A quadruped robot and its control system

Abstract

This invention belongs to the field of quadruped robots, and relates to a quadruped robot and its control system. It includes a head structure and leg structures mounted on the body structure. The invention integrates knowledge from computer science, bionics, biomechanics, materials science, and other disciplines in its external design, breaking through the conventional wheeled structure based on the SLAM algorithm. It designs a non-wheeled robot whose movement speed can match the data acquisition speed of the SLAM algorithm, making innovation and breakthroughs in the field of combining SLAM technology with indoor legged mobile robots. Based on SLAM localization technology, it can achieve dynamic target following. Combined with a six-microphone array voice module and an offline speech recognition engine, it realizes technologies such as sound source localization, voice navigation, and voice remote control for the intelligent robot, further expanding the experience level of human-machine voice intelligent interaction.

Description

Technical Field

[0001] This invention belongs to the field of quadruped robots and relates to a quadruped robot and its control system. Background Technology

[0002] Currently, autonomous route planning and real-time tracking during robot movement have become a hot topic in the robotics field. The realization of this technology is also a crucial prerequisite for determining whether a fully autonomous robot can perform autonomous behavior. With autonomous driving technology gradually becoming a social phenomenon, SLAM (Simultaneous Localization and Mapping), which enables autonomous localization, obstacle avoidance, and route planning, has entered the public eye. This invention proposes a method for real-time localization and mapping based on LiDAR and ultra-wideband positioning technology to achieve multi-level data fusion. This method enables an independently designed quadruped robot to perform autonomous route planning and tracking in a completely unknown environment, and achieves intelligent human-machine interaction functions such as sound source localization, voice navigation, and voice control. This robot has the following application scenarios:

[0003] 1. Family companionship

[0004] Due to various factors, many people living alone cannot alleviate their anxiety about being alone through the companionship of pets. This robot can autonomously track people by detecting human voices, thus meeting people's daily pet companionship needs and becoming a valuable assistant in the home.

[0005] 2. Healthcare

[0006] Nowadays, healthcare is gradually gaining attention. Robots can be equipped with medical kits and use the YOLO algorithm to identify the types of medicines and store them in a database, enabling people to access and transport medicines.

[0007] 3. Monitoring and Management

[0008] The robot's head is equipped with a camera, enabling it to survey and explore its environment. Furthermore, the robot's path planning allows for patrolling and comprehensive monitoring, reducing the consumption of human and financial resources. It is suitable for numerous scenarios, including large factories, construction sites, and deserts.

[0009] 4. Transportation

[0010] Robots can transport designated items by planning routes within a given environment, making them suitable for logistics and transportation, especially for sorting small packages. Furthermore, robots can cope with harsh environments, reducing transportation costs.

[0011] In conclusion, as a novel quadruped robot, its unique advantages and broad application prospects will undoubtedly enable it to play an important role in family companionship, healthcare, and monitoring and transportation.

[0012] Invention patent CN115610539A designs an autonomous route robot, which is a tracked robot with external wheels. The external wheels are set inside the connecting frame. In use, the control board on the inner control frame is connected to the drive motor through the connecting wire, so that the motor is powered on and rotates, which drives the bevel gear at the top to rotate. Then, the bevel gears on both sides drive the drive wheels on both sides to rotate. The drive wheels drive the circulating track connected on the outside to rotate. The tracked drive robot can perform path planning in all terrains.

[0013] Invention patent CN108360428A designs a sanitation robot and its control method for automatic garbage pickup, transportation, and unloading. After the robot automatically moves to a designated location through path planning, the front-end garbage-picking robotic arm picks up the garbage and stores it in a garbage bin. The sanitation robot of this invention can inspect, transport, and unload garbage in public places such as scenic spots, parks, and residential areas in real time, eliminating the need for sanitation workers to clean regularly every day, and significantly improving its intelligence.

[0014] Overall, significant progress has been made in robot localization and path planning. Currently, wheeled and tracked robots are the most common, and no related patents have been published in the field of quadruped robots. Summary of the Invention

[0015] The above-mentioned technical problems of the present invention are mainly solved by the following technical solutions:

[0016] A quadruped robot, characterized in that it includes a head structure and leg structures disposed on a body structure.

[0017] In the aforementioned quadruped robot, the leg structure is a foot key, which adopts a hemispherical structure. The combination of the cylinder and the cuboid with two through holes in the hemispherical structure is used to install the screws of the servo motor. The foot connecting key acts as a bridge to support the foot key and is connected to the body structure through the leg key.

[0018] The aforementioned quadruped robot's body structure includes a thoracic cover plate, an upper thoracic plate, a lower thoracic plate, and a tail connector. The thoracic cover plate has cutouts on both sides for storing sensors, and the front side is open to facilitate the insertion of additional devices, forming a snap-fit ​​with them. The upper thoracic plate is connected to the cover plate using screws and through holes. The rectangular slot in the middle is a battery compartment for storing the robot's power supply battery. The blank rectangular areas around the perimeter are for easy wiring. The four small rectangular through holes on the outermost side are for connecting to the servos of the leg structure. The outer ring through holes are used for connecting the tail connector to the thoracic cover plate. The remaining through holes are used to fix the robot's internal sensors and control board.

[0019] In the aforementioned quadruped robot, the head structure has a hollow center to facilitate integration with servo motors.

[0020] A control system for a quadruped robot, characterized in that it comprises:

[0021] Mobile robot control module: a serial 16-channel servo control board, used for connection via COM port, to zero all servos of the robot and control the rotation of the servos;

[0022] ROS-RVIZ module: Used to display various states of the robot during robot control, such as the movement state and force state of the robotic arm, so as to display various data and facilitate operators to debug and control.

[0023] Sensor unit module: includes laser sensor, gyroscope, and camera; the laser sensor is used to determine the position and measure the distance to both sides of the robot, enabling the robot to measure the distance to sudden objects in the surrounding environment and avoid direct collisions with other objects; the gyroscope is used for the robot's self-balancing adjustment, real-time detection of the robot's balance status, and maintaining a stable speed during tracking; the camera is connected to the main control board via USB to monitor the real-time environment and to identify and track human beings;

[0024] Data processing module: This module processes the acquired sound and human images. The acquired sound is translated into hexadecimal encoding using a speech recognition sensor, and then conditional control statements are used to initiate the process. The human images are converted into binary matrices, and the matrices are processed and calculated to obtain the corresponding graphics.

[0025] Real-time display module: Used to display the robot's current status information;

[0026] Data acquisition module: used to acquire the robot's current position coordinates, motion information, and voice reception;

[0027] Communication module: Used for robot to interact with users, recognize specific voice signals from the human body, and provide responses and corresponding instructions.

[0028] In the aforementioned control system, the data processing module performs positioning correction based on SLAM and UWB positioning fusion, including:

[0029] The coordinates are collected and preprocessed, and the data includes LiDAR data and UWB positioning data.

[0030] For LiDAR data, it is defined as a particle in a particle swarm and initialized.

[0031] After updating the location using the odometer, the local extreme values ​​are calculated and evaluated. Based on the evaluation results, a choice is made as to whether to update the location directly or to use UWB positioning data for location updates.

[0032] In the aforementioned control system, when calculating the local extremum of the position after updating the position via the odometer, the position is obtained through maximum likelihood estimation, and:

[0033] If the calculated local extremum is the local extremum at the current position, then the local mean at the current position is retained as the local extremum.

[0034] If the calculated local extremum is not the local extremum of the current position, it is compared with the current local extremum. If the error is within 5%, it is considered to be the result of equipment inaccuracy, and the local extremum of the current position is still taken. If the error exceeds 5%, the calculated local extremum is used.

[0035] In the aforementioned control system, when the data processing module performs human detection path planning, the MediaPipe algorithm is used to identify and dynamically track human posture. MediaPipe detects images in motion and, based on angle markings, uses a large number of motion images as a training set to achieve posture and behavior recognition. Image ranging and spatial geometry methods are then used to convert the data into three-dimensional coordinate information. These coordinate sets are then used to solve for the angles of each joint of the human body using spatial vector methods, and this information is fed back to the robot. The robot then determines the human posture based on images captured by its real-time camera and, combined with a path planning algorithm, approaches the human body. Finally, the robot's voice recognition sensor is used to locate the sound source, specifically including:

[0036] In a static state, image data captured by a camera is acquired; the source location of human speech is obtained; laser SALM and UWB are activated to acquire current environmental information and perform joint mapping; for image data, the acquired RGB color image is first converted to grayscale image using OpenCV, and then dilation and contraction processing is performed; the MediaPipe library is used to recognize and detect the human body. If a human body is detected, the human body's posture is determined and combined with the mapping output as coordinate position 1, and the position is adjusted using a gyroscope to prepare to move towards coordinate position 1; if no human body is detected, coordinate position 1 is set to empty; the coordinates of the sound source position are processed and judged. If the sound source position coincides with coordinate position 1, the position moves towards coordinate position 1; if they do not coincide, the sound source position replaces coordinate position 1, and the orientation is adjusted using a gyroscope to prepare to move towards the sound source position;

[0037] While in motion, the system acquires image data captured by the camera; during movement, it updates and optimizes the image in real time to ensure the reliability of the current image; it uses a laser sensor to determine the distance between the laser beams and obstacles in real time, and takes reactive actions when necessary, shifting laterally away from the current path, then using a gyroscope to rotate the current orientation and replan the path; for image data, the acquired RGB color image is first converted to grayscale using OpenCV, and then dilation and contraction processing is performed; the MediaPipe library is used to identify and detect human bodies. If a human body is detected, the posture of the human body is determined and combined with the map output as coordinate position 2, and the gyroscope is used to adjust the position, preparing to move towards coordinate position 1; if no human body is detected, no action is taken, and the system moves towards the sound source coordinate position 1 acquired when stationary; if a termination command is encountered, the movement stops immediately.

[0038] In the aforementioned control system, under static conditions, the specific path planning includes:

[0039] The UWB positioning system uses an algorithm based on the Time Difference of Arrival (TDOA). This algorithm transmits signals from a tag to multiple base stations. Based on the time difference of the tag signal arriving at each base station, the distance difference between the tag and each base station is obtained. Finally, the tag coordinates are calculated using a hyperbolic model. The UWB base stations are installed in fixed locations and their coordinates are known, namely BSi(x1,y1) (i = 1, 2, 3). The coordinates of the tag MS to be determined are (x, y), and the distance from the tag to the base station is di (i = 1, 2, 3). The current position is calculated using formulas (1) and (2).

[0040]

[0041]

[0042] The laser radar emitter module emits a laser beam at an angle. The beam hits the surface of the object to be measured and is reflected back to the receiver module. The laser radar receiver module based on the triangulation model usually obtains the static position and coordinates of the CMOS camera as the initial state.

[0043] In the aforementioned control system, under dynamic conditions, the specific path planning includes:

[0044] First, the odometer data is converted into robot pose change data using a wheeled odometer. This data is then fed into a Bayesian filter to initially calculate the predicted pose, obtaining the pose information for the next moment. The prediction accuracy depends entirely on the accuracy of the wheeled odometer. Next, a particle filter method is used to fuse radar observations to correct the local positioning results. The local positioning results are then transformed into the global coordinate system through coordinate transformation, and further Gaussian blurring is performed within the global coordinate system. Finally, UWB observations are used to correct the positioning results in the global coordinate system. The position of each particle is corrected using the information observed by UWB to obtain new mean and variance. The new coordinates and variance are then fed back into the established global map to update the map. This effectively eliminates the accumulated error in LiDAR SLAM through UWB, suppresses the relatively scattered positioning results, and ensures the accuracy of the mapping.

[0045] This invention has the following advantages: 1. The quadrupedal hemispherical design ensures the robot's freedom of movement and speed during operation. Even in the event of collisions or falls, it can maintain its working state without external adjustment, ensuring basic stability. 2. The multi-data fusion algorithm optimizes the traditional SLAM algorithm, reducing the cumulative error caused by turning in traditional SLAM and effectively improving the coordinate measurement accuracy in non-outdoor environments. This provides a good opportunity for the application of functional robots for indoor mobility, where demand is rapidly increasing. 3. The design integrates knowledge from computer science, bionics, biomechanics, materials science, and other disciplines, breaking through the conventional wheeled structure based on SLAM algorithms. It designs a non-wheeled robot whose movement speed matches the data acquisition speed of the SLAM algorithm, making an innovation and breakthrough in the field of combining SLAM technology with indoor legged mobile robots. 4. Based on SLAM positioning technology, dynamic target following can be achieved. Combined with a six-microphone array voice module and an offline speech recognition engine, it enables technologies such as sound source localization, voice navigation, and voice remote control for intelligent robots, further expanding the experience level of human-computer voice intelligent interaction. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the overall appearance and structure of the present invention.

[0047] Figure 2 This is a schematic diagram of the head bond structure.

[0048] Figure 3 This is a schematic diagram of the foot-like key structure.

[0049] Figure 4 This is a schematic diagram of the foot connection key structure.

[0050] Figure 5 This is a schematic diagram of the leg key structure.

[0051] Figure 6 This is a schematic diagram of the thoracic ventral sac structure.

[0052] Figure 7 This is a schematic diagram of the upper thoracic plate.

[0053] Figure 8 This is a schematic diagram of the inferior thoracic plate.

[0054] Figure 9 This is a schematic diagram of the tail connector structure.

[0055] Figure 10 This is a schematic diagram of the MediaPipe algorithm.

[0056] Figure 11 This is a flowchart for human body tracking.

[0057] Figure 12 This is a schematic diagram of the robot's backend management platform.

[0058] Figure 13 This is a schematic diagram of the overall design framework of the control software.

[0059] Figure 14 This is a schematic diagram illustrating the principle of fusion between lidar and UWB.

[0060] Figure 15 This is a UWB location map.

[0061] Figure 16 This is a diagram illustrating the principle of lidar positioning.

[0062] Figure 17 This is a flowchart of the SLAM and UWB positioning fusion process. Detailed Implementation

[0063] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0064] Example:

[0065] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0066] In terms of overall structural design, this invention provides a novel quadruped robot, such as... Figure 1 The image shows the overall 3D structure of the robot, including the head structure, leg structure, and torso structure.

[0067] like Figure 2 The image shows the robot's head structure. We used a biomimetic modeling method to simulate and design a shape similar to a pet dog, with a hollowed-out center to facilitate integration with servo motors.

[0068] Figure 3 Foot key, Figure 4 Foot connection key Figure 5 These three parts together constitute the robot's leg structure. Figure 3 The foot key adopts a hemispherical structure, which makes it easy to adjust the mass distribution of the eccentric push device inside the robot. The combination of the cylinder and the cuboid with two through holes on it is to connect the screws of the servo motor. Figure 4 The foot connection key acts as a bridge to support the leg, and... Figure 3 and Figure 5 The connection features chamfered edges for added stability and aesthetics, while through holes at key locations ensure proper installation of the servo motors and guarantee the robot's flexibility. Figure 5 The robot's leg key, the cylindrical side of which connects to the servo motor, is located at... Figure 7 upper chest plate and Figure 8 The hollowed-out diagonal cuboid design between the lower thoracic plates ensures the key's load-bearing capacity while reducing material waste. This position is perpendicular to the overall thoracic cavity, ensuring the robot's range of motion and leg freedom.

[0069] Figure 6 Thoracic cover plate, Figure 7 upper chest plate, Figure 8 Inferior thoracic plate Figure 9 The tail section connects to form the robot's body. Figure 6 The thoracic cover is inspired by the fluid ramp of a car's exterior. The smooth, steep ramp structure ensures high space utilization and makes the robot more biomimetic. The hollowed-out sides are for storing sensors, such as laser sensors and gyroscopes for distance measurement, which are easy to modify. In addition, the front side is open to facilitate the installation of additional devices, forming a snap-fit ​​with the additional devices. Figure 7 The upper thoracic plate is connected to the through-hole with screws. Figure 6 The cover plate is connected, and the rectangular slot in the middle is the battery compartment to store the robot's power supply battery. The blank rectangular areas around it are auxiliary designs to facilitate wiring. The four small rectangular through holes on the outermost side are used to connect to the servo motors of the leg structure. The outer ring through holes are used to connect the tail connection key to the chest cover plate. The remaining through holes are used to fix the robot's internal sensors and control board. Figure 8 The lower and upper thoracic plates have the same length and width, and the surface is not porous. This layer mainly serves as an expansion layer. The external openings are the same as those in the upper cavity, and the interior is reserved for the position of sensors. Figure 9 The tail connector is used to fix the upper and lower plates of the thoracic cavity, fixing the width between the upper and lower plates of the robot. The rectangular space in the middle is the reserved position for the tail.

[0070] In human detection, the MediaPipe algorithm is used for human posture recognition and dynamic tracking. For example... Figure 10 As shown, the Mediapipe algorithm estimates the positions of various joints in the human body, treating it as a whole image and using a neural network for feature extraction. Each step defines different images and absolute coordinates within those images, calculates the node positions, and then transforms them into the same coordinate system. In motion, Mediapipe detects images in motion and, based on angle markings, uses a large number of motion images as a training set to achieve posture and behavior recognition. It then converts the image ranging and spatial geometry methods into three-dimensional coordinate information, and uses these coordinate sets to solve for the angles of each joint in the human body using the spatial vector method, providing feedback to the robot. Figure 11 By deploying Mediapipe on the robot and enabling the robot's camera, the robot can determine the human's posture based on images captured by the robot's real-time camera and use a path planning algorithm to approach the human. Combined with the robot's voice recognition sensor, it can achieve the effect of sound source localization.

[0071] With the continuous development of sensor technology, researchers have discovered that each single sensor has inherent technical limitations that cannot be overcome. For example, ultra-wideband positioning technology has good penetration capabilities, but it is greatly affected by metallic objects in the environment and generates significant noise; lidar operates stably, but it cannot penetrate for measurement, and the laser diverges when it encounters transparent media in the environment, leading to large positioning errors. However, indoor mobile robots not only operate in complex environments, but also require high accuracy and stability in positioning and navigation for their tasks. Therefore, high-precision positioning and navigation of indoor mobile robots cannot be achieved with a single sensor; it is necessary to adopt a multi-sensor data fusion approach to allow different sensors to compensate for each other's shortcomings. Therefore, this study uses a combined system based on UWB and lidar to investigate the positioning and navigation of mobile robots in complex indoor environments. The overall design framework of the control software is as follows: Figure 13 As shown.

[0072] SLAM (Single-Landed Awareness and Mapping) is characterized by low cost and rich environmental information acquisition, providing not only positioning information but also constructing corresponding point cloud maps. Due to the problem of error accumulation, a positioning method capable of providing absolute positioning information is needed to correct this. Ultra-wideband (UWB) positioning systems, with their high bandwidth and strong penetration capabilities, can provide absolute location services, making them suitable for indoor positioning applications. The fusion principle is as follows... Figure 14 As shown.

[0073] The UWB positioning system uses an algorithm based on Time Difference of Arrival (TDOA). This algorithm transmits signals from a tag to multiple base stations. Based on the time difference of the tag's signal arriving at each base station, the distance difference between the tag and each base station is obtained. Finally, the tag coordinates are calculated using a hyperbolic model. Compared with the Time of Arrival (TOA) algorithm, this algorithm does not require clock synchronization between the base station and the tag. It only requires clock synchronization between the reference, thus having the advantages of fewer communication cycles and higher positioning accuracy. Its positioning principle is shown in the figure.

[0074] The UWB base station is installed in a fixed location and its coordinates are known. The coordinates are BS i(x1,y1) (i=1,2,3). The coordinates of the tag MS to be obtained are (x,y). The distance from the tag to the base station is di (i=1,2,3). The location tag at a certain moment is calculated as shown in formula (1) and formula (2).

[0075]

[0076]

[0077] The principle of lidar for positioning is to obtain ranging information from the environment using a laser beam. LiDAR typically consists of three modules: a laser transmitter module, a receiver module, and a circuit module. Although the 2D lidar used in this study has a reduced measurement range and accuracy, it is lower in cost and more compact, better meeting the needs of mobile robots in complex indoor environments. Therefore, it is widely used in indoor mobile robot positioning and navigation. The principle of the triangulation model is shown in the figure. The lidar transmitter module emits a laser beam at an angle. The beam hits the surface of the object being measured and reflects back to the receiver module. The receiver module of a lidar based on the triangulation model is typically a CMOS camera. The positioning principle is as follows... Figure 16 As shown.

[0078] This paper presents a combined localization method that uses UWB observations to correct LiDAR SLAM predictions for a robot. First, a wheeled odometry system converts odometer data into robot pose change information, which is then fed into a Bayesian filter to initially calculate the predicted pose, obtaining the pose information for the next time step. The prediction accuracy depends entirely on the accuracy of the wheeled odometry system. Next, a particle filter method is used to fuse radar observations to correct the local localization results. Coordinate transformation is then applied to convert the local localization results to the global coordinate system, followed by further Gaussian blurring within the global coordinate system. Finally, UWB observations are used to correct the local localization results in the global coordinate system. The position of each particle is corrected using the information obtained from UWB observations to obtain new mean and variance. These new coordinates and variances are then fed back into the established global map to update the map. The advantage of this combined method is that it effectively eliminates accumulated errors in LiDAR SLAM through UWB while suppressing the dispersion of localization results, ensuring the accuracy of the mapping. The fusion process is as follows: Figure 17 As shown.

[0079] like Figure 13 The intelligent robot backend management interface program, developed based on PyQt and PySide, enables the management and use of robots, featuring four functions: graph construction, autonomous localization, dynamic tracking, and voice interaction. By configuring relevant modules, the robot can autonomously plan routes and track itself in real time during movement, allowing it to mimic pets and achieve companionship and intelligent interaction.

[0080] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A control system for a quadruped robot, characterized in that, The quadruped robot includes a body structure, a head structure and leg structures mounted on the body structure; the control system includes: Mobile robot control module: a serial 16-channel servo control board, used for connection via COM port, to zero all servos of the robot and control the rotation of the servos; ROS-RVIZ module: Used to display various states of the robot during robot control, including the movement state and force state of the robotic arm, so as to display various data and facilitate operators to debug and control. Sensor unit module: includes laser sensor, gyroscope, and camera; laser sensor is used for position determination, measuring the distance to both sides of the robot, enabling the robot to measure the distance to sudden objects in the surrounding environment and avoid direct collisions with other objects; gyroscope is used for robot self-balancing adjustment, real-time detection of the robot's balance status, and maintaining a stable speed during tracking; camera is connected to the main control board via USB, monitoring the real-time environment, recognizing and tracking human beings; Data processing module: This module processes the acquired sound and human images. It uses a voice recognition sensor to translate the sound into hexadecimal encoding and then uses conditional control statements to initiate the process. It converts the human image into a binary matrix, processes and performs calculations on the matrix to obtain the corresponding graphic. Real-time display module: Used to display the robot's current status information; Data acquisition module: used to acquire the robot's current position coordinates, motion information, and voice reception; Communication module: used for robot-user interaction, recognizing specific speech transmitted by the human body, and providing responses and corresponding instructions; wherein, the data processing module performs localization correction based on SLAM and UWB localization fusion, including: The coordinates are collected and preprocessed. The data processing module includes LiDAR data and UWB positioning data. For LiDAR data, it is defined as a particle in a particle swarm and initialized. After updating the location using the odometer, the local extreme values ​​are calculated and judged. Based on the judgment result, it is decided whether to directly update the location or use UWB positioning data to update the location. Specifically, when calculating the local extrema of the position after updating the position using odometry, the position is obtained through maximum likelihood estimation, and: If the calculated local extremum is the local extremum at the current position, then the local mean at the current position is retained as the local extremum. If the calculated local extremum is not the local extremum at the current location, it is compared with the current local extremum. If the error is within 5%, it is considered to be the result of equipment inaccuracy, and the local extremum at the current location is still taken. If the error exceeds 5%, the calculated local extremum is used.

2. The control system for the quadruped robot according to claim 1, characterized in that, The leg structure is a foot key, which adopts a hemispherical structure. The two through holes of the hemispherical structure are used to install the screws of the servo motor. The foot connecting key acts as a bridge to support the foot key and is connected to the body structure through the leg key.

3. The control system for the quadruped robot according to claim 1, characterized in that, The body structure includes a thoracic cover plate, an upper thoracic plate, a lower thoracic plate, and a tail connector. The thoracic cover plate has cutouts on both sides for inserting sensors, which snap together with additional devices. The upper thoracic plate is connected to the cover plate with screws and through holes. The rectangular slot in the middle is a battery compartment for storing the robot's power supply battery. The blank rectangular areas around the perimeter are for easy wiring. The four small rectangular through holes on the outermost side are for connecting to the servos of the leg structure. The outer ring through holes are used for connecting the tail connector to the thoracic cover plate. The remaining through holes are used to fix the robot's internal sensors and control board.

4. The control system for the quadruped robot according to claim 1, characterized in that, The head structure has a hollow center, which facilitates its integration with a servo motor.

5. The control system for the quadruped robot according to claim 1, characterized in that, When the data processing module performs human detection path planning, it uses the MediaPipe algorithm to identify and dynamically track human posture. MediaPipe detects images in motion and, based on angle markings, uses a large number of motion images as a training set to achieve posture and behavior recognition. It then converts the image ranging and spatial geometry methods into three-dimensional coordinate information. From these coordinate sets, it uses the spatial vector method to solve for the angles of each joint of the human body and feeds this information back to the robot. Finally, it determines the human posture based on images captured by the robot's real-time camera and, combined with the path planning algorithm, approaches the human. Finally, it uses the robot's speech recognition sensor to achieve sound source localization, specifically including: In a static state, capture image data from the camera; obtain the source location of human speech; The system initiates laser SLAM and UWB to acquire current environmental information and performs joint mapping. For image data, the acquired RGB color image is first converted to grayscale using OpenCV, and then dilation and contraction processing is performed. The MediaPipe library is used to identify and detect human bodies. If a human body is detected, its posture is determined and combined with the map output as coordinate position 1. The gyroscope is used to adjust the position, preparing to move towards coordinate position 1. If no human body is detected, coordinate position 1 is left empty. The coordinates of the sound source position are processed and judged. If the sound source position coincides with coordinate position 1, the system moves towards coordinate position 1. If they do not coincide, the sound source position is used instead of coordinate position 1, and the gyroscope is used to adjust the orientation, preparing to move towards the sound source position. While in motion, the system acquires image data captured by the camera; during movement, it updates and optimizes the image in real time to ensure the reliability of the current image; it uses a laser sensor to determine the distance between the laser beams and obstacles in real time, and takes reactive actions when necessary, shifting laterally away from the current path, then using a gyroscope to rotate the current orientation and replan the path; for image data, the acquired RGB color image is first converted to grayscale using OpenCV, and then dilation and contraction processing is performed; the MediaPipe library is used to identify and detect human bodies. If a human body is detected, the posture of the human body is determined and combined with the map output as coordinate position 2, and the gyroscope is used to adjust the position, preparing to move towards coordinate position 1; if no human body is detected, no action is taken, and the system moves towards the sound source coordinate position 1 acquired when stationary; if a termination command is encountered, the movement stops immediately.

6. The control system for the quadruped robot according to claim 5, characterized in that, In a static state, the specific path planning includes: The UWB positioning system employs an algorithm based on the time difference of arrival (TDOA). This algorithm involves a tag transmitting signals to multiple base stations. The distance difference between the tag and each base station is calculated based on the time difference of the tag's signal arrival at each station. Finally, the tag's coordinates are determined using a hyperbolic model. The UWB base stations are installed in fixed locations with known coordinates, which are B, C, and D. i (x i ,y i (i = 1,2,3), the coordinates of the tag MS to be determined are (x,y), and the distance from the tag to the base station is d. i (i = 1, 2, 3), use formulas (1) and (2) to calculate the current position: （1） （2） The laser radar transmitter module emits a laser beam at an angle. The beam hits the surface of the object to be measured and is reflected back to the receiver module. The laser radar receiver module, based on a triangulation model, uses a CMOS camera to obtain the static position and coordinates as the initial state.

7. The control system for the quadruped robot according to claim 5, characterized in that, In a dynamic state, specific path planning includes: First, the odometer data is converted into robot pose change data using a wheeled odometer. This data is then fed into a Bayesian filter to initially calculate the predicted pose, obtaining the pose information for the next moment. The prediction accuracy depends entirely on the accuracy of the wheeled odometer. Next, a particle filter method is used to fuse radar observations to correct the local positioning results. The local positioning results are then transformed into the global coordinate system through coordinate transformation, and further Gaussian blurring is performed within the global coordinate system. Finally, UWB observations are used to correct the positioning results in the global coordinate system. The position of each particle is corrected using the information observed by UWB to obtain new mean and variance. The new coordinates and variance are then fed back into the established global map to update the map. This effectively eliminates the accumulated error in LiDAR SLAM through UWB, suppresses the relatively scattered positioning results, and ensures the accuracy of the mapping.

Images