Intelligent navigation system and method for tea picking robot

Abstract

The invention discloses an intelligent navigation system and method for a tea leaf picking robot, and the system is characterized in that an industrial control terminal carries out the construction of a three-dimensional vector space for a tea garden based on a panoramic image and a structured light image obtained in real time, and carries out the planning of a picking path through a path planning model stored in local; and converting the generated planning result into a control signal and transmitting the control signal to an execution mechanism of the picking robot. According to the system and the method, the real-time performance of path planning and the fluency of picking work are improved; the cloud server can provide cloud computing service interfaces for a plurality of industrial control terminals at the same time, management is convenient, and the overall use cost is reduced. According to the system, other sensors except a camera do not need to be used, the hardware equipment and software processing cost is reduced, the path planning model is stored locally, data are processed through the local model, the data processing speed and the picking work fluency are improved, and large-scale application and popularization of mechanical picking are facilitated.

Description

Technical Field

[0001] The present application relates to the technical field of tea picking robot control, and in particular to an intelligent navigation system and method for a tea picking robot. Background Art

[0002] As various tea varieties are widely planted across the country, my country's tea production has gradually increased. During the tea picking season, a large number of people are needed in the tea garden to pick and process tea, and the pickers need to spend a lot of time and energy. With the continuous development of new technologies, picking robots can realize mechanized tea picking and work continuously around the clock in designated areas. During the picking process, the picking path needs to be effectively planned, but at present, due to the slow path planning speed in the tea garden environment, the work efficiency of the picking robot is generally low, making it difficult to achieve large-scale mechanized tea picking.

[0003] In order to ensure the accuracy of path planning, existing picking path planning methods usually configure a variety of environmental information sensors, integrate the environmental information obtained by each sensor, and then transmit the data to the cloud server for calculation, and finally pick tea leaves according to the planned path. However, this method will increase the hardware cost, which is not conducive to the large-scale promotion of picking robots, and may have high computational complexity when dealing with complex tea garden scenes. For example, the data format of each sensor needs to be converted, corrected and fused, resulting in a long planning time. In addition, this method relies on cloud servers for data processing, which has high requirements for the stability of network connections. Once a network failure or instability occurs, it may affect the accuracy and reliability of path planning.

[0004] Therefore, it is necessary to propose an intelligent navigation system and method for a tea picking robot, which can reduce the use cost of the picking robot and perform rapid path planning for the picking robot in complex tea garden scenes, thereby improving the work efficiency of the picking robot and realizing the promotion and application of mechanical tea picking. Summary of the invention

[0005] The present application provides an intelligent navigation system and method for a tea picking robot, which is used to solve the technical problems in the prior art that the picking robot has slow path planning speed, is overly dependent on network environment conditions, has low picking efficiency, and cannot be promoted on a large scale.

[0006] In a first aspect, the present application provides an intelligent navigation system for a tea picking robot, including a cloud server and a plurality of industrial control terminals that are communicatively connected to the cloud server; The industrial control terminal is used to obtain panoramic images and structured light images in real time, establish a tea garden model based on the structured light images, determine the picking status information based on the panoramic images and the tea garden model, and upload the panoramic images, structured light images and picking status information to the cloud server; it is also used to download the path planning model of the cloud server to the local according to the preset update rules, and issue control instructions to the picking robot based on the local path planning model and picking status information; The cloud server is used to establish a path planning model, receive panoramic images, structured light images and picking status information sent by each industrial control terminal, optimize the path planning model according to the panoramic images and picking status information, and send the optimized path planning model to each industrial control terminal based on preset publishing rules.

[0007] Furthermore, the industrial control terminal includes a visual perception module, a data analysis module, a communication module and a planning control module; The visual perception module is used to obtain panoramic images and structured light images; The data analysis module is used to establish a tea garden model according to the structured light image, determine multiple target operation areas according to the tea garden model, and determine the picking status information of the picking robot by combining the multiple target operation areas and the panoramic image; The communication module is used to upload panoramic images, structured light images and picking status information to the cloud server in real time, and is also used to download the path planning model from the cloud server according to preset update rules; The planning and control module is used to determine the travel route based on the path planning model downloaded to the local computer, according to the target operation area and picking status information, and issue control instructions to the picking robot.

[0008] Furthermore, the planning control module determines the travel route based on the path planning model downloaded to the local computer and sends a control instruction to the picking robot according to the target operation area and the picking status information, including: Send a first instruction to make the picking robot move from a starting point along a first direction, and when a tea road in a target operation area is identified, send a movement stop instruction, and obtain a travel distance according to the state information of the picking robot and the position information of the tea road; Sending a second instruction to the picking robot to move in a second direction according to the travel distance, and sending a tea picking instruction when it is confirmed that the picking robot has reached the target operation area; Sending a third instruction to enable the picking robot to pick tea leaves in the target operation area and move along the second direction to the boundary of the operation area; Detect whether there are any obstacles in the first direction; when an obstacle appears, send the fourth instruction to make the picking robot return to the axis of the first direction where the starting point is located, and move along the axis of the first direction. During the movement, detect the tea road of the next target working area in real time, and send the fifth instruction to make the picking robot reach the next target working area; if no obstacle appears, turn at the current position, and send the sixth instruction to make the picking robot move along the first direction to the tea road of the next target working area.

[0009] Further, the planning control module includes a wheel odometer; The travel distance is calculated based on the rotation frequency of the moving wheels of the picking robot using a wheel odometer; the control instructions include steering data and throttle and brake opening data.

[0010] Furthermore, the visual perception module includes multiple groups of binocular cameras arranged in front, to the left, to the right and behind the picking robot and a structured light camera arranged on the top of the picking robot. Panoramic images are obtained through the multiple groups of binocular cameras, and structured light images are obtained through the structured light camera.

[0011] Furthermore, the picking status information includes at least the position information of the picking robot, the tea road position information of the target operation area, the road condition information and the distance between tea leaves in the target operation area.

[0012] Furthermore, the backbone layer of the path planning model adopts a residual neural network and a plurality of FPN feature fusion structures, and the head layer supports multi-task processing, and the tasks include at least object detection, landmark recognition, and tea ceremony recognition.

[0013] Furthermore, the path planning model uses a Monte Carlo tree search algorithm to search for paths, and uses a preset decision value evaluation function to evaluate the paths, and selects the optimal travel path according to the evaluation results.

[0014] Furthermore, the processor of the data analysis module adopts a graphics processor.

[0015] On the other hand, the present application also provides an intelligent navigation method for a tea picking robot, which is applied to any of the above-mentioned intelligent navigation systems of the tea picking robot, comprising: Establish a path planning model through the cloud server and send the path planning model to the industrial control end based on the preset publishing rules; Panoramic images and structured light images are acquired in real time through the industrial control terminal, a tea garden model is established according to the structured light image, the picking status information is determined based on the panoramic image and the tea garden model, and the panoramic image, structured light image and picking status information are uploaded to the cloud server; the path planning model of the cloud server is downloaded to the local computer, and control instructions are issued to the picking robot based on the local path planning model and picking status information.

[0016] The beneficial effects of the present application are as follows: the present application provides an intelligent navigation system and method for a tea picking robot, which has the following advantages: 1. This system uses a pure visual solution to complete the path planning problem of the tea-picking robot. The camera observes the surrounding environment of the tea-picking robot in real time. It does not require many different types of other sensors, which reduces the cost of hardware and software. 2. This system adopts a connected architecture that combines local computing on the industrial control end with self-optimized cloud service models. The trained detection model does not require too much computing power for local computing. The trained model can be put into operation after being remotely sent by the server. A single-stage target detection model is used for obstacle detection, while ensuring the real-time performance of path planning. 3. The industrial control end of this system constructs a three-dimensional vector space of the tea garden based on the collected visual data, and models it in real time during the journey to realize the digitization of the map information of the tea garden; 4. The cloud server of this system collects on-site samples in real time during the work of the industrial control end to follow up the training and expand the sample library, solving the problem of low recognition rate caused by insufficient samples brought by pre-training; 5. The cloud server of this system can provide cloud computing service interfaces to multiple industrial control terminals at the same time, which facilitates management and reduces the overall cost of use.

[0017] In summary, the system of the present application can achieve very low latency in the process of path planning, improves the real-time control and smoothness of work, and reduces software and hardware costs and management difficulty, which is helpful to achieve large-scale promotion of mechanized tea picking. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A system structure diagram of an embodiment of an intelligent navigation system for a tea picking robot provided in this application; Figure 2 A path display diagram of a path navigation embodiment provided by this application; Figure 3 This is an application structure framework diagram of an embodiment of the intelligent navigation system of the tea picking robot provided in this application. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0020] Before introducing the embodiments of the present application, the concept of the present application is first introduced. The navigation method of the existing picking robot has the following disadvantages: First, many sensors are needed, which is very costly. In the existing technology, due to the complex environment of tea gardens, in order to accurately locate the tea path and the picking robot, it is necessary to obtain environmental information through various types of sensors, but this means that the hardware cost and the software cost of analyzing and integrating the data need to be increased, which hinders the large-scale promotion of picking robots; Secondly, the current planning of picking routes often relies on cloud servers for data processing, which requires high stability of network connections. Once a network failure or instability occurs, the accuracy and reliability of route planning may be affected. In the tea garden scenario, the stability of the network cannot be guaranteed, and abnormal situations such as disconnection may occur, affecting the progress of tea picking and reducing production efficiency.

[0021] In response to the above problems, the present application provides an intelligent navigation system and method for a tea picking robot, in which the industrial control end realizes the perception of the environment through a pure visual solution, collects raw visual data, constructs a three-dimensional vector space for the tea garden through the collected visual data, and downloads the path planning model from the cloud server. The behavior and path planning of the picking robot are realized based on the three-dimensional vector space and the path planning model downloaded locally, and the generated planning results are converted into control signals and transmitted to the actuator of the picking robot. The above method improves the real-time performance of path planning and the smoothness of picking work. In addition, the cloud server can provide cloud computing service interfaces to multiple industrial control ends at the same time, which is convenient for management, reduces the overall cost of use, and contributes to the large-scale promotion of mechanized picking.

[0022] Combination Figure 1 As shown, a specific embodiment of the present application discloses an intelligent navigation system 100 for a tea picking robot, including a plurality of industrial control terminals 101 and a cloud server 102; The industrial control terminal 101 is used to obtain panoramic images and structured light images in real time, establish a tea garden model based on the structured light images, determine picking status information based on the panoramic images and the tea garden model, and upload the panoramic images, structured light images and picking status information to the cloud server 102; it is also used to download the path planning model of the cloud server 102 to the local according to the preset update rules, and issue control instructions to the picking robot based on the local path planning model and picking status information; The cloud server 102 is used to establish a path planning model, receive panoramic images, structured light images and picking status information sent by each industrial control terminal 101, optimize the path planning model according to the panoramic images and picking status information, and send the optimized path planning model to the corresponding industrial control terminal based on preset publishing rules.

[0023] Compared with the prior art, this system realizes the perception of the environment through a pure visual solution. The industrial control end constructs a three-dimensional vector space of the tea garden based on the panoramic image and structured light image obtained in real time through the collected visual data, and downloads the path planning model from the cloud server. The picking path is planned using the path planning model stored locally, and the generated planning result is converted into a control signal and transmitted to the actuator of the picking robot. The above method improves the real-time performance of path planning and the fluency of picking work. In addition, the cloud server can provide cloud computing service interfaces to multiple industrial control ends at the same time, which is convenient for management, reduces the overall cost of use, and contributes to the large-scale promotion of mechanized picking. The system of this embodiment only needs to rely on visual perception sensors, and does not require a variety of different types of other sensors, which reduces the cost of hardware and software. In addition, the path planning model of the industrial control end is stored locally, and the data is processed by the local model, which can reduce the delay, improve the real-time degree of data processing, and improve the fluency of picking work; the cloud server can provide cloud computing service interfaces to multiple clients at the same time, which reduces the difficulty and cost of management, and contributes to the large-scale promotion of mechanized tea picking.

[0024] In different seasons, the growth status and leaf characteristics of different tea plants will change, and the road conditions of the tea road will also be different. Updating the recognition model according to different seasons, automatically downloading the corresponding detection model from the server and deploying it can better adapt to the characteristic changes of tea plants in different seasons, help the picking robot optimize the operation strategy, adjust the picking path, better adapt to the actual situation of the tea garden in the current season, and improve the path planning and picking effect.

[0025] In a specific embodiment, the preset update rule includes: determining the download frequency according to the setting of the current season; The preset publishing rule is: when the industrial computer is idle, the system will automatically issue and update the detection model according to seasonal needs to achieve continuous optimization of model performance.

[0026] It should be noted that when the industrial control end collects visual data, it will transmit the sample data collected on site in real time to the cloud server. If the network signal is poor, the picture will be temporarily saved locally. After the network is restored, the distance between tea trees and tea leaves, the path between tea tree rows (tea road) and shooting time information will be uploaded to the cloud server. The cloud server expands the training sample library by receiving the actual collected data to solve the problem of low accuracy of the path planning model caused by insufficient samples brought by pre-training.

[0027] In a preferred embodiment, the industrial control terminal includes a visual perception module, a data analysis module, a communication module and a planning control module; The visual perception module is used to obtain panoramic images and structured light images; The data analysis module is used to establish a tea garden model according to the structured light image, determine multiple target operation areas according to the tea garden model, and determine the picking status information of the picking robot by combining the multiple target operation areas and the panoramic image; The communication module is used to upload panoramic images, structured light images and picking status information to the cloud server in real time, and is also used to download the path planning model from the cloud server according to preset update rules; The planning and control module is used to determine the travel route based on the path planning model downloaded to the local computer and the target working area and picking status information, and issue control instructions to the picking robot.

[0028] In a preferred embodiment, the visual perception module includes multiple groups of binocular cameras arranged in front, to the left, to the right and behind the picking robot and a structured light camera arranged on the top of the picking robot. Panoramic images are obtained through the multiple groups of binocular cameras, and structured light images are obtained through the structured light camera.

[0029] By setting up multiple groups of binocular cameras in front, to the left, to the right and behind the picking robot, a 360-degree all-round perception of the surrounding environment of the picking robot can be achieved, improving the ability to detect environmental changes and obstacles; the structured light camera can perceive the depth and distance of the environment by acquiring structured light images, which helps to more accurately identify the working area and provide an accurate data basis for panoramic modeling of the tea garden.

[0030] It should be noted that, in actual applications, more groups of cameras can be used according to the existing hardware conditions, or other technologies can be used to acquire environmental data.

[0031] In a specific embodiment, the industrial control end is divided into two parallel working threads when performing image data acquisition: The first thread is responsible for uploading the collected images to the cloud server through the communication module. The upload can be based on the TCP / IP protocol, and the transmission network can be a 5G network. The industrial control end transmits the distance between tea trees and tea leaves, the path between tea tree rows (tea road), and the shooting time information to the cloud server in real time. The cloud server automatically classifies and stores this data and includes it in the data set library. The data is then automatically labeled and retrained through the pre-trained model, and then manually reviewed to produce a model with better performance. In some scenarios, the industrial control end will also upload the detection speed and accuracy data to the cloud in real time for subsequent data analysis and statistics; The second thread is to perform pre-processing such as denoising on the image to optimize the image quality for model detection, thereby improving the detection and path planning accuracy of the model.

[0032] In a specific embodiment, modeling of tea gardens based on structured light images can also be achieved using a neural network model, which completes three-dimensional space transformation based on structured light image data, replacing the traditional method of fusing sensor information such as lidar. During the processing, the BEV (top view) space conversion layer is introduced into the neural network model. The BEV space conversion layer converts the data of the robot's surrounding environment from its own coordinate system to the top view coordinate system, eliminating the influence of geometric shape changes such as ground slope, which helps to more intuitively understand the environment and obstacle distribution around the robot, and realize accurate perception and prediction of object depth information through the neural network model.

[0033] In a preferred embodiment, the planning control module determines the travel route based on the path planning model downloaded to the local according to the target operation area and the picking status information, and issues a control instruction to the picking robot, including: Send a first instruction to make the picking robot move from a starting point along a first direction, and when a tea road in a target operation area is identified, send a movement stop instruction, and obtain a travel distance according to the state information of the picking robot and the position information of the tea road; Sending a second instruction to the picking robot to move in a second direction according to the travel distance, and sending a tea picking instruction when it is confirmed that the picking robot has reached the target operation area; Sending a third instruction to enable the picking robot to pick tea leaves in the target operation area and move along the second direction to the boundary of the operation area; Detect whether there are any obstacles in the first direction; when an obstacle appears, send the fourth instruction to make the picking robot return to the axis of the first direction where the starting point is located, and move along the axis of the first direction. During the movement, detect the tea road of the next target working area in real time, and send the fifth instruction to make the picking robot reach the next target working area; if no obstacle appears, turn at the current position, and send the sixth instruction to make the picking robot move along the first direction to the tea road of the next target working area.

[0034] Based on the path planning model downloaded to the local computer, combined with the target operation area and picking status information, instructions are sent to enable the robot to move along the planned path. This improves the real-time response, reduces system latency, avoids excessive reliance on network conditions during path planning, and improves the smoothness of the picking robot when picking tea, thereby improving work efficiency.

[0035] Combine the following Figure 2 The above control process is explained with a specific example. Figure 2 As shown, assuming that the current starting position of the picking robot is point A in the figure, the first direction is the x-axis in the figure, and the second direction is the y-axis in the figure, then the control method of the picking robot by the industrial control end is as follows: The picking robot uses the detected tea road as its own driving trajectory. The robot first cruises along the x-axis. When the binocular camera recognizes the tea road, the robot stops moving and uses the binocular vision target detection algorithm to determine the exact location of the tea road. Subsequently, the robot moves along the y-axis, while using its eight cameras to build a 360-degree panoramic view to confirm whether it has reached a specific work area. After confirming arrival, a picking instruction is issued. After picking is completed, the robot continues to move along the y-axis to the boundary. With the help of the 360-degree visual perception system, the robot can detect whether there are obstacles in the x-axis direction.

[0036] If encountering an obstacle, the robot will return to the x-axis where the starting point A is located, and move along the original x-axis path, and detect the tea road in real time to reach the next picking area; if there is no obstacle, the robot will turn to the x-axis direction at the current y-axis position, and repeat this cycle until it reaches the end point B.

[0037] In a preferred embodiment, the picking status information includes at least the position information of the picking robot, the position information of the tea road, the distance between tea leaves in the target picking area and the surface condition information of the tea road.

[0038] In a preferred embodiment, the planning control module includes a wheel odometer; The travel distance is calculated based on the rotation frequency of the moving wheels of the picking robot using a wheel odometer; the control instructions include steering data and throttle and brake opening data.

[0039] Through the tea road road condition information, tea road location information and robot location information, the throttle opening and steering angle can be calculated to accurately reach the tea picking area.

[0040] In a preferred embodiment, the backbone layer of the path planning model adopts a residual neural network and a plurality of FPN feature fusion structures, and the head layer supports multi-task processing, and the tasks include at least object detection, landmark recognition, and tea ceremony recognition.

[0041] In addition, in order to improve the accuracy of the model's data processing, a virtual camera is introduced to construct standardized data to solve the slight deviations caused by different camera external parameters. A virtual standard camera layer is used to process image data through camera calibration external parameters to ensure data consistency and provide calibrated input for the backbone neural network.

[0042] In some embodiments, the data acquired by the visual perception module may be video data. When processing the video data, a spatiotemporal sequence feature layer may be introduced into the path planning model. By processing the spatiotemporal sequence of the video instead of static images, the short-term memory capacity of the perception network is improved, and the problem of partial feature loss caused by instantaneous perception is solved, so that the model can better understand dynamic scenes, and when moving obstacles appear in the picking scene, it can make accurate judgments and optimize the route.

[0043] In a preferred embodiment, the path planning model uses a Monte Carlo Tree Search algorithm (MCTS algorithm) to search for paths, and uses a preset decision value evaluation function to evaluate the paths, and selects the optimal travel path according to the evaluation results.

[0044] The value function is obtained by learning the full scene features through neural networks, and the MCTS algorithm is used to guide the path search to improve the real-time decision-making. This learning model enables the tea-picking robot to make more intelligent and real-time decisions in complex environments.

[0045] In a preferred embodiment, the processor of the data analysis module adopts a graphics processor. The built-in GPU acceleration function is used to execute multiple threads simultaneously, which is suitable for processing large-scale data parallel computing tasks. By decomposing the task into multiple parallel computing units and executing these computing units simultaneously on the GPU, the processing speed of the task can be accelerated.

[0046] Combine the following Figure 3 The overall process is described in detail. Figure 3 The complete processing process of this system is shown. The left side shows the data processing and operation of the industrial control end, and the right side shows the data processing process of the cloud server. The industrial control end obtains structured light image data and 360 panoramic image data through a 3D structured light camera and four sets of binocular cameras respectively. Based on the acquired image data, a three-dimensional scene model of the tea garden is established, the target area for picking is determined, and the image data is uploaded to the cloud server. The cloud server optimizes the path planning model based on the acquired data. The industrial control end controls the throttle, brake, etc. of the picking robot based on the model saved locally, so that the picking robot can achieve efficient tea picking.

[0047] The embodiment of the present application further provides an intelligent navigation method for a tea picking robot, which is applied to any of the above-mentioned intelligent navigation systems for tea picking robots, and includes: Establish a path planning model through the cloud server and send the path planning model to the industrial control end based on the preset publishing rules; Panoramic images and structured light images are acquired in real time through the industrial control terminal, a tea garden model is established according to the structured light image, the picking status information is determined based on the panoramic image and the tea garden model, and the panoramic image, structured light image and picking status information are uploaded to the cloud server; the path planning model of the cloud server is downloaded to the local computer, and control instructions are issued to the picking robot based on the local path planning model and picking status information.

[0048] The intelligent navigation system and method of the tea picking robot provided in this embodiment realizes the perception of the environment through a pure visual solution. The industrial control end constructs a three-dimensional vector space of the tea garden based on the panoramic image and structured light image obtained in real time through the collected visual data, and downloads the path planning model from the cloud server. The picking path is planned using the path planning model stored locally, and the generated planning results are converted into control signals and transmitted to the actuator of the picking robot. The above method improves the real-time performance of path planning and the smoothness of picking work. In addition, the cloud server can provide cloud computing service interfaces to multiple industrial control ends at the same time, which is convenient for management, reduces the overall cost of use, and contributes to the large-scale promotion of mechanized picking.

[0049] This system only needs to rely on visual perception sensors, and does not require a variety of other sensors of different types, which reduces the cost of hardware and software. In addition, the path planning model of the industrial control end is stored locally. By processing data through the local model, it can reduce latency, improve the real-time degree of data processing, and improve the smoothness of picking work; the cloud server can provide cloud computing service interfaces to multiple clients at the same time, reducing the difficulty and cost of management, and helping to achieve large-scale promotion of mechanized tea picking.

[0050] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0051] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent navigation system for a tea picking robot, characterized in that: It includes a cloud server and multiple industrial control terminals communicating with the cloud server; The industrial control terminal is used to obtain panoramic images and structured light images in real time, establish a tea garden model based on the structured light image, determine the picking status information based on the panoramic image and the tea garden model, and upload the panoramic image, structured light image and picking status information to the cloud server; It is also used to download the path planning model of the cloud server to the local according to the preset update rules, and issue control instructions to the picking robot based on the local path planning model and picking status information; the industrial control end includes a visual perception module, a data analysis module, a communication module and a planning control module; The visual perception module is used to obtain panoramic images and structured light images; The data analysis module is used to establish a tea garden model according to the structured light image, determine multiple target operation areas according to the tea garden model, and determine the picking status information of the picking robot by combining the multiple target operation areas and the panoramic image; The communication module is used to upload panoramic images, structured light images and picking status information to the cloud server in real time, and is also used to download the path planning model from the cloud server according to preset update rules; The planning and control module is used to determine the travel route based on the path planning model downloaded to the local computer and the target operation area and picking status information, and issue control instructions to the picking robot; The planning control module determines the travel route based on the path planning model downloaded to the local computer and the target operation area and picking status information, and issues control instructions to the picking robot, including: Send a first instruction to make the picking robot move from a starting point along a first direction, and when a tea road in a target operation area is identified, send a movement stop instruction, and obtain a travel distance according to the state information of the picking robot and the position information of the tea road; Sending a second instruction to the picking robot to move in a second direction according to the travel distance, and sending a tea picking instruction when it is confirmed that the picking robot has reached the target operation area; Sending a third instruction to enable the picking robot to pick tea leaves in the target operation area and move along the second direction to the boundary of the operation area; Detect whether there is an obstacle in the first direction; when an obstacle appears, send a fourth instruction to make the picking robot return to the axis of the first direction where the starting point is located, and move along the axis of the first direction, and detect the tea path of the next target operation area in real time during the movement, and send a fifth instruction to make the picking robot reach the next target operation area; if no obstacle appears, turn at the current position, and send a sixth instruction to make the picking robot move along the first direction to the tea path of the next target operation area; The cloud server is used to establish a path planning model, receive panoramic images, structured light images and picking status information sent by each industrial control terminal, optimize the path planning model according to the panoramic images and picking status information, and send the optimized path planning model to each industrial control terminal based on preset publishing rules.

2. The intelligent navigation system of the tea picking robot according to claim 1, characterized in that: The planning control module includes a wheel odometer; The travel distance is calculated based on the rotation frequency of the moving wheels of the picking robot according to the wheel odometer; the control instruction includes steering data and throttle and brake opening data.

3. The intelligent navigation system of the tea picking robot according to claim 1, characterized in that: The visual perception module includes multiple groups of binocular cameras arranged in front of, to the left, to the right and behind the picking robot and a structured light camera arranged on the top of the picking robot. Panoramic images are obtained through the multiple groups of binocular cameras, and structured light images are obtained through the structured light camera.

4. The intelligent navigation system of the tea picking robot according to claim 1, characterized in that: The picking status information at least includes the position information of the picking robot, the tea road position information of the target operation area, the road surface condition information and the distance between tea leaves in the target operation area.

5. The intelligent navigation system of the tea picking robot according to claim 1, characterized in that: The backbone layer of the path planning model adopts a residual neural network and a plurality of FPN feature fusion structures, and the head layer supports multi-task processing, and the tasks include at least object detection, landmark recognition, and tea ceremony recognition.

6. The intelligent navigation system of the tea picking robot according to claim 5, characterized in that: The path planning model uses a Monte Carlo tree search algorithm to search for paths, and uses a preset decision value evaluation function to evaluate the paths, and selects the optimal travel path based on the evaluation results.

7. The intelligent navigation system for tea picking robots according to claim 1, characterized in that: The processor of the data analysis module is a graphics processor.

8. An intelligent navigation method for a tea picking robot, characterized in that: The intelligent navigation system for a tea picking robot according to any one of claims 1 to 7 is characterized by comprising: Establish a path planning model through the cloud server and send the path planning model to the industrial control end based on the preset publishing rules; Panoramic images and structured light images are acquired in real time through the industrial control terminal, a tea garden model is established according to the structured light image, the picking status information is determined based on the panoramic image and the tea garden model, and the panoramic image, structured light image and picking status information are uploaded to the cloud server; the path planning model of the cloud server is downloaded to the local computer, and control instructions are issued to the picking robot based on the local path planning model and picking status information.

Images