navigation robot and includes: a sensor module for acquiring surrounding environment data; a mapping module for drawing a complete map of the external environment based on SLAM using data transmitted by the sensor module; a path planning module for performing path planning and realizing navigation; a voice control module for responding to user needs based on topic publishing, and realizing human-computer interaction and voice recognition enhancement; a task unloading and allocation module for unloading and allocating tasks using fog computing technology; and a communication module for realizing communication between the navigation robot and the cloud and PC. Compared with the prior art, the present invention has the advantages of being able to process large amounts of map data and having a wide range of applications.

Description

Technical Field

[0001] The present invention relates to the field of robot navigation, and in particular to an intelligent voice-controlled navigation system based on fog computing technology. Background Art

[0002] With the rapid development of artificial intelligence technology, the development of intelligent voice robots has also made great progress and gradually appeared in daily life. However, as the application scenarios become more and more extensive, the requirements for robots are also getting higher and higher, which leads to the fact that voice navigation robots cannot meet the needs in some application scenarios. The most important ones include the following points:

[0003] When the robot is in a complex environment, such as when the surrounding environment is very noisy, the robot cannot distinguish between instructions and surrounding noise. The most direct manifestation is that after we issue an instruction, the robot does not respond or work according to the specified instruction. The main solution to this type of problem is to perform noise reduction processing on the input voice signal to minimize the impact of environmental noise. When the target command signal is weak, it can be enhanced. In general, the goal is to reduce environmental noise and increase the target signal.

[0004] Another problem is that the memory of the processors carried by most current robots is very limited, which means that when performing large-scale mapping or navigation work, the internal processors are far from meeting the actual requirements. The current solution is to use a lower-complexity algorithm to avoid insufficient memory, but this will result in lower accuracy of the final result. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent voice-controlled navigation system based on fog computing technology, which has high accuracy and wide application range.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] An intelligent voice-controlled navigation system based on fog computing technology, the system being arranged on a navigation robot, comprising:

[0008] Sensor module: used to obtain surrounding environment data;

[0009] Mapping module: Based on SLAM, it uses the data transmitted by the sensor module to draw a complete map of the external environment;

[0010] Path planning module: used for path planning and navigation;

[0011] Voice control module: used to respond to user needs based on topic publishing, and to achieve human-computer interaction and voice recognition enhancement;

[0012] Task offloading and allocation module: uses fog computing technology for task offloading and allocation;

[0013] Communication module: used to realize communication between the navigation robot and the cloud and PC.

[0014] The mapping module specifically performs the following steps:

[0015] During the movement of the robot, the encoder and IMU are used to calculate the odometer information, and the robot's motion model is used to obtain the initial estimate of the robot's position and posture.

[0016] The laser data obtained by the laser sensor mounted on the robot is combined with the observation model to accurately correct the robot's posture and obtain the precise positioning of the robot;

[0017] Based on precise positioning, laser data is added to the grid map to control the robot's movement in the environment, and laser data is continuously added to eventually complete the construction of the entire scene map.

[0018] The point cloud data generated during the mapping process is transmitted to the cloud for storage and calculation, and the computing tasks are assigned to the cloud to reduce the memory usage on the robot side.

[0019] The path planning includes global path planning and local path planning. Path planning is performed based on existing map information, i.e., static path planning, to find an optimal path from the starting point to the target point. The RRT global path planning algorithm is used. The path planning algorithm continuously generates state points randomly in the search space. If the point is located in a collision-free position, the node closest to the node in the search tree is found as the reference node, and the path is extended from the reference node until the distance between the target node and the search tree is within a certain range.

[0020] The voice control module combines the map constructed by the gmapping service in the SLAM module and the target point location information obtained by the navigation service to control the robot by voice. The voice recognition terminal sends a wake-up signal and prepares for voice input. After receiving the voice signal, the SDK's voice recognition function is called in the Publisher main loop. After successful recognition, the resultFlag variable is set to publish the recognized string.

[0021] The speech recognition enhancement adopts LMS adaptive filter to perform noise reduction, which specifically includes initialization, calculation of estimated value of filter output, calculation of estimation error and updating weight coefficient at the next moment.

[0022] Assume that the input signal sequence for speech recognition enhancement is x i(n), the expected output signal is d(n), and the error signal is defined as:

[0023]

[0024] Where M represents the filter order, w i is the weight coefficient, by finding the optimal weight coefficient w i , so that the error signal e(n) is minimized;

[0025] The iterative formula of the weight coefficient is:

[0026] W(k+1)=W(k)-μ▽(k)

[0027] Where W(k) represents the weight coefficient; μ is the convergence factor, which is used to control the convergence rate; ▽(k) represents the gradient factor; and the convergence factor value range is as follows:

[0028]

[0029] Among them, λ max is the maximum eigenvalue of the correlation matrix. When it tends to infinity, the weight vector converges to the optimal Wiener solution;

[0030] Use iterative descent method to solve the convergence factor and gradient factor;

[0031] The gradient factor expression is calculated as:

[0032]

[0033] Then, the iterative formula of weight coefficient is updated as follows:

[0034] W(k+1)=W(k)+2μe(k)X(k)

[0035] Where X(k) represents the input.

[0036] The method of responding to user needs based on topic publishing includes the following steps:

[0037] The PC side obtains the voice information set by the user;

[0038] The navigation robot and the PC communicate through the network to determine the IP addresses between them;

[0039] Modify the respective environment variables;

[0040] Write the publisher and subscriber codes, publish the topic name to the navigation robot, and the navigation robot will perform related operations after receiving it.

[0041] The task offloading and allocation module optimizes task processing by minimizing the total node energy, which includes the total offloading energy, the fog node total energy and the total transmission energy, wherein:

[0042] The total unloading energy is expressed as:

[0043]

[0044] Among them, u n represents the computing task of mobile robot n, K represents the number of robots, and b n represents the power consumption when the robot is unloaded, represents unloading energy;

[0045] The total energy of the fog node is expressed as:

[0046]

[0047] Among them, N c It represents the number of CPU cycles required to calculate 1 bit in the fog node, E f represents the energy consumption per cycle of fog node calculation, u n N c E f The computational energy consumption of offloading tasks for the mobile robot n;

[0048] The total transmission energy is expressed as:

[0049]

[0050] Among them, O n is the task analysis result of mobile robot n;

[0051] The objective function of the task offloading and allocation module is determined according to the total energy of the node:

[0052]

[0053] The task unloading and allocation module needs to determine the CPU occupancy before performing task unloading and allocation. The CPU occupancy is determined based on the relationship model between device power consumption and device CPU occupancy rate:

[0054] p=p idle +η(p max -p idle )

[0055] Among them, p idle is the power consumption when the CPU is idle, p max is the CPU full load power consumption, η is the resource occupancy rate of the device CPU;

[0056] The relationship between device power consumption and CPU frequency is:

[0057] E=k(f) 2 C

[0058] Among them, f is the main frequency of the device CPU, C is the CPU cycle required to process data, and k is the effective conversion capacitance of the device CPU chip.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] (1) Traditional robots cannot perform large-scale mapping and navigation tasks, largely because the robot's own processor has limited data processing capabilities and is difficult to perform big data calculations. Taking the above problems into consideration, this invention adds fog computing technology to the robot's mapping and navigation tasks. Compared with the currently popular cloud computing, fog computing has the advantages of being more energy-efficient and more efficient, and can adapt to application scenarios with large amounts of data.

[0061] (2) The present invention aims at how a robot can plan a path to reach a target point in an unknown environment. The RRT path planning algorithm is added to the planning process, so that the robot can reach the target point in an unknown environment and can search high-dimensional space quickly and effectively.

[0062] (3) Considering that the robot is always easily disturbed by the external environment in the actual environment, which makes the robot unable to recognize the target command well, the present invention adds an LMS adaptive filter to the robot's voice control module to reduce the impact of surrounding noise on the robot's voice command reception, and at the same time enhances the effective voice part so that the robot can better complete the task corresponding to the command.

[0063] (4) The point cloud data generated by the robot during the mapping process is transmitted to the cloud, and fog computing technology is used to solve computing tasks that require high processor power and memory in special scenarios, thereby reducing the memory usage on the robot side and improving computing processing capabilities.

[0064] (5) The present invention is based on SLAM for mapping and uses lidar as a positioning sensor. It can still work without GPS and has excellent robustness and accurate positioning, which is conducive to establishing more accurate areas and performing path planning.

[0065] (6) When the environment is relatively large, the present invention determines the CPU occupancy rate by the device power consumption so as to arrange the task unloading and allocation more reasonably. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 It is a schematic diagram of the system structure of the present invention;

[0067] Figure 2Flowchart of map construction and path planning for robots;

[0068] Figure 3 The path planned by the path planning algorithm adopted by the present invention;

[0069] Figure 4 It is a flow chart of voice control;

[0070] Figure 5 This is a flow chart of the speech recognition program;

[0071] Figure 6 This is a node tree diagram when the voice controls the movement of the car in the present invention;

[0072] Figure 7 Configure diagrams for multi-machine communication;

[0073] Figure 8 For the map built in rviz;

[0074] Fig. 9 Figure 2. Simulation results enhanced for speech recognition.

[0075] Fig.10 A framework diagram for integrating fog computing into robots;

[0076] Fig.11 This is a diagram of the fog computing framework;

[0077] Fig.12 Image of a robot launching a fog network. DETAILED DESCRIPTION

[0078] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0079] An intelligent voice-controlled navigation system based on fog computing technology, the system is arranged on a navigation robot, such as Figure 1 As shown, including:

[0080] 1) Sensor module: used to obtain surrounding environment data.

[0081] 2) Mapping module: Based on SLAM, it uses the data transmitted by the sensor module to draw a complete map of the external environment.

[0082] The mapping module specifically performs the following steps:

[0083] During the movement of the robot, the encoder and IMU are used to calculate the odometer information, and the robot's motion model is used to obtain the initial estimate of the robot's position and posture.

[0084] The laser data obtained by the laser sensor mounted on the robot is combined with the observation model to accurately correct the robot's posture and obtain the precise positioning of the robot;

[0085] Based on precise positioning, laser data is added to the grid map to control the robot's movement in the environment, and laser data is continuously added to eventually complete the construction of the entire scene map.

[0086] The point cloud data generated during the mapping process is transmitted to the cloud for storage and calculation, and the computing tasks are assigned to the cloud to reduce the memory usage on the robot side.

[0087] 3) Path planning module: used for path planning and navigation.

[0088] After mapping the real environment, the robot needs to be tested on navigation tasks.

[0089] The key technologies mainly involved include global map, self-positioning, path planning, motion control and environmental perception. The theoretical knowledge involved in the global map is SLAM. In ROS, the present invention adopts gmapping for the implementation of SLAM. At the same time, in order to complete SLAM, the perception sensor adopted is a laser radar. Self-positioning uses the function package for positioning provided by ROS: amcl. For path planning, including global path planning and local path planning, path planning is carried out based on the existing map information, that is, static path planning, to find an optimal path from the starting point to the target point, and the RRT global path planning algorithm is used; the path planning algorithm continuously generates state points randomly in the search space. If the point is in a collision-free position, the node closest to the node in the search tree is found as the reference node, and it is extended from the reference node until the distance between the target node and the search tree is terminated within a certain range.

[0090] The RRT global path planning algorithm process includes:

[0091] 3-1) Initialize an environment, including a map, a starting point, and an end point;

[0092] 3-2) Randomly sample state point x from the environment rand ;

[0093] 3-3) Find the node x closest to the sampling point from the constructed tree near ;

[0094] 3-4) x near and x randConnect them, the direction of this connection is the direction of the tree, set a step size, Stepsize as the step size of the tree growth once, grow a step size in the direction of tree growth, and a new node x will be generated at the end of the growth new ;

[0095] 3-5) Determine from x near to x rand Whether it passes through the obstacle, if so, abandon the new node, if not, then x new The node is added to the tree;

[0096] 3-6) Start looping from 3-2), randomly sampling state points from the environment, and looking for the distance x from the tree near The node closest to the node is x new ;

[0097] 3-7) Start the tree growth, at x near and x rand Grow a step in the connection direction to the new node x new ;

[0098] 3-8) Determine from x near to x rand Whether to pass through the obstacle, if so, abandon the point. If not, set x new The node is added to the tree;

[0099] 3-9) Repeat the above tree growth process until the distance from the new node generated by the tree to the target point is less than one step length, then terminate the tree growth.

[0100] The process of mapping and path planning is as follows Figure 2 As shown. Figure 3 As shown, the thick black frame represents a randomly generated obstacle, and the thicker black line represents a path plan generated by the above algorithm.

[0101] 4) Voice control module: used to respond to user needs based on topic publishing, and to achieve human-computer interaction and voice recognition enhancement.

[0102] like Figure 4 As shown in the figure, when the robot recognizes the wake-up word, it will automatically enter the standby mode and wait for the next instruction. If the wake-up word is not recognized, the robot will remain in sleep mode. Each instruction issued to the robot has a keyword. If the robot recognizes the keyword of the instruction, it will execute the corresponding driver program to complete the function to be implemented. After the previous instruction is executed, the robot will return to standby mode and wait for the next instruction. If the received command is to shut down, the robot will enter sleep mode and wait for the next wake-up.

[0103] The voice control module combines the map built by the gmapping service in the SLAM module and the target point location information obtained by the navigation service to control the robot by voice. The voice recognition terminal sends a wake-up signal and prepares for voice input. After receiving the voice signal, the SDK's voice recognition function is called in the Publisher main loop. After successful recognition, the resultFlag variable is set to publish the recognized string.

[0104] Specifically, Figure 5 As shown, first initialize ROS and declare Publisher and Subscriber. Subscriber receives the wake-up signal, determines whether wakeFlag is equal to 1, and calls the sub-function Wake(). Otherwise, it outputs Logout. If it is, it calls demo_mic() to wait for voice input. The waiting time is adjusted to 10 seconds. After detecting voice input, it starts to call the iFlytek SDK voice recognition function to determine whether resultFlag is 0. If it is, VoiceWordsPub.publish(msg) is completed and Logout is output. Otherwise, it directly outputs Logout.

[0105] Figure 6 The diagram shows the correlation between nodes and topics when the car is controlled by voice. Move_base is a node related to navigation and path planning, and voice_goal is a node that issues instructions for the car to reach the target point through voice. The two nodes communicate through the two topics move_base / status and move_base / goal. Move_base / goal is used to send the target point to the car's move_base node, and the car will feedback the current status through move_base / status. When the car is controlled by voice, multi-machine communication is involved. Figure 7 As shown in the figure, the process of setting up multi-machine communication in ROS includes checking the IP addresses of the host and slave through ifconfig, adding the IP address and host name of the other host in / etc / hosts / , testing whether the communication is normal through ping, and finally adding the ROS host name address and its own name in ~ / .bashrc. The car is issued with the command to reach the specified location through voice, and the slave receives the voice information, transmits the target command to the host through the topic, and the car plans a new path according to the current road conditions. Figure 8 Maps built for robots.

[0106] In the actual environment, there is a lot of noise in the environment where the robot is located, so noise reduction processing is necessary. In the present invention, speech recognition enhancement uses LMS adaptive filter for noise reduction, which specifically includes the following steps:

[0107] 4-1) Initialize the filter W(k):

[0108]

[0109] 4-2) Calculate the estimated value of the filter output: y(k) = W T (k)X(k);

[0110] 4-3) Calculate the estimated error: e(k) = d(k) - y(k);

[0111] 4-4) Update the weight coefficient at time k+1: W(k+1)=W(k)+μe(k)X(k);

[0112] 4-5) Update k and repeat 4-2)-4-4) until the convergence condition is met.

[0113] Among them, the updating method of the weight coefficient in 4-4) is obtained by the following method:

[0114] Assume that the input signal sequence for speech recognition enhancement is x i (n), the expected output signal is d(n), and the error signal is defined as:

[0115]

[0116] Where M represents the filter order, w i is the weight coefficient, by finding the optimal weight coefficient w i , so that the error signal e(n) is minimized;

[0117] The iterative formula of the weight coefficient is:

[0118] W(k+1)=W(k)-μ▽(k)

[0119] Where W(k) represents the weight coefficient; μ is the convergence factor, which is used to control the convergence rate; ▽(k) represents the gradient factor; and the convergence factor value range is as follows:

[0120]

[0121] Among them, λ max is the maximum eigenvalue of the correlation matrix. When it tends to infinity, the weight vector converges to the optimal Wiener solution;

[0122] Use iterative descent method to solve the convergence factor and gradient factor;

[0123] The gradient factor expression is calculated as:

[0124]

[0125] Then, the iterative formula of weight coefficient is updated as follows:

[0126] W(k+1)=W(k)+2μe(k)X(k)

[0127] Where X(k) represents the input.

[0128] Speech recognition enhancement algorithm simulation diagram Fig. 9 As shown in the figure, a noise-free speech is input, Gaussian noise is added to the speech, and the LMS algorithm is used to achieve noise reduction and speech enhancement. After the algorithm simulation converges, the speech signal after noise reduction is obtained.

[0129] The method of responding to user needs based on topic publishing includes the following steps:

[0130] The PC side obtains the voice information set by the user;

[0131] The navigation robot and the PC communicate through the network to determine the IP addresses between them;

[0132] Modify the respective environment variables;

[0133] Write the publisher and subscriber codes, publish the topic name to the navigation robot, and the navigation robot will perform related operations after receiving it.

[0134] 5) Task offloading and allocation module: fog computing technology is used for task offloading and allocation.

[0135] This embodiment uses the EdgeX Foundry edge computing framework as the implementation platform for fog computing. EdgeX is a general open framework that is open and compatible enough to allow any device to be digitized by defining Device, Device Service, and Deviceprofile. The container uses Docker, which has many advantages, such as more efficient use of system resources, faster startup time, consistent operating environment, continuous payment and deployment, easier migration, and easier maintenance and expansion downloads.

[0136] The EdgeX Foundry microservice collection consists of four microservice layers and two enhanced basic system services. The four microservice layers include a series of services from physical domain data collection to information domain data processing. The other two basic system services provide support services for the four service layers. The four microservice layers from the physical layer to the application layer are: device service layer (device service), core service layer (core service), supporting service layer (supporting service), and open service layer (export service). Two enhanced basic services: security and system management.

[0137] In EdgeX, the South Side and North Side are distinguished. The South Side refers to the source of data from all IoT devices, sensors, actuators, etc. The North Side refers to the cloud, where data is stored, aggregated, analyzed, and integrated. In this design, the laser radar, camera, and voice input sensor on the robot serve as the South Side, and the cloud server serves as the North Side of EdgeX.

[0138] The framework diagram for integrating fog computing into robots is as follows Fig.10 As shown in the figure. Its framework diagram consists of five parts: hardware platform, software platform, robot operating system (ROS is used here), support layer and application layer. The hardware platform is mainly composed of robot sensors, communication hardware platform and robot hardware platform. The software platform is used to manage fog nodes, including Linux and Android operating systems, communication software, fog nodes and service discovery, etc. When using OpenLTE, the communication software includes eNode B and EPC functions. The robot operating system uses ROS, on which the robot modeling, architecture and debugging functions can be completed. Relying on the nodes and communications of ROS itself, there is no need to consider how to use multi-threaded programming. The support layer is mainly used to support application security issues in the fog computing process. The application layer mainly applies fog computing technology to currently popular fields, such as robot SLAM, artificial intelligence, etc.

[0139] like Fig.11 , Fig.12As shown, multiple robots can share the resources obtained by each sensor, and the robots are connected to the fog network through a wireless network. Here, the lidar is mainly used for SLAM mapping. At the same time, the robot's visual sensor also has a camera. Target recognition and semantic segmentation of objects in the environment can be performed during the movement of the robot. These are achieved through deep learning algorithms. The training and updating of deep learning algorithms are mainly achieved through the cloud, and the trained models are downloaded to the fog node or the robot itself. In this invention, the postman tool was used to test the use of http service to send data to EdgeX and obtain sensor data respectively. A robot equipped with a sensor is used as an edge device. A personal PC is used as a fog node and cloud for fog computing. The unbuntu operating system on the robot collects sensor data and uploads it to the EdgeX node. EdgeX transmits data to the robot. The test completes the communication between the robot and the edge device. If necessary, the EdgeX fog node will transmit data to the cloud. The specific process of its construction steps is as follows:

[0140] Build a value descriptor in Postman to describe a value, tell EdgeX what format the data is in and what to use to mark the data;

[0141] Upload a device configuration file. EdgeX integrates device configuration files as a way to easily add new devices. A device configuration file is essentially a template that describes a device, its data format, and supported commands.

[0142] When applying fog computing technology to voice navigation robots, the problem that needs to be considered is how to effectively offload tasks and allocate tasks when the amount of data is large, so as to reduce the large amount of processing calculations on the robot side. Its task offloading model is as follows: Fig.11 As shown in Figure 1, the model includes the cloud network layer, fog network layer and device layer. The top layer is the central cloud which is far away from the IoT devices. It has the most abundant computing and storage resources and serves as the network control and scheduling center. It also provides task offloading services. Due to the limited local resources, the heterogeneous devices in the device layer send offloading requests to the fog network layer. The fog network layer is composed of a large number of fog nodes, which mainly provide computing and storage services for IoT devices. It is also responsible for transmitting IoT device data to the central cloud for computing and storage. First, the IoT device from the device layer sends an offloading request. Secondly, the ID makes an offloading decision based on the network performance constraints or channel condition constraints. The offloading task can be processed by the fog node or the cloud to meet the business needs. Finally, the fog node or the cloud returns the calculation result to the ID to complete the task offloading process.

[0143] The task offloading and allocation module optimizes task processing by minimizing the total node energy, which includes the total offloading energy, the fog node total energy and the total transmission energy, wherein:

[0144] The total unloading energy is expressed as:

[0145]

[0146] Among them, u n represents the computing task of mobile robot n, K represents the number of robots, and b n represents the power consumption when the robot is unloaded, represents unloading energy;

[0147] The total energy of the fog node is expressed as:

[0148]

[0149] Among them, N c It represents the number of CPU cycles required to calculate 1 bit in the fog node, E f represents the energy consumption per cycle of fog node calculation, u n N c E f The computational energy consumption of offloading tasks for the mobile robot n;

[0150] The total transmission energy is expressed as:

[0151]

[0152] Among them, O n is the task analysis result of mobile robot n;

[0153] The objective function of the task offloading and allocation module is determined according to the total energy of the node:

[0154]

[0155] The task unloading and allocation module needs to determine the CPU occupancy before performing task unloading and allocation. The CPU occupancy is determined based on the relationship model between device power consumption and device CPU occupancy rate:

[0156] p=p idle +η(p max -p idle )

[0157] Among them, p idle is the power consumption when the CPU is idle, p max is the CPU full load power consumption, η is the resource occupancy rate of the device CPU;

[0158] The relationship between device power consumption and CPU frequency is:

[0159] E=k(f) 2 C

[0160] Among them, f is the main frequency of the device CPU, C is the CPU cycle required to process data, and k is the effective conversion capacitance of the device CPU chip.

[0161] 6) Communication module: used to realize communication between the navigation robot and the cloud and PC.

[0162] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. An intelligent voice-controlled navigation system based on fog computing technology, characterized in that: The system is arranged on a navigation robot and comprises: Sensor module: used to obtain surrounding environment data; Mapping module: Based on SLAM, it uses the data transmitted by the sensor module to draw a complete map of the external environment; Path planning module: used for path planning and navigation; Voice control module: used to respond to user needs based on topic publishing, and to achieve human-computer interaction and voice recognition enhancement; Task offloading and allocation module: uses fog computing technology for task offloading and allocation; Communication module: used to realize communication between the navigation robot and the cloud and PC; The task offloading and allocation module optimizes task processing by minimizing the total node energy, which includes the total offloading energy, the fog node total energy and the total transmission energy. The total unloading energy is expressed as: in, u n Represents a mobile robot n The computational tasks K Indicates the number of robots, b n represents the power consumption when the robot is unloaded, represents unloading energy; The total energy of the fog node is expressed as: in, N c Indicates the number of CPU cycles required to calculate 1 bit in the fog node. E f represents the energy consumption per cycle calculated by the fog node, u n N c E f For mobile robots n The computational energy consumption of offloaded tasks; The total transmission energy is expressed as: in, O n It is a mobile robot n The results of the task analysis; The objective function of the task offloading and allocation module is determined according to the total energy of the node: 。 2. According to claim 1, an intelligent voice-controlled navigation system based on fog computing technology is characterized in that: The mapping module specifically performs the following steps: During the movement of the robot, the encoder and IMU are used to calculate the odometer information, and the robot's motion model is used to obtain the initial estimate of the robot's position and posture. The laser data obtained by the laser sensor mounted on the robot is combined with the observation model to accurately correct the robot's posture and obtain the precise positioning of the robot; Based on precise positioning, laser data is added to the grid map to control the robot's movement in the environment, and laser data is continuously added to eventually complete the construction of the entire scene map.

3. The intelligent voice-controlled navigation system based on fog computing technology according to claim 1, characterized in that: The point cloud data generated during the mapping process is transmitted to the cloud for storage and calculation, and the computing tasks are assigned to the cloud to reduce the memory usage on the robot side.

4. The intelligent voice-controlled navigation system based on fog computing technology according to claim 1, characterized in that: The path planning includes global path planning and local path planning. Path planning is performed based on existing map information, i.e., static path planning, to find an optimal path from the starting point to the target point. The RRT global path planning algorithm is used. The path planning algorithm continuously generates state points randomly in the search space. If the point is located in a collision-free position, the node closest to the point in the search tree is found as the reference node, and the path is extended from the reference node until the distance between the target node and the search tree is within a certain range.

5. The intelligent voice-controlled navigation system based on fog computing technology according to claim 1, characterized in that: The voice control module combines the map constructed by the gmapping service in the SLAM module and the target point location information obtained by the navigation service to control the robot by voice. The voice recognition terminal sends a wake-up signal and prepares for voice input. After receiving the voice signal, the SDK's voice recognition function is called in the Publisher main loop. After successful recognition, the resultFlag variable is set to publish the recognized string.

6. The intelligent voice-controlled navigation system based on fog computing technology according to claim 1, characterized in that: The speech recognition enhancement adopts LMS adaptive filter to perform noise reduction, which specifically includes initialization, calculation of estimated value of filter output, calculation of estimation error and updating weight coefficient at the next moment.

7. The intelligent voice-controlled navigation system based on fog computing technology according to claim 6, characterized in that: Assume that the input signal sequence for speech recognition enhancement is x i ( n ), the expected output signal is d ( n ), the error signal is defined as: in, M represents the filter order, w i is the weight coefficient, by finding the optimal weight coefficient w i , so that the error signal e ( n ) minimum; The iterative formula of the weight coefficient is: in, represents the weight coefficient; is the convergence factor, which is used to control the convergence rate; Represents the gradient factor; the convergence factor value range is as follows: in, is the maximum eigenvalue of the correlation matrix. When it tends to infinity, the weight vector converges to the optimal Wiener solution; Use iterative descent method to solve the convergence factor and gradient factor; The gradient factor expression is calculated as: Then, the iterative formula of weight coefficient is updated as follows: in, Represents input.

8. The intelligent voice-controlled navigation system based on fog computing technology according to claim 1, characterized in that: The method of responding to user needs based on topic publishing includes the following steps: The PC side obtains the voice information set by the user; The navigation robot and the PC communicate through the network to determine the IP addresses between them; Modify the respective environment variables; Write the publisher and subscriber codes, publish the topic name to the navigation robot, and the navigation robot will perform related operations after receiving it.

9. The intelligent voice-controlled navigation system based on fog computing technology according to claim 1, characterized in that: The task unloading and allocation module needs to determine the CPU occupancy before performing task unloading and allocation. The CPU occupancy is determined based on the relationship model between device power consumption and device CPU occupancy rate: in, p idle is the power consumption when the CPU is idle, p max The power consumption of the CPU at full load. η The resource usage of the device CPU; The relationship between device power consumption and CPU frequency is: in, f is the device CPU main frequency, C The CPU cycles required to process the data, k Effectively convert capacitors for device CPU chips.

Images