Design method, system and equipment for driving path of unmanned ship and medium

Abstract

The invention discloses an unmanned ship driving path design method, system, device and medium, and relates to the technical field of unmanned ship control, and the method comprises the steps: optimizing a neural network model through a reward algorithm, taking the navigation path of a virtual unmanned ship as a label of environmental data to train the optimized neural network model, and constructing a planning path model; inputting the position and attitude data of a to-be-detected virtual unmanned ship into the planning path model for many times to obtain a plurality of planning paths capable of avoiding obstacles in the virtual environment; and obtaining expected values according to the plurality of planning paths, and selecting a planning result corresponding to the highest expected value in the expected values as the driving path of the unmanned ship. According to the method, the encountered obstacles can be avoided on the planned path, and the optimal path meeting the requirements can be accurately selected.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned ship control, and in particular to a design method, system, equipment and medium for an unmanned ship driving path. Background Art

[0002] Unmanned and intelligent shipbuilding are important directions for the development of today's ships. Related products and technologies such as remote driving of unmanned ships are constantly emerging in the market.

[0003] In the prior art, the methods for controlling the movement of unmanned boats mainly include: artificial potential field method and A* algorithm and other path planning methods. The artificial potential field method is based on the idea of ​​building a repulsive field around obstacles and a gravitational field around the target point. The unmanned boat is subjected to repulsive and gravitational forces in the composite field composed of these two potential fields. The combined force of repulsive and gravitational forces guides the movement of the controlled object and searches for collision-free obstacle avoidance paths. The A* algorithm evaluates the path cost through a heuristic function to find the shortest path from the starting point to the target point.

[0004] However, the above method cannot replan an accurate path that meets the requirements after avoiding obstacles. Summary of the invention

[0005] The embodiments of the present invention provide a method, system, device and medium for designing a driving path of an unmanned ship, which can solve the problem in the prior art that an accurate path that meets the requirements cannot be replanned after avoiding obstacles.

[0006] An embodiment of the present invention provides a method for designing a travel path of an unmanned ship, comprising the following steps:

[0007] The method comprises the following steps: 1. simulating the navigation path of a virtual unmanned ship in a virtual environment in real time, wherein the virtual unmanned ship is simulated by the position and posture data of the unmanned ship, and the virtual environment is simulated by the environmental data including the real-time obstacle data; optimizing the neural network model by means of a reward algorithm, and training the optimized neural network model by taking the navigation path of the virtual unmanned ship as a label of the environmental data to construct a planning path model, wherein the planning path model is used to plan a path that can avoid obstacles in the virtual environment; inputting the position and posture data of the virtual unmanned ship to be detected into the planning path model multiple times, and obtaining multiple planning paths that can avoid obstacles in the virtual environment; obtaining an expected value according to the multiple planning paths, and selecting a planning result corresponding to the highest expected value among the expected values ​​as the driving path of the unmanned ship.

[0008] Furthermore, the planning path model is constructed, and the specific steps include:

[0009] The training rules are set, and the training rules include: in the set number of training times, the navigation target point of the virtual unmanned ship is fixed unchanged; after the set number of training times is completed, the navigation target point of the virtual unmanned ship is randomly generated; the reward rules are set, and the reward rules include: according to the distance between the virtual unmanned ship and the obstacle, the distance between the virtual unmanned ship and the navigation target point, and the angle between the heading of the virtual unmanned ship and the target point, the behavior of the virtual unmanned ship during the training process is scored; the navigation path of the virtual unmanned ship is used as a label of environmental data, and the neural network model is trained to plan a path that can avoid obstacles in the virtual environment; during the training process, the neural network is optimized according to the score and the mean square error, and the parameters of the neural network are updated; the training is completed to obtain a planning path model.

[0010] Furthermore, the obtaining of the expected value comprises the following specific steps:

[0011] The expected value indicators are set as the length and consumption time of the path, and weights are assigned; an expected evaluation is performed on each planned path; the evaluation results are arranged in order, and the result with the shortest path length and the shortest consumption time is selected from the arranged results as the driving path of the unmanned boat.

[0012] An embodiment of the present invention provides a system for designing a travel path of an unmanned ship, comprising:

[0013] A data simulation module is used to simulate in real time the navigation path of a virtual unmanned ship in a virtual environment, wherein the virtual unmanned ship is simulated by the position and posture data of the unmanned ship, and the virtual environment is simulated by environmental data including real-time obstacle data; a model training module is used to optimize the neural network model through a reward algorithm, and train the optimized neural network model using the navigation path of the virtual unmanned ship as a label of the environmental data to construct a planning path model, and the planning path model is used to plan a path that can avoid obstacles in the virtual environment; an obstacle avoidance module is used to input the position and posture data of the virtual unmanned ship to be detected into the planning path model multiple times to obtain multiple planning paths that can avoid obstacles in the virtual environment; based on the multiple planning paths, an expected value is obtained, and a planning result corresponding to the highest expected value among the expected values ​​is selected as the driving path of the unmanned ship.

[0014] An embodiment of the present invention provides a computer device, comprising: a memory and a processor; the memory stores a computer program, and the processor implements the above-mentioned method for designing a driving path of an unmanned ship when executing the computer program.

[0015] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for designing a driving path of an unmanned ship described above is implemented.

[0016] The embodiments of the present invention provide a method, system, device and medium for designing a travel path of an unmanned ship. Compared with the prior art, the advantages thereof are as follows:

[0017] The neural network model is optimized through a reward algorithm, and the navigation path of the virtual unmanned ship is used as a label of environmental data to train the optimized neural network model to construct a planning path model; the position and posture data of the virtual unmanned ship to be detected are input into the planning path model multiple times to obtain multiple planned paths that can avoid obstacles in the virtual environment; and according to the expected value, a path that meets the requirements is selected from the multiple planned paths as the driving path, and the optimal path that meets the expected value can be obtained after avoiding obstacles. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of a virtual training environment provided by an embodiment of the present invention;

[0019] Figure 2 A schematic diagram of a reward function provided by an embodiment of the present invention;

[0020] Figure 3 The overall workflow diagram provided for the embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below.

[0022] An embodiment of the present invention provides a method for designing a travel path of an unmanned ship, comprising the following steps:

[0023] Step 1: Real-time simulation of the navigation path of the virtual unmanned ship in the virtual environment. The virtual unmanned ship is simulated by the position and posture data of the unmanned ship, and the virtual environment is simulated by the environmental data including real-time obstacle data. Figure 1 As shown, the red and blue lines are the hull coordinate system, the green wireframe is the obstacle, and the red dot is the target point.

[0024] Step 2: Optimize the neural network model through the reward algorithm, and use the navigation path of the virtual unmanned ship as the label of the environmental data to train the optimized neural network model to build a planning path model. The planning path model is used to plan a path that can avoid obstacles in the virtual environment.

[0025] Step 3: Input the position and posture data of the virtual unmanned ship to be detected into the planning path model multiple times to obtain multiple planning paths that can avoid obstacles in the virtual environment; obtain the expected value based on the multiple planning paths, and select the planning result corresponding to the highest expected value among the expected values ​​as the driving path of the unmanned ship.

[0026] The specific design of the present invention is:

[0027] The data acquisition part and automatic navigation part of the unmanned ship, as well as the digital mapping part of the unmanned ship in the cloud.

[0028] Data acquisition part. It is used to obtain the six free attitude data such as the rolling of the unmanned ship, the position, the distance from obstacles and its direction through the radar, attitude sensor and GPS module set on the unmanned ship; the navigation program running on the ship-side computer collects the data output by the sensor, and uses the neural network model to realize navigation and obstacle avoidance in a changing environment; the sensor collected data is uploaded to the cloud server through the ship-side computer.

[0029] The automatic navigation part is used to obtain the distance between obstacles and the current position data of the unmanned ship based on the laser radar and GPS modules, and realize automatic navigation and obstacle avoidance through the automatic navigation and obstacle avoidance model trained in the virtual environment by the neural network training script. After the unmanned ship is started, the ship-side computer imports the information obtained by the radar and GPS modules into the neural network related program, which will output appropriate behavioral instructions to realize the automatic navigation of the unmanned ship in the specified path. When the unmanned ship works in formation, it can sail in formation according to the set values ​​under the premise of realizing the autonomous navigation of a single device. The automatic navigation part also includes a danger perception and response part: when the minimum value of the distance data collected from the obstacle is lower than the set threshold, it will stop automatic navigation and keep the position of the unmanned ship unchanged, and send relevant messages to the cloud.

[0030] The digital mapping part of the cloud server will realize the display of the actual working environment model and the unmanned ship model on the web page based on the javascript language and the three.js library. The relative position of the virtual unmanned ship in the virtual environment is determined by the received GPS position information, and the current unmanned ship posture is mapped on the virtual unmanned ship through the received unmanned ship posture information. The actual unmanned ship motion path is modified by modifying the coordinate points on the web page, and the user can use it conveniently on multiple platforms through the web page construction.

[0031] 1. For the control of unmanned boats, the methods used are specifically introduced as follows:

[0032] Determine the starting point and end point of the unmanned boat's voyage, plan its approximate running path through the map, obtain a certain number of path points and upload them to the cloud server. After the unmanned boat passes the self-test, try to connect to the cloud server. If the connection fails, the unmanned boat will sail according to the last running data. If the connection is successful, the unmanned boat will obtain the current set running path and start automatic navigation. The unmanned boat obtains the distance information between the unmanned boat and the surrounding obstacles, the unmanned boat's six-degree-of-freedom attitude information, the unmanned boat's heading and position information through the set radar equipment, attitude sensor and GPS module; the system is based on the neural network model constructed by the DQN algorithm, which consists of an input layer with 20 neurons, two hidden layers with these 256 neurons, and an output layer with these three neurons. By pre-training the neural network in an artificially constructed virtual environment, the program will automatically score each behavior of the virtual unmanned boat based on multiple indicators such as environmental changes during training, and optimize the relevant parameters of each node in the hidden layer of the neural network based on the score, so that the neural network can more stably achieve the goal of navigation and obstacle avoidance. Import the neural network model that meets the requirements into the relevant script, and obtain and process the 18-dimensional distance data, the current unmanned ship heading data, and the distance data from the target point obtained by the laser radar, a total of 20 state data. The neural network will perform expected calculations for three output behaviors based on these 20 data, and select the behavior with the highest expected value to execute. Based on the previous training in the virtual environment, this behavior will be able to guide the unmanned ship to achieve the target functions of navigation and obstacle avoidance, such as Figure 2 As shown. Based on radar, GPS and other equipment that can obtain data in real time, the program can timely obtain the current movement status of the unmanned ship and its surrounding environment data. If there is any change in the environment, the neural network can synchronously process the relevant data and make relevant responses, which enables the unmanned ship to stably realize navigation and obstacle avoidance functions in a changing environment.

[0033] Based on the self-defined topic name when publishing messages in the MQTT protocol, each unmanned ship is numbered, and the length of the path each unmanned ship has traveled since its startup is obtained by processing the GPS signal of each unmanned ship. The unmanned ship navigation system will control the power output of each unmanned ship based on the set standard spacing to ensure that the difference in the length of the navigation path between adjacent numbered unmanned ships can be maintained within a range with the set standard spacing as the midpoint. By modifying the set standard spacing, the unmanned ship formation can work in a queue or alternately operate in a certain pattern.

[0034] When a single unmanned ship is working, it travels at a relatively low speed and obtains more accurate point cloud data of the environment within the radar detection range through the installed global laser radar. When working in an unmanned ship formation, each unmanned ship can sail at a relatively fast speed, and each ship will record the point cloud data of the surrounding environment. After uploading the relevant files to the server, the data of each file will be merged to obtain a relatively complete point cloud data of the navigation environment, and the navigation environment of the unmanned ship will be modeled through a program written based on the PCL library.

[0035] Second, the design of the unmanned ship also includes the following contents:

[0036] By controlling the speed of a single unmanned ship or having multiple unmanned ships scan together, more accurate point cloud data of the unmanned ship's operating environment can be obtained, and environmental modeling can be achieved on this basis.

[0037] When the unmanned boat starts, it will perform a self-check. If a module fails to start normally, a corresponding report will be sent to the cloud. When the distance between the unmanned boat and the obstacle is less than the set threshold, it will stop navigating and control the unmanned boat to stay at the current position, and send a report to the cloud.

[0038] The unmanned navigation system has a built-in GPS receiver and attitude sensor.

[0039] The distance and direction data between the unmanned ship and the obstacle are obtained through the laser radar set by the unmanned ship, the current heading of the unmanned ship is obtained through the attitude sensor, and the current position of the unmanned ship is obtained through the GPS module. The ship-side computer receives a series of raw data and processes it into decimal data. These data are imported into the relevant program, which will process the data based on the neural network model and output appropriate behavior instructions to achieve automatic navigation and obstacle avoidance of the unmanned ship in the target path.

[0040] Based on javascript and three.js library, a virtual environment corresponding to the real world is created on the web page. The relative position of the virtual unmanned ship in the virtual environment is determined by the received GPS position information, and the mapping is realized on the virtual unmanned ship through the received unmanned ship posture information. The path of the real unmanned ship is modified by modifying the coordinate points on the web page, and the user can use it conveniently on multiple platforms through the web page construction.

[0041] By keeping the difference in the length of the unmanned ship's running path within a set range, the unmanned ships can achieve formation navigation according to certain rules. The overall workflow is shown in the figure below: Figure 3 shown.

[0042] 3. Summarize the specific advantages of the present invention:

[0043] 1. By using three-dimensional modeling to display the surrounding environment of the unmanned ship when it is moving, instead of the traditional method of displaying relevant information through image transmission, this method allows observers to monitor the current situation of the unmanned ship from a third-party perspective, and can obtain the posture and surrounding environment of the unmanned ship more quickly and intuitively.

[0044] 2. By using neural networks to achieve automatic navigation of unmanned ships, replacing traditional manual control or path planning methods, the unmanned ships can be more adaptable to automatic navigation and obstacle avoidance in changing environments, and can respond to sudden appearances such as cruise ships or other objects and avoid related obstacles.

[0045] 3. By using relevant environmental mapping methods, users can more conveniently obtain the three-dimensional modeling of the environment where the unmanned boat operates in the waters without using additional equipment to obtain relevant data.

[0046] An embodiment of the present invention provides a system for designing a travel path of an unmanned ship, comprising:

[0047] The data simulation module is used to simulate the navigation path of the virtual unmanned ship in the virtual environment in real time. The virtual unmanned ship is simulated by the position and posture data of the unmanned ship, and the virtual environment is simulated by the environmental data containing real-time obstacle data. The model training module is used to optimize the neural network model through the reward algorithm, and the navigation path of the virtual unmanned ship is used as the label of the environmental data to train the optimized neural network model to construct a planning path model. The planning path model is used to plan a path that can avoid obstacles in the virtual environment. The obstacle avoidance module is used to input the position and posture data of the virtual unmanned ship to be detected into the planning path model multiple times to obtain multiple planning paths that can avoid obstacles in the virtual environment; according to the multiple planning paths, the expected value is obtained, and the planning result corresponding to the highest expected value among the expected values ​​is selected as the driving path of the unmanned ship.

[0048] An embodiment of the present invention provides a computer device, including: a memory and a processor; the memory stores a computer program, and when the processor executes the computer program, the steps of a method for designing a driving path of an unmanned ship are implemented.

[0049] An embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of a method for designing a driving path of an unmanned ship are implemented.

[0050] A specific embodiment is as follows:

[0051] The present invention comprises: an unmanned ship end data acquisition part, an automatic navigation part and an unmanned ship digital mapping part in the cloud.

[0052] The data acquisition part is used to obtain obstacle location information, current attitude and heading information of the unmanned ship through the laser radar, attitude sensor, electronic compass and other equipment installed on the unmanned ship; the unmanned ship is connected to the Internet through the 4G module, and the data collected by other devices are uploaded to the cloud server.

[0053] The digital mapping part of the cloud server will realize the display of the actual working environment model and the unmanned ship model on the web page based on the javascript language and the three.js library. The relative position of the virtual unmanned ship in the virtual environment is determined by the received GPS position information, and the current unmanned ship posture is mapped on the virtual unmanned ship through the received unmanned ship posture information. The actual unmanned ship motion path is modified by modifying the coordinate points on the web page, and the user can use it conveniently on multiple platforms through the web page construction.

[0054] The automatic navigation part of the unmanned ship uses the radar equipment, attitude sensor and GPS module installed on the unmanned ship to obtain data such as the distance information between the unmanned ship and surrounding obstacles, the six-degree-of-freedom attitude information of the unmanned ship, the heading and position information of the unmanned ship; it accepts and correlates data through the existing neural network model and outputs appropriate behavior instructions. Since the neural network model can process the current environmental information in real time, it enables the unmanned ship to realize navigation and obstacle avoidance functions in a changing environment.

[0055] For the water environment mapping part, based on the global laser radar set on the unmanned ship, a single unmanned ship or a fleet of unmanned ships is used to obtain the environmental point cloud data of the water surface of the relevant water area. After uploading the relevant point cloud data to the cloud server, the water surface environment of the unmanned ship navigation water area is reconstructed based on the point cloud data.

[0056] This system uses the following specific components:

[0057] The ship-side computer uses Raspberry Pi 4B+8GB memory, which is an ARM-based microcomputer motherboard with basic PC functions. It can directly run programs based on languages ​​such as Python when the environment is fully configured.

[0058] The advantages of this equipment are:

[0059] 1. Small size: As a micro motherboard, it maintains a small size while being able to realize the basic functions of a PC, making it easier to arrange on the device.

[0060] 2. Connect the underlying hardware: Raspberry Pi has 40 GPIO pins, which can directly interact with sensors and other devices, making the expansion and replacement of related equipment easier.

[0061] 3. Compatible with multiple languages: Raspberry Pi has a corresponding operating system, and users can choose different languages ​​for development work according to their own abilities.

[0062] 4. Low cost: Compared with daily computers, Raspberry Pi is low-cost and relatively friendly to development work.

[0063] The positioning adopts the WTRTK960+RTK_BOT_470P30 module. This combination can achieve high-precision global positioning. The advantages are:

[0064] 1. High-precision positioning: This module combination can achieve centimeter-level positioning through RTK positioning, helping the navigation system to better realize its functions.

[0065] 2. Excellent anti-interference ability: Its built-in anti-interference unit ensures the accuracy of positioning in complex electromagnetic environments.

[0066] 3. Multi-system integration: The module has high sensitivity and can realize single system integration of satellite navigation systems such as BDS and GPS, as well as multi-system integration in any combination.

[0067] 4. Small size: The module is small in size and easy to arrange in the unmanned ship.

[0068] 5. Can be connected via serial port: The module can communicate directly through the serial port, which is highly compatible with Raspberry Pi.

[0069] The radar uses the Midao MID-360 laser radar, which is a semi-global laser radar with the following advantages:

[0070] 1. Good waterproof performance: The radar has IP65 waterproof level, which can adapt well to the working environment on water, and will not cause equipment damage or failure due to water ingress.

[0071] 2. Can work in strong light environment: The radar has a strong ability to resist ambient light and can still work normally under the light when the unmanned ship is working during the day.

[0072] 3. Larger measurement range: For small unmanned ships, its 40m measurement radius can provide the unmanned ship with sufficient detection data to achieve obstacle avoidance function.

[0073] 4. Larger field of view angle: The field of view angle of -7° to 60° can obtain sufficient environmental point cloud data and obtain appropriate point cloud data.

[0074] The 4G module uses the mcuzone Qualcomm 4G full network expansion board. The advantages of this device are:

[0075] 1. It is easy to use and is compatible with phone cards from multiple companies including China Mobile and China Telecom, making it convenient to use in practice.

[0076] 2. The connection is stable. The device starts quickly, has a stable connection during use, and is not easy to disconnect.

[0077] 3. Fast upstream and downstream network speeds provide more timely data to the cloud.

[0078] The attitude sensor uses the WT1-IMU module, which has the following advantages:

[0079] 1. Small size, its size of 15*15*2mm makes it easy to install in various parts.

[0080] 2. High measurement accuracy. The measurement accuracy of 0.5 degrees in the x and y directions is sufficient to meet the detection requirements of the posture of small unmanned ships.

[0081] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for designing a travel path of an unmanned ship, characterized in that: The following steps are involved: Simulating in real time the navigation path of a virtual unmanned ship in a virtual environment, wherein the virtual unmanned ship is simulated by the position and posture data of the unmanned ship, and the virtual environment is simulated by environmental data including real-time obstacle data; The neural network model is optimized by a reward algorithm, and the navigation path of the virtual unmanned ship is used as a label of environmental data to train the optimized neural network model to construct a planning path model, wherein the planning path model is used to plan a path that can avoid obstacles in the virtual environment; The position and posture data of the virtual unmanned ship to be detected are input into the planning path model multiple times to obtain multiple planning paths that can avoid obstacles in the virtual environment; based on the multiple planning paths, the expected value is obtained, and the planning result corresponding to the highest expected value among the expected values ​​is selected as the driving path of the unmanned ship.

2. The method for designing a travel path of an unmanned ship according to claim 1, characterized in that: The specific steps of constructing the planning path model include: Setting training rules, wherein the training rules include: during the set number of training times, the navigation target point of the virtual unmanned ship is fixed and remains unchanged; after completing the set number of training times, the navigation target point of the virtual unmanned ship is randomly generated; Setting reward rules, wherein the reward rules include: scoring the behavior of the virtual unmanned ship during training according to the distance between the virtual unmanned ship and the obstacle, the distance between the virtual unmanned ship and the navigation target point, and the angle between the heading of the virtual unmanned ship and the target point; The navigation path of the virtual unmanned ship is used as the label of the environmental data, and the neural network model is trained to plan a path that can avoid obstacles in the virtual environment; During the training process, the neural network is optimized based on the score and mean square error, and the parameters of the neural network are updated; After training is completed, the planning path model is obtained.

3. The method for designing a travel path of an unmanned ship according to claim 1, characterized in that: The specific steps of obtaining the expected value include: Set the expected value indicators as the length and time of the path, and assign weights; Perform expectation evaluation on each planned path; The evaluation results are arranged in order, and the result with the shortest path length and the shortest time consumption is selected as the driving path of the unmanned boat.

4. A system for designing a travel path for an unmanned ship, characterized in that: include: A data simulation module, used for simulating in real time the navigation path of a virtual unmanned ship in a virtual environment, wherein the virtual unmanned ship is simulated by the position and posture data of the unmanned ship, and the virtual environment is simulated by environmental data including real-time obstacle data; A model training module is used to optimize the neural network model through a reward algorithm, and use the navigation path of the virtual unmanned ship as a label of environmental data to train the optimized neural network model to build a planning path model, wherein the planning path model is used to plan a path that can avoid obstacles in the virtual environment; The obstacle avoidance module is used to input the position and posture data of the virtual unmanned ship to be detected into the planning path model multiple times to obtain multiple planning paths that can avoid obstacles in the virtual environment; according to the multiple planning paths, the expected value is obtained, and the planning result corresponding to the highest expected value among the expected values ​​is selected as the driving path of the unmanned ship.

5. A computer device comprising: Memory and processor; The memory stores a computer program, and is characterized in that when the processor executes the computer program, a method for designing a driving path of an unmanned ship according to any one of claims 1 to 3 is implemented.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, a method for designing a travel path of an unmanned ship according to any one of claims 1 to 3 is implemented.

Images