A water rescue robot master control system

Abstract

The application belongs to the technical field of water rescue equipment control, and provides a water rescue robot master control system.The system comprises: a master CPU module; a communication module, integrated with a LoRa command control link and a Mesh self-organizing network data transmission link, forming a dual-means communication system; a navigation positioning module, integrated with Beidou and inertial navigation positioning and fusion vision and hearing positioning; a path planning module, based on a grid map for global path planning; a path tracking module, using an improved integral line-of-sight method for accurate path tracking; a collision avoidance module, fusing a dynamic window method and a speed obstacle method to realize dynamic obstacle avoidance; a propeller drive control module, using a double closed loop control structure to coordinate control of the heading and the speed; and a life buoy throwing control module, used for autonomous decision and execution of life buoy throwing.The application, through multi-module cooperation and advanced algorithm fusion, significantly improves the autonomy, accuracy, collaborative operation ability and reliability of the rescue robot in complex sea conditions.

Description

Technical Field

[0001] This invention relates to the field of water rescue equipment control technology, and in particular to a main control system for a water rescue robot. Background Technology

[0002] Existing water rescue systems primarily rely on manually operated lifeboats or remotely controlled rescue equipment for rescue operations. Traditional lifeboats require rescuers to personally navigate to the location of the person in the water, while remotely controlled rescue equipment, although controllable remotely, generally uses satellite positioning. Furthermore, existing equipment mostly employs fixed communication methods.

[0003] The existing technologies mentioned above have the following shortcomings: First, the positioning accuracy is low. Relying solely on satellite navigation results in significant positioning errors in a maritime environment, making it difficult to meet the accuracy requirements for close-range rescue. Second, they lack intelligent collision avoidance capabilities, making them prone to collisions with other vessels or obstacles in complex sea conditions. Third, their path planning capabilities are weak, failing to autonomously plan the optimal rescue path based on the real-time environment. Fourth, their communication methods are limited, hindering reliable data sharing and task coordination when multiple robots are working collaboratively. Fifth, the throwing of rescue devices relies on manual judgment, leading to inaccurate timing and affecting the success rate of rescue operations. Summary of the Invention

[0004] This invention provides a main control system for a water rescue robot to overcome the shortcomings of existing technologies.

[0005] This invention provides a main control system for a water rescue robot, comprising:

[0006] The main control CPU module integrates a multi-core processor and a neural network processing unit, which is used to coordinate multiple modules and realize intelligent recognition functions; The communication module is electrically connected to the main control CPU module. The communication module includes a LoRa command and control link unit and a Mesh self-organizing network data transmission link unit to form a dual-means communication system for realizing data sharing when multiple robots work together. The navigation and positioning module is electrically connected to the main control CPU module. The navigation and positioning module adopts a two-level positioning combination method, which achieves precise positioning through the fusion of visual recognition and sound source positioning. It includes: a first-level positioning unit, which integrates a Beidou GNSS receiver and a MEMS inertial measurement unit, and achieves coarse positioning through a multi-source signal fusion algorithm; and a second-level positioning unit, which includes a binocular camera and a microphone array. The path planning module is connected to the main control CPU module and adopts a global path planning algorithm based on a grid map. The path tracking module interacts with the path planning module and the navigation and positioning module. The path tracking module adopts an improved integral line-of-sight guidance strategy, including a forward-looking distance adaptive adjustment unit and a dimension-reduced state observer. The forward-looking distance adaptive adjustment unit dynamically adjusts the forward-looking distance according to the lateral tracking error and the navigation speed. The dimension-reduced state observer estimates the time-varying sideslip angle online and compensates for it, so that the robot can travel according to the planned path. The collision avoidance module is electrically connected to the main control CPU module. It integrates the dynamic window method and the speed obstacle method. The collision avoidance module includes a speed obstacle judgment unit, a feasible speed window calculation unit, and an obstacle avoidance decision unit. The speed obstacle judgment unit judges the collision risk by calculating whether the relative speed vector between the robot and the obstacle falls within the speed obstacle cone. The feasible speed window calculation unit removes speed vectors located within the finite speed obstacle cone based on the dynamic window. The obstacle avoidance decision unit generates a safe local obstacle avoidance path. The propeller drive control module is electrically connected to the main control CPU module and adopts a dual closed-loop control structure. The dual closed-loop control structure includes a kinematic control outer loop and a dynamic control inner loop. The kinematic control outer loop calculates the heading and speed commands based on the position error. The dynamic control inner loop includes a heading and speed coordination control unit and a thruster optimization control unit. The heading and speed coordination control unit calculates the lateral turning force and the left and right longitudinal thrust based on the heading and speed commands. The thruster optimization control unit uses a multi-objective optimization method to solve for the optimal left and right nozzle velocities. The lifebuoy throwing control module is electrically connected to the main control CPU module. It is used to autonomously determine the throwing timing based on the relative position and posture of the robot and the person in the water, and push the lifebuoy shell out.

[0007] According to the present invention, a main control system for a water rescue robot includes a LoRa command and control link unit for transmitting instructions between the robot and the control console, and a Mesh self-organizing network data transmission link unit that adopts a distributed, decentralized, multi-hop relay communication method to support dynamic networking of point-to-point, point-to-many, and many-to-many connections, for realizing data sharing during multi-robot collaborative operations.

[0008] According to the present invention, a main control system for a water rescue robot is provided, wherein the first-level positioning unit is an inertial navigation module including a GNSS subunit and a MEMS subunit, the GNSS subunit supporting multi-frequency signal reception, and the MEMS subunit including a three-axis accelerometer and a three-axis gyroscope; The second-level positioning unit includes a binocular camera and a microphone array. The binocular camera includes a visible light camera and a near-infrared camera. The optical axes of the two cameras in the binocular camera are parallel and arranged horizontally. The microphone array includes multiple high-sensitivity microphones. The spatial filtering characteristics of the microphone array are used to estimate the location of the sound source by employing the time difference of arrival method or the controllable waveform positioning method.

[0009] According to the present invention, a main control system for a water rescue robot, when the path planning module performs global path planning, converts the marine environment into a grid map containing obstacle information, performs expansion processing on the edges of obstacles, expands the search in multiple directions from the starting grid, calculates the distance from each grid to the starting point and records the parent node information, and when the ending grid is found, it traces back to the starting point from the ending point according to the parent node information to generate the globally optimal path.

[0010] According to the main control system of a water rescue robot provided by the present invention, in the path tracking module, the forward sight distance adaptive adjustment unit dynamically adjusts the forward sight distance according to the lateral tracking error and the navigation speed; the dimensionality reduction state observer estimates the time-varying sideslip angle online and compensates for it by defining intermediate variables including sideslip angle and lateral tracking error, and the expected heading angle is calculated from the path tangent angle, lateral tracking error, forward sight distance and sideslip angle estimate.

[0011] According to the main control system of a water rescue robot provided by the present invention, in the collision avoidance module, the speed obstacle judgment unit is used to perform safety expansion processing on the obstacle, calculate the relative speed vector between the robot and the obstacle, and determine whether the angle between the relative speed vector and the obstacle collision vector is less than the critical angle. If it is less than the critical angle, there is a risk of collision. The feasible speed window calculation unit introduces the maximum allowable collision occurrence time, constructs a finite speed obstacle cone, and determines whether the speed vector is a dangerous speed. The obstacle avoidance decision unit combines the evaluation function of the dynamic window method, including the target distance function, the obstacle distance function, and the speed evaluation function, and selects the optimal speed vector after comprehensive scoring.

[0012] According to the main control system of a water rescue robot provided by the present invention, the expression for the expansion radius for safety expansion processing of obstacles is:

[0013] in, Where is the radius of expansion. Let the radius be the obstacle. For the robot radius, For safety margin; The expression for the critical angle is:

[0014] in, It is the critical angle. The distance between the robot and the obstacle; The expression for determining the dangerous speed is:

[0015] in, Location of the obstacle. For the robot's position, Let be the relative velocity vector between the robot and the obstacle. This represents the maximum permissible time for a collision to occur.

[0016] According to the present invention, a main control system for a water rescue robot is provided, wherein the kinematic control outer loop establishes a position error state equation and uses a nonlinear controller to calculate the heading and speed commands based on the position error; the dynamic control inner loop includes a heading and speed coordination control unit and a thruster optimization control unit, wherein the heading and speed coordination control unit calculates the lateral steering force and the left and right longitudinal thrust based on the heading and speed commands, and the thruster optimization control unit uses a multi-objective optimization method to solve for the left and right nozzle velocities.

[0017] According to the present invention, a main control system for a water rescue robot is provided, wherein the main control CPU module has a heterogeneous eight-core processor architecture, and the neural network processing unit provides 6 TOPS of AI computing power, supporting real-time inference of various neural network models including CNN convolutional neural network, RNN recurrent neural network and LSTM long short-term memory network.

[0018] The main control system for a water rescue robot provided by the present invention further includes: The task allocation module, electrically connected to the main control CPU module, is used to allocate tasks when multiple robots work together. The task allocation module adopts an integer programming method, comprehensively considers constraints such as the distance between the robot and the person who fell into the water, the robot's battery power, communication distance, and the number of lifebuoys carried, establishes a multi-objective optimization function, and solves the optimal task allocation scheme.

[0019] This invention provides a main control system for a water rescue robot. Firstly, by employing a dual-means communication system, combining a LoRa command and control link with a Mesh self-organizing network data transmission link, it ensures reliable command transmission while enabling data sharing and collaborative operation among multiple robots. This effectively solves the communication conflicts and data transmission instability problems of traditional single communication methods in complex sea conditions, significantly improving the communication reliability and collaborative efficiency of the rescue system. Secondly, this invention employs a two-stage positioning combination method. The first stage utilizes multi-source signal fusion of BeiDou GNSS and an inertial measurement unit for coarse positioning, while the second stage achieves precise positioning through audio-visual fusion of binocular cameras and a microphone array. This significantly improves positioning accuracy, meeting the high-precision requirements of close-range rescue and ensuring the robot can accurately approach people in the water. Thirdly, this invention employs a collision avoidance strategy that combines dynamic windowing and velocity obstacle methods. It constructs finite velocity obstacle cones and calculates feasible velocity windows in real time. This invention enables robots to anticipate collision risks and avoid them in time. Compared to traditional single collision avoidance methods, obstacle avoidance response is faster, safety margin is greater, and the probability of collision accidents is effectively reduced. In addition, the improved integral line-of-sight guidance strategy of this invention, through adaptive adjustment of forward-looking distance and real-time compensation of sideslip angle by a dimensionality-reduced state observer, allows the robot to accurately track the planned path even under wave interference and water flow influence. The path tracking error is reduced compared to traditional methods, ensuring the stability and controllability of the rescue process. This invention also provides a dual closed-loop control structure. Through the coordinated operation of the kinematic outer loop and the dynamic inner loop, precise control of heading and speed is achieved, enabling the robot to maintain a stable attitude in complex sea conditions. This creates favorable conditions for the accurate throwing of lifebuoys. Moreover, the intelligent throwing decision unit can autonomously determine the best throwing time based on the relative position, attitude, and speed of the robot and the person in the water. Compared with manual judgment, the throwing success rate is improved, significantly enhancing rescue efficiency and success rate. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This invention provides a schematic diagram of the overall structure of the main control system for a water rescue robot. Figure 2 A schematic diagram of the navigation and positioning module structure in the main control system of a water rescue robot provided by the present invention; Figure 3 This invention provides a schematic diagram of the collision avoidance module structure in the main control system of a water rescue robot.

[0022] Reference numerals: 100, Main control CPU module; 200, Communication module; 300, Navigation and positioning module; 400, Path planning module; 500, Path tracking module; 600, Collision avoidance module; 700, Propeller drive control module; 800, Lifebuoy throwing control module; 310. First-level positioning unit; 311. GNSS subunit; 312. MEMS subunit; 320. Second-level positioning unit; 321. Binocular camera; 322. Microphone array; 610. Speed ​​obstacle judgment unit; 620. Feasible speed window calculation unit; 630. Obstacle avoidance decision unit. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, embodiments of this invention, and should not be construed as limiting the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. In the description of this invention, it should be understood that the terminology used is for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0024] The embodiments of the present invention are described below with reference to the figures.

[0025] like Figure 1 As shown, the present invention provides a main control system for a water rescue robot, comprising: The main control CPU module 100 integrates a multi-core processor and a neural network processing unit to coordinate multiple modules and realize intelligent recognition functions.

[0026] Furthermore, this invention uses the main control CPU module as the brain and decision-making center of the entire system. This module establishes a stable and reliable electrical connection and data interaction channel with the communication module, navigation and positioning module, path planning module, path tracking module, collision avoidance module, propeller drive control module, and lifebuoy throwing control module through internal circuits and a data bus. The main control CPU module is responsible for overall task scheduling and resource coordination. Its integrated multi-core processor and dedicated neural network processing unit (NPU) constitute a powerful heterogeneous computing core. The multi-core processor is responsible for handling complex system tasks and routine calculations, while the NPU is dedicated to performing real-time inference of deep learning models, providing the necessary computing power support for the robot's intelligent recognition functions, such as visual recognition of people falling into the water.

[0027] The communication module 200 is electrically connected to the main control CPU module 100. The communication module 200 includes a LoRa command and control link unit and a Mesh self-organizing network data transmission link unit, forming a dual-means communication system for realizing data sharing when multiple robots work together.

[0028] Furthermore, the communication module provides dual-means communication, ensuring reliable transmission of commands and data in various environments. Specifically, the LoRa command and control link unit is responsible for exchanging key commands (such as start, return, and mode switching) and status information with the rear control console. Its advantages lie in its long transmission distance and low power consumption, ensuring smooth basic command. Meanwhile, the Mesh self-organizing network data transmission link unit constructs a distributed, decentralized, and resilient network, supporting dynamic networking and multi-hop relay for point-to-point, point-to-many, and many-to-many connections. When multiple robots work collaboratively, this link enables them to communicate directly with each other, sharing location, perception data, and other information, forming collective intelligence. Even if a communication path is interrupted, the network can automatically reconstruct, greatly improving the system's robustness and collaborative efficiency in operations across vast water areas.

[0029] The navigation and positioning module 300 is electrically connected to the main control CPU module 100. The navigation and positioning module 300 adopts a two-level positioning combination method to achieve precise positioning through the fusion of visual recognition and sound source positioning. It includes: a first-level positioning unit 310, which integrates a Beidou GNSS receiver and a MEMS inertial measurement unit to achieve coarse positioning through a multi-source signal fusion algorithm; and a second-level positioning unit 320, which includes a binocular camera 321 and a microphone array 322.

[0030] Furthermore, the navigation and positioning module employs a two-stage combined positioning mode, achieving seamless integration from wide-area search to precise close-range approach. The first-stage positioning unit integrates a high-precision BeiDou GNSS receiver and a MEMS inertial measurement unit. Through advanced multi-source signal fusion algorithms, it complements the absolute position information provided by satellites with the continuous attitude and motion information provided by the inertial navigation system, effectively overcoming the problems caused by satellite signal loss or jumps at sea, and achieving stable, coarse positioning on turbulent sea surfaces. When the robot navigates to a certain range of the person in the water, the second-stage positioning unit is activated. Its binocular camera calculates the precise distance and orientation of the target using stereo vision principles, while the microphone array uses the time difference of sound waves arriving at different microphones or beamforming technology to direct the person's cries for help. Visual and auditory information are fused in the main control CPU, ultimately achieving meter-level or even sub-meter-level positioning accuracy at close range, far exceeding that of a single technology, providing crucial support for the final rescue operation.

[0031] The path planning module 400 is connected to the main control CPU module 100 and adopts a global path planning algorithm based on a grid map.

[0032] Furthermore, the path planning module of this invention is responsible for planning a safe and efficient global route from the starting point to the location of the person who fell into the water. This module operates based on a grid map. First, it digitizes the complex marine environment (including static obstacles such as islands and reefs) into individual grid cells, and appropriately expands the edges of obstacles to allow for a safety margin. Then, it executes a planning algorithm, starting from the starting grid cell and expanding the search in multiple directions, calculating and updating the shortest distance from each grid cell to the starting point in real time, while recording its parent node information. When the search algorithm reaches the grid cell containing the target point, it generates a globally optimal safe path by backtracking through these parent nodes.

[0033] The path tracking module 500 interacts with the path planning module 400 and the navigation and positioning module 300. The path tracking module 500 adopts an improved integral line-of-sight guidance strategy, including a forward-looking distance adaptive adjustment unit and a dimension-reduced state observer. The forward-looking distance adaptive adjustment unit dynamically adjusts the forward-looking distance according to the lateral tracking error and the navigation speed. The dimension-reduced state observer estimates the time-varying sideslip angle online and compensates for it, so that the robot can travel according to the planned path.

[0034] Furthermore, the path tracking module provided by this invention aims to ensure that the robot can accurately navigate along the planned path. This invention employs an improved integral line-of-sight guidance strategy, which includes a forward-looking distance adaptive adjustment unit. This unit dynamically adjusts the forward-looking distance based on the robot's real-time navigation speed and the lateral tracking error along the path. When the error is large or the speed is slow, a smaller forward-looking distance is used to quickly approach the path; when the error is small or the speed is fast, the forward-looking distance is increased to ensure smooth navigation and avoid overshoot. In addition, regarding hull sideslip caused by wind, waves, and currents at sea, the path tracking module's built-in dimensionality-reduced state observer can estimate the time-varying sideslip angle online in real time and compensate for it when calculating the desired heading angle, thereby significantly improving the path tracking accuracy and stability under adverse sea conditions.

[0035] The collision avoidance module 600, electrically connected to the main control CPU module 100, integrates the dynamic window method and the speed obstacle method. The collision avoidance module 600 integrates the dynamic window method and the speed obstacle method and includes a speed obstacle judgment unit 610, a feasible speed window calculation unit 620, and an obstacle avoidance decision unit 630. The speed obstacle judgment unit 610 judges the collision risk by calculating whether the relative speed vector between the robot and the obstacle falls within the speed obstacle cone. The feasible speed window calculation unit 620 removes speed vectors located within the finite speed obstacle cone based on the dynamic window. The obstacle avoidance decision unit 630 generates a safe local obstacle avoidance path.

[0036] Furthermore, the collision avoidance module is used to deal with dynamic obstacles that suddenly appear during navigation (such as other vessels). Specifically, this invention integrates the dynamic window method with the speed obstacle method. The speed obstacle judgment unit calculates the relative velocity vector between the robot and the obstacle and determines whether it falls within a speed obstacle cone formed by the obstacle's shape and safety radius. If so, a collision risk is identified. The feasible speed window calculation unit then eliminates dangerous speed options located within a finite speed obstacle cone (considering the maximum allowable collision time) from all possible speeds generated by the dynamic window method. Finally, the obstacle avoidance decision unit comprehensively evaluates the remaining safe speeds based on factors such as target distance, obstacle distance, and speed, and selects the optimal speed command to generate a local obstacle avoidance path that is both safe and maximizes the path towards the target.

[0037] The propeller drive control module 700 is electrically connected to the main control CPU module 100 and adopts a dual closed-loop control structure. The dual closed-loop control structure includes a kinematic control outer loop and a dynamic control inner loop. The kinematic control outer loop calculates the heading and speed commands based on the position error. The dynamic control inner loop includes a heading and speed coordination control unit and a thruster optimization control unit. The heading and speed coordination control unit calculates the lateral steering force and the left and right longitudinal thrust based on the heading and speed commands. The thruster optimization control unit uses a multi-objective optimization method to solve for the optimal left and right nozzle velocities.

[0038] Furthermore, the propeller drive control module is responsible for controlling the robot's motion. For the propeller drive control module, this invention employs a dual closed-loop control structure to achieve precise decoupling and coordinated control of heading and speed. The outer loop's kinematic control loop calculates the heading and speed commands required for path tracking based on the instructions from the path tracking module and the real-time position feedback from the navigation and positioning module. The inner loop's dynamic control loop is more refined, further divided into: a heading and speed coordination control unit that converts the outer loop commands into the lateral steering force and longitudinal thrust of the left and right propellers required to achieve the motion; and a propeller optimization control unit that, based on a multi-objective optimization model, calculates the optimal rotational speed or nozzle velocity commands for the left and right propellers to most accurately and efficiently generate the required thrust and steering force.

[0039] The lifebuoy throwing control module 800 is electrically connected to the main control CPU module 100. It is used to autonomously determine the throwing timing based on the relative position and posture of the robot and the person in the water, and push the lifebuoy shell out.

[0040] Furthermore, the lifebuoy throwing control module receives precise relative position and attitude information from the navigation and positioning module, as well as the main control CPU's judgment of the lifebuoy's state based on visual recognition. Combining this information, the module autonomously decides the optimal throwing time, ensuring that the throwing command is triggered at the moment when the robot's posture is most stable and the lifebuoy is most likely to be obtained by the lifebuoy after being thrown. The built-in lifebuoy is then accurately and quickly launched towards the target area through methods such as electronic ignition.

[0041] The LoRa command and control link unit is used to transmit commands to the console, and the Mesh self-organizing network data transmission link unit adopts a distributed, decentralized, multi-hop relay communication method, supporting point-to-point, point-to-many, and many-to-many dynamic networking, which is used to realize data sharing when multiple robots work together.

[0042] Furthermore, the communication module of this invention adopts a dual-means communication system design, including two independent but collaborative communication subsystems: a LoRa command and control link unit and a Mesh self-organizing network data transmission link unit. The LoRa command and control link unit is electrically connected to the main control CPU module via a serial port interface, primarily responsible for command transmission between the control console and the rescue robot. It employs low-power, long-distance communication technology to ensure the reliability and real-time performance of command transmission under complex sea conditions. The Mesh self-organizing network data transmission link unit is also electrically connected to the main control CPU module, employing a distributed, decentralized, multi-hop relay communication method. Each node in the system acts as both a data sender and a data relayer, supporting dynamic networking modes such as point-to-point, point-to-many, and many-to-many. It enables beyond-line-of-sight communication by relaying data between a robot node and the control console when the distance is too great.

[0043] The dual-means communication architecture of this invention enables efficient data sharing when multiple robots are working together. Each robot can exchange location information, environmental perception data, and task execution status, which significantly improves the reliability and coverage of the communication system. Even if a single communication link is interfered with or blocked, the system can still maintain a connection through another communication method, avoiding the communication interruption problem that is easy to occur in traditional single communication methods. At the same time, the multi-hop relay function greatly extends the communication distance, enabling rescue robots to work together over a wider area.

[0044] In specific implementation, the LoRa command and control link unit adopts the domestic E70-DTU (433NW30) module, with dimensions of 82mm×62mm×25mm, a default serial port baud rate of 115200bps, an operating temperature range of -40℃ to +85℃, and a communication distance of 2.5km; the Mesh self-organizing network data transmission link unit adopts the domestic SmartMesh airborne module SINOSUN-SM, which supports 20MHz and 40MHz bandwidth, has a receiving sensitivity of -103dBm, an average latency of 7ms, a maximum effective transmission distance of 30km, adopts a terminal interface, has good shock resistance, and an operating temperature range of -40℃ to +80℃.

[0045] like Figure 2 As shown, the first-level positioning unit 310 is an inertial navigation module including a GNSS subunit 311 and a MEMS subunit 312. The GNSS subunit 311 supports multi-frequency signal reception, and the MEMS subunit 312 includes a three-axis accelerometer and a three-axis gyroscope. The second-level positioning unit 320 includes a binocular camera 321 and a microphone array 322. The binocular camera 321 includes a visible light camera and a near-infrared camera. The optical axes of the two cameras in the binocular camera 321 are parallel and horizontally arranged. The microphone array 322 includes multiple high-sensitivity microphones. The spatial filtering characteristics of the microphone array are used to estimate the direction of the sound source using the time difference of arrival method or the controllable waveform positioning method.

[0046] Specifically, the navigation and positioning module of this invention adopts a two-level positioning combination method, achieving progressive high-precision positioning from long distance to short distance through the organic combination of coarse and precise positioning. The first-level positioning unit is an inertial navigation module including a GNSS subunit and a MEMS subunit. The GNSS subunit receives Beidou satellite signals through an antenna, supporting multi-frequency signal reception to improve positioning accuracy and anti-interference capability. The MEMS subunit includes a three-axis accelerometer and a three-axis gyroscope. The three-axis accelerometer measures the robot's linear acceleration in three orthogonal directions, and the three-axis gyroscope measures the robot's angular velocity around three axes. The data from the two subunits are transmitted to the main control CPU module for fusion processing through an internal data bus. The second-level positioning unit includes a binocular camera and a microphone array. The binocular camera includes a visible light camera and a near-infrared camera. The two cameras are fixedly mounted on the front of the robot, with parallel and horizontally arranged optical axes. The depth information of the target is obtained by calculating the parallax of the same target in the images of the two cameras. The visible light camera is used for visual positioning under normal daylight conditions, and the near-infrared camera is used for visual positioning under night or low light conditions. The microphone array consists of multiple high-sensitivity microphones arranged in a specific geometric pattern. Each microphone is connected to the audio acquisition interface of the main control CPU module via a signal line. By utilizing the spatial filtering characteristics of the microphone array, the location of the sound source is estimated by calculating the time difference of the sound signal arriving at each microphone or by using a controllable waveform localization method.

[0047] The beneficial effect of the two-level positioning combination method of the present invention is that the first-level positioning solves the problem of long-distance navigation, enabling the robot to accurately drive to the approximate area where the person who fell into the water is located. The second-level positioning achieves precise positioning at close range through audio-visual fusion, improving the positioning accuracy to the meter level or even the sub-meter level. Especially when the impact of sea waves is large and the satellite positioning accuracy decreases, the audio-visual fusion positioning can still maintain high accuracy, ensuring that the robot can accurately approach the person who fell into the water.

[0048] In specific implementation, the first-level positioning unit adopts the UniStrong UGM532 GNSS MEMS inertial integrated navigation module. The BeiDou GNSS receiver supports multi-frequency signal reception, and the RTK positioning accuracy reaches 2cm+1ppm. The MEMS inertial measurement unit includes a three-axis accelerometer and a three-axis gyroscope, with an attitude accuracy of 0.5° and a heading accuracy of 0.1°. The multi-source signal fusion algorithm adopts the extended Kalman filter method to establish a 19-dimensional system state vector, including position error, velocity error, attitude angle error, gyroscope drift error, accelerometer zero bias error, spatial lever error, and time asynchrony error. Through iterative updates of the state equation and measurement equation, the optimal fusion of BeiDou positioning information and inertial navigation information is achieved.

[0049] When performing global path planning, the path planning module 400 converts the marine environment into a grid map containing obstacle information, expands the obstacle edges, expands the search in multiple directions from the starting grid, calculates the distance from each grid to the starting point and records the parent node information, and when the ending grid is found, it traces back to the starting point from the ending point based on the parent node information to generate the globally optimal path.

[0050] Furthermore, the path planning module is connected to the main control CPU module via an internal data bus, receiving position information from the navigation and positioning module and environmental information from the perception system. During global path planning, the marine environment is first converted into a grid map containing obstacle information. The continuous ocean space is discretized into regular grid cells, each assigned a value of 0 or 1, where 0 represents a navigable area and 1 represents an obstacle area. Obstacle edges are inflated, meaning that grid cells within a certain range around the obstacle are also marked as obstacles. The inflated distance is typically set to the robot's safe radius plus a certain safety margin to prevent the robot from colliding with obstacles during navigation.

[0051] The path search process starts from the starting grid and expands in eight directions: east, west, south, north, and southeast, northeast, southwest, and northwest. It calculates the cumulative distance from each searched grid to the starting grid and records the parent node information of that grid, i.e., which grid it originated from. The search process employs a shortest distance-first strategy, selecting the unvisited grid with the smallest distance to the starting grid for expansion at each step, until the destination grid is reached.

[0052] Once the destination grid is found, the search proceeds upwards from the destination grid by tracing back the parent node information of each grid in turn until the starting grid is reached. This generates a complete path from the starting point to the destination grid, which is the globally optimal path.

[0053] The beneficial effects of the path planning method of the present invention are that it can comprehensively consider all obstacles in the marine environment, and the planned path can effectively avoid static obstacles such as islands and reefs, while ensuring the shortest path length, reducing sailing time and energy consumption. At the same time, the grid map representation is simple and intuitive, with high computational efficiency, which can meet the requirements of real-time path planning.

[0054] In the path tracking module 500, the forward look distance adaptive adjustment unit dynamically adjusts the forward look distance based on the lateral tracking error and the navigation speed; the dimensionality reduction state observer estimates and compensates for the time-varying sideslip angle online by defining intermediate variables including the sideslip angle and the lateral tracking error, and the expected heading angle is calculated from the path tangent angle, the lateral tracking error, the forward look distance, and the estimated sideslip angle.

[0055] Furthermore, the path tracking module of the present invention interacts with the path planning module and the navigation and positioning module, receiving the planned target path from the path planning module and the robot's real-time position and heading information from the navigation and positioning module.

[0056] Specifically, the module employs an improved integral line-of-sight guidance strategy, comprising two key sub-units: a forward-looking distance adaptive adjustment unit and a dimension-reduced state observer. The forward-looking distance adaptive adjustment unit dynamically adjusts the forward-looking distance based on the current lateral tracking error and navigation speed. The lateral tracking error refers to the vertical distance between the robot's current position and the target path. When the lateral tracking error is large, the forward-looking distance is set smaller, enabling the robot to quickly approach the target path. When the lateral tracking error is small and the navigation speed is high, the forward-looking distance is set larger to avoid path oscillations caused by frequent robot heading adjustments.

[0057] The reduced-dimensional state observer estimates and compensates for the time-varying sideslip angle online by defining an intermediate variable that includes the sideslip angle and lateral tracking error. The sideslip angle is the deviation angle between the robot's actual motion direction and its heading caused by external disturbances such as ocean currents and sea winds. This angle is difficult to measure directly. The observer estimates the real-time value of the sideslip angle by continuously observing the robot's motion state and calculating using a mathematical model. The calculation of the desired heading angle comprehensively considers the path tangent angle, lateral tracking error, forward look distance, and the estimated sideslip angle. The path tangent angle represents the tangential direction of the target path at the current point. The desired heading angle is obtained by comprehensively calculating these parameters and then passed to the propeller drive control module to achieve heading control.

[0058] Furthermore, the operation process of the path tracking module of this invention will be explained below with reference to specific algorithms and parameter calculation formulas.

[0059] In operation, the path tracking module of this invention aims to calculate the desired heading angle based on the initial global path using an improved integral line-of-sight guidance strategy, input the desired heading angle into the heading controller for tracking control, and obtain the initial navigation command for the rescue robot.

[0060] First, calculate the tangent angle of the path based on the projection point of the rescue robot's current position on the preset approach path.

[0061] In calculating the path tangent direction angle, this invention plans a preset approach path for the rescue robot based on the target's position and direction of movement in the target proximity parameters. This path is a curve pointing from the rescue robot's current position to the predicted position of the target. Specifically, the current position of the rescue robot is read. Find the point closest to the current position on the preset proximity path and use it as the projection point. Subsequently, the tangent to the path is calculated at the projection point. The direction of the tangent is determined by the derivative of the path at that point, while the tangent direction angle is... The angle between the tangent direction and the X-axis is given by... calculate, and Let be the derivatives of the path in the Y and X directions at the projection point.

[0062] Secondly, the path tracking module calculates the lateral tracking error of the rescue robot relative to the preset approach path, and calculates the adaptive forward-looking distance based on the lateral tracking error and the rescue robot's speed; the expression for the adaptive forward-looking distance is:

[0063] in, For minimum forward sight distance, The first adjustment parameter is... This is the second adjustment parameter. This is the third adjustment parameter. For lateral tracking error, Speed ​​of the rescue robot.

[0064] To address tracking errors, this invention calculates the current position of the rescue robot. The vertical distance to the preset approach path is used as the lateral tracking error. Specifically, a perpendicular line is drawn from the current position to the path; the intersection of the perpendicular line and the path is the foot of the perpendicular; the length of the perpendicular line is the lateral tracking error. When the rescue robot is on the right side of the path Take a positive value, and a negative value if it is on the left; in addition, the vertical distance is calculated using the formula for the distance from a point to a curve. For a parameterized path, traverse the path points to find the point closest to the rescue robot, and the distance from that point to the rescue robot is the vertical distance. .

[0065] Subsequently, the path tracking module constructs an observer in a reduced-dimensional state and outputs an estimated sideslip angle value through the observer; The auxiliary state update equation for the observer is:

[0066] in, This is the updated auxiliary state for the observer. For observation error, For observer gain, The current heading angle, The tangent direction angle of the path; The expression for the estimated sideslip angle is:

[0067] in, This is the estimated sideslip angle for the output.

[0068] Finally, based on the adaptive forward look-ahead distance and the estimated sideslip angle, the desired heading angle is calculated to obtain the initial navigation command; The expression for the desired heading angle is:

[0069] in, The desired heading angle.

[0070] When obtaining initial navigation commands, this invention calculates the ratio of lateral tracking error to adaptive forward look-ahead distance. The angle correction is obtained by comparing the arctangent values. Then, the path tangent direction angle and angle correction are added together, and the sideslip angle estimate is subtracted to obtain the desired heading angle. The final expected heading angle constitutes the initial navigation command, and the command data structure contains the expected heading angle value and a timestamp.

[0071] The beneficial effect of the improved path tracking method is that the adaptive adjustment of the forward sight distance enables the robot to achieve good tracking performance under different navigation speeds and different tracking errors, avoiding the problems of oscillation at high speeds and slow tracking speed at low speeds caused by the traditional fixed forward sight distance method. The online estimation and compensation of the sideslip angle significantly improves the robustness of the system to external disturbances, and the robot can still accurately track the planned path even in severe sea conditions with large winds, waves and currents.

[0072] like Figure 3 As shown, in the collision avoidance module 600, the speed obstacle judgment unit 610 is used to perform safety expansion processing on the obstacle and calculate the relative speed vector between the robot and the obstacle, and determine whether the angle between the relative speed vector and the obstacle collision vector is less than the critical angle. If it is less than the critical angle, there is a risk of collision. The feasible speed window calculation unit 620 introduces the maximum allowable collision occurrence time, constructs a finite speed obstacle cone, and determines whether the speed vector is a dangerous speed. The obstacle avoidance decision unit 630 combines the evaluation function of the dynamic window method, including the target distance function, the obstacle distance function, and the speed evaluation function, and selects the optimal speed vector after comprehensive scoring.

[0073] Furthermore, the collision avoidance module is electrically connected to the main control CPU module, receiving obstacle information from the perception system, including parameters such as the obstacle's position, speed, and size. The module integrates two collision avoidance strategies: the dynamic window method and the speed obstacle method, and includes three sub-units: a speed obstacle judgment unit, a feasible speed window calculation unit, and an obstacle avoidance decision unit.

[0074] The speed obstacle judgment unit first performs safety expansion processing on the detected obstacles, adding the robot's radius plus a safety margin to the actual radius of the obstacle, thus treating the robot as a point mass and simplifying subsequent calculations. Next, it calculates the relative velocity vector between the robot and the obstacle (the robot's velocity minus the obstacle's velocity). It then checks if the angle between this relative velocity vector and the vector connecting the obstacle and the robot is less than a critical angle. The critical angle is calculated using an arcsine function, with the parameter being the ratio of the expanded radius to the distance from the robot to the obstacle. If the angle is less than the critical angle, it indicates a collision risk. The feasible speed window calculation unit introduces the concept of a maximum permissible collision time based on the dynamic window method, constructing a finite speed obstacle cone. It does not simply exclude all possible collision speeds, but only those speeds that will collide within the maximum permissible time. It calculates the ratio of the magnitude of the difference between the robot and obstacle position vectors to the magnitude of the relative velocity vector, and checks if this ratio is less than the maximum permissible collision time. If it is less, the speed is judged as a dangerous speed and removed from the feasible speed window. The obstacle avoidance decision unit combines multiple evaluation functions of the dynamic window method to make a comprehensive decision. The target distance function evaluates the distance between the robot and the target position after the robot travels at the speed. The obstacle distance function evaluates the distance between the robot and the nearest obstacle after the robot travels at the speed. The speed evaluation function evaluates the magnitude of the speed. The evaluation functions are weighted and summed to obtain a comprehensive score. The speed vector with the best comprehensive score is selected as the obstacle avoidance speed command.

[0075] The beneficial effects of the fusion collision avoidance method of the present invention are that the speed obstacle method takes into account the motion state of the obstacle and can predict the future collision risk, while the dynamic window method takes into account the dynamic constraints of the robot and ensures that the planned obstacle avoidance speed is actually achievable by the robot. The fusion of the two methods makes the collision avoidance decision both forward-looking and feasible, significantly improving the success rate and safety of collision avoidance, especially when facing moving obstacles.

[0076] The expression for the expansion radius of the obstacle for safety expansion treatment is as follows:

[0077] in, Where is the radius of expansion. Let the radius be the obstacle. For the robot radius, For safety margin; The expression for the critical angle is:

[0078] in, It is the critical angle. The distance between the robot and the obstacle; The expression for determining the dangerous speed is:

[0079] in, Location of the obstacle. For the robot's position, Let be the relative velocity vector between the robot and the obstacle. This represents the maximum permissible time for a collision to occur.

[0080] Specifically, in the implementation of the collision avoidance module, this invention performs safety expansion processing on obstacles to ensure the first step of safe collision avoidance. The expansion radius is obtained by adding the obstacle radius, the robot radius, and the safety margin. The obtained expansion radius determines the minimum safe distance that the robot must maintain from the obstacle.

[0081] The critical angle is calculated by using the arcsine function to determine the ratio of the expansion radius to the distance between the robot and the obstacle. This angle defines the angle of the velocity obstacle cone. When the relative velocity vector falls within this cone, it means there is a risk of collision.

[0082] The dangerous speed determination in this invention is achieved by calculating the magnitude of the difference between the obstacle's position vector and the robot's position vector (i.e., the distance between them), then calculating the magnitude of the relative velocity vector (i.e., the relative velocity). Dividing these two values ​​yields the time required to reach the collision location while maintaining the current relative velocity. If this time is less than the set maximum permissible collision time, the speed is determined to be a dangerous speed. The establishment of these mathematical models provides the collision avoidance algorithm with a rigorous theoretical foundation and reliable computational basis. Model parameters such as safety margins and the maximum permissible collision time can be adjusted according to actual sea conditions and mission requirements, giving the system good adaptability.

[0083] The kinematic control outer loop establishes a position error state equation and uses a nonlinear controller to calculate heading and speed commands based on the position error. The dynamic control inner loop includes a heading and speed coordination control unit and a thruster optimization control unit. The heading and speed coordination control unit calculates the lateral steering force and left and right longitudinal thrust based on the heading and speed commands. The thruster optimization control unit uses a multi-objective optimization method to solve for the left and right nozzle velocities.

[0084] Furthermore, the propeller drive control module is electrically connected to the main control CPU module, and adopts a dual closed-loop control structure to achieve precise control of the robot's heading and speed. The kinematic control outer loop serves as the outer control loop, establishing a position error state equation. This equation describes the dynamic relationship between the error between the robot's actual position and the desired position. A nonlinear controller processes the position error. Considering the nonlinear characteristics of the robot's motion on the water surface, the controller is designed using nonlinear control theory. Based on the position error, the desired heading and speed commands are calculated, and these commands are transmitted to the dynamic control inner loop.

[0085] The dynamic control inner loop, as the inner control loop, includes two sub-units: the heading and speed coordination control unit and the thruster optimization control unit. The heading and speed coordination control unit receives heading and speed commands from the outer loop. Considering that the robot is equipped with two propellers and there is a coupling relationship between heading and speed, this unit calculates the required lateral steering force and left and right longitudinal thrust through a coordinated control algorithm. The lateral steering force is mainly used to change the robot's heading, the difference between the left and right longitudinal thrusts generates steering torque, and the sum of the left and right longitudinal thrusts determines the robot's forward thrust.

[0086] The propeller optimization control unit of the present invention receives commands for lateral steering force and left and right longitudinal thrust, and uses a multi-objective optimization method to solve for the flow velocity of the left and right nozzles. The optimization objectives include minimizing thrust error, minimizing steering force error, and minimizing the change in control quantity. The optimal flow velocity values ​​of the left and right nozzles are obtained by solving the multi-objective optimization algorithm, and these flow velocity commands are transmitted to the propeller driver through control signals.

[0087] The beneficial effects of the dual closed-loop control structure of this invention are that the outer loop control ensures position tracking accuracy, the inner loop control ensures dynamic response performance, and the cascaded cooperation of the two loops enables the system to have both good steady-state accuracy and fast dynamic response. In particular, the heading and speed coordination control solves the control coupling problem unique to dual propeller robots, and the thruster optimization control makes the working state of the left and right propellers reach the optimal level, improving energy utilization efficiency.

[0088] The main control CPU module 100 has a heterogeneous eight-core processor architecture, and the neural network processing unit provides 6 TOPS of AI computing power, supporting real-time inference of various neural network models including CNN convolutional neural network, RNN recurrent neural network and LSTM long short-term memory network.

[0089] Furthermore, the main control CPU module of this invention serves as the core processing unit of the entire system, responsible for data acquisition, processing, decision-making, and issuing control commands. The module features a heterogeneous eight-core processor architecture. Heterogeneous architecture refers to the presence of different types of cores within the processor, optimized for different types of computational tasks. By rationally allocating computational tasks to different types of cores, power consumption can be reduced while maintaining computational performance. The integrated neural network processing unit is specifically designed to accelerate the computation of artificial intelligence algorithms, providing considerable AI computing power and supporting real-time inference for various neural network models, including CNN convolutional neural networks, RNN recurrent neural networks, and LSTM long short-term memory networks.

[0090] The high-performance main control CPU module of this product provides powerful general-purpose computing capabilities, enabling it to process data and control tasks from multiple modules simultaneously. The neural network processing unit endows the system with powerful artificial intelligence capabilities, allowing the robot to intelligently identify its environment and make autonomous decisions, significantly improving the system's intelligence level and autonomous operation capabilities.

[0091] In specific implementation, the main control CPU module adopts the Rockchip RK3588 industrial-grade chip, which has a heterogeneous octa-core processor architecture with four ARM Cortex-A76 cores and four ARM Cortex-A55 cores. The Cortex-A76 core has a main frequency of 2.4GHz, and the Cortex-A55 core has a main frequency of 1.8GHz. The neural network processing unit adopts a high-performance AI computing architecture, providing 6TOPS of AI computing power and supporting real-time inference of various neural network models such as CNN convolutional neural network, RNN recurrent neural network and LSTM long short-term memory network.

[0092] This also includes: The task allocation module is electrically connected to the main control CPU module 100 and is used to allocate tasks when multiple robots work together. The task allocation module adopts an integer programming method, comprehensively considers constraints such as the distance between the robot and the person who fell into the water, the robot's battery power, communication distance and the number of lifebuoys carried, establishes a multi-objective optimization function, and solves the optimal task allocation scheme.

[0093] Furthermore, this invention also provides a task allocation module, which is electrically connected to the main control CPU module and is used for task allocation when multiple robots are working collaboratively. When multiple people need to be rescued and multiple rescue robots are available in the system, the task allocation module uses an integer programming method to optimize the allocation of rescue tasks. The module comprehensively considers multiple constraints and optimization objectives, including the distance between the robot and the person in the water (the closer the robot, the shorter the rescue time), the robot's remaining battery power (a robot with insufficient battery power may not be able to complete the round trip), the communication distance between the robot and the control console (a robot outside the communication range cannot receive instructions or transmit data), and the number of lifebuoys carried by the robot (a robot with insufficient lifebuoys cannot perform the rescue task).

[0094] In multi-task processing, this invention establishes a multi-objective optimization function, which weights and combines multiple optimization objectives such as minimizing rescue time, minimizing energy consumption, and maximizing communication reliability to form a comprehensive optimization objective function. An integer programming algorithm is used to solve this optimization problem. Integer programming refers to optimization problems where decision variables must take integer values. In task allocation, each robot either performs a task or does not perform a task, making it a typical integer programming problem. By solving this problem, the optimal task allocation scheme is obtained, i.e., determining which robot should rescue which person who has fallen into the water.

[0095] The intelligent task allocation of this invention significantly improves the collaborative operation efficiency of multi-robot systems, avoids the waste of resources caused by multiple robots going to the same person who has fallen into the water at the same time, and also avoids the situation where some people who have fallen into the water are not rescued by robots for a long time. By optimizing the allocation, the overall rescue time is minimized and the success rate is maximized, giving full play to the advantages of multi-robot systems. It has particularly important practical significance in large-scale maritime accident rescue.

[0096] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0097] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A main control system for a water rescue robot, characterized in that, include: The main control CPU module integrates a multi-core processor and a neural network processing unit, which is used to coordinate multiple modules and realize intelligent recognition functions; The communication module is electrically connected to the main control CPU module. The communication module includes a LoRa command and control link unit and a Mesh self-organizing network data transmission link unit to form a dual-means communication system for realizing data sharing when multiple robots work together. The navigation and positioning module is electrically connected to the main control CPU module. The navigation and positioning module adopts a two-level positioning combination method, which achieves precise positioning through the fusion of visual recognition and sound source positioning. It includes: a first-level positioning unit, which integrates a Beidou GNSS receiver and a MEMS inertial measurement unit, and achieves coarse positioning through a multi-source signal fusion algorithm; and a second-level positioning unit, which includes a binocular camera and a microphone array. The path planning module is connected to the main control CPU module and adopts a global path planning algorithm based on a grid map. The path tracking module interacts with the path planning module and the navigation and positioning module. The path tracking module adopts an improved integral line-of-sight guidance strategy, including a forward-looking distance adaptive adjustment unit and a dimension-reduced state observer. The forward-looking distance adaptive adjustment unit dynamically adjusts the forward-looking distance according to the lateral tracking error and the navigation speed. The dimension-reduced state observer estimates the time-varying sideslip angle online and compensates for it, so that the robot can travel according to the planned path. The collision avoidance module is electrically connected to the main control CPU module. It integrates the dynamic window method and the speed obstacle method. The collision avoidance module includes a speed obstacle judgment unit, a feasible speed window calculation unit, and an obstacle avoidance decision unit. The speed obstacle judgment unit judges the collision risk by calculating whether the relative speed vector between the robot and the obstacle falls within the speed obstacle cone. The feasible speed window calculation unit removes speed vectors located within the finite speed obstacle cone based on the dynamic window. The obstacle avoidance decision unit generates a safe local obstacle avoidance path. The propeller drive control module is electrically connected to the main control CPU module and adopts a dual closed-loop control structure. The dual closed-loop control structure includes a kinematic control outer loop and a dynamic control inner loop. The kinematic control outer loop calculates the heading and speed commands based on the position error. The dynamic control inner loop includes a heading and speed coordination control unit and a thruster optimization control unit. The heading and speed coordination control unit calculates the lateral turning force and the left and right longitudinal thrust based on the heading and speed commands. The thruster optimization control unit uses a multi-objective optimization method to solve for the optimal left and right nozzle velocities. The lifebuoy throwing control module is electrically connected to the main control CPU module. It is used to autonomously determine the throwing timing based on the relative position and posture of the robot and the person in the water, and push the lifebuoy shell out.

2. The main control system for a water rescue robot according to claim 1, characterized in that, The LoRa command and control link unit is used to transmit commands to the console. The Mesh self-organizing network data transmission link unit adopts a distributed, decentralized, multi-hop relay communication method, which supports point-to-point, point-to-many, and many-to-many dynamic networking, and is used to realize data sharing when multiple robots work together.

3. The main control system for a water rescue robot according to claim 1, characterized in that, The first-level positioning unit is an inertial navigation module that includes a GNSS subunit and a MEMS subunit. The GNSS subunit supports multi-frequency signal reception, and the MEMS subunit includes a three-axis accelerometer and a three-axis gyroscope. The second-level positioning unit includes a binocular camera, which includes a visible light camera and a near-infrared camera. The optical axes of the two cameras in the binocular camera are parallel and arranged horizontally. The microphone array includes multiple high-sensitivity microphones. The spatial filtering characteristics of the microphone array are used to estimate the location of the sound source by employing the time difference of arrival method or the controllable waveform positioning method.

4. The main control system for a water rescue robot according to claim 1, characterized in that, When performing global path planning, the path planning module converts the marine environment into a grid map containing obstacle information, expands the obstacle edges, expands the search in multiple directions from the starting grid, calculates the distance from each grid to the starting point and records the parent node information, and when the ending grid is found, it traces back to the starting point from the ending point based on the parent node information to generate the globally optimal path.

5. The main control system for a water rescue robot according to claim 1, characterized in that, In the path tracking module, the forward look distance adaptive adjustment unit dynamically adjusts the forward look distance based on the lateral tracking error and the navigation speed; the dimensionality reduction state observer estimates and compensates for the time-varying sideslip angle online by defining intermediate variables including the sideslip angle and the lateral tracking error, and the expected heading angle is calculated from the path tangent angle, the lateral tracking error, the forward look distance, and the estimated sideslip angle.

6. The main control system for a water rescue robot according to claim 1, characterized in that, In the collision avoidance module, the speed obstacle judgment unit is used to perform safety expansion processing on obstacles and calculate the relative speed vector between the robot and the obstacle. It determines whether the angle between the relative speed vector and the obstacle collision vector is less than the critical angle. If it is less than the critical angle, there is a risk of collision. The feasible speed window calculation unit introduces the maximum allowable collision occurrence time, constructs a finite speed obstacle cone, and determines whether the speed vector is a dangerous speed. The obstacle avoidance decision unit combines the evaluation function of the dynamic window method, including the target distance function, obstacle distance function, and speed evaluation function, and selects the optimal speed vector after comprehensive scoring.

7. The main control system for a water rescue robot according to claim 6, characterized in that, The expression for the expansion radius of an obstacle under safety expansion treatment is: in, Where is the radius of expansion. Let the radius be the obstacle. For the robot radius, For safety margin; The expression for the critical angle is: in, It is the critical angle. The distance between the robot and the obstacle; The expression for determining the dangerous speed is: in, Location of the obstacle. For the robot's position, Let be the relative velocity vector between the robot and the obstacle. This represents the maximum permissible time for a collision to occur.

8. The main control system for a water rescue robot according to claim 1, characterized in that, The kinematic control outer loop establishes a position error state equation and uses a nonlinear controller to calculate heading and speed commands based on the position error. The dynamic control inner loop includes a heading and speed coordination control unit and a thruster optimization control unit. The heading and speed coordination control unit calculates the lateral steering force and left and right longitudinal thrust based on the heading and speed commands. The thruster optimization control unit uses a multi-objective optimization method to solve for the left and right nozzle velocities.

9. The main control system for a water rescue robot according to claim 1, characterized in that, The main control CPU module has a heterogeneous eight-core processor architecture, and the neural network processing unit provides 6 TOPS of AI computing power, supporting real-time inference of various neural network models, including CNN convolutional neural networks, RNN recurrent neural networks and LSTM long short-term memory networks.

10. The main control system for a water rescue robot according to claim 1, characterized in that, Also includes: The task allocation module, electrically connected to the main control CPU module, is used to allocate tasks when multiple robots work together. The task allocation module adopts an integer programming method, comprehensively considers constraints such as the distance between the robot and the person who fell into the water, the robot's battery power, communication distance, and the number of lifebuoys carried, establishes a multi-objective optimization function, and solves the optimal task allocation scheme.

Images