Robotic arm path planning method, device and readable storage medium for tissue culture seedling transplanting

Abstract

This invention proposes a path planning method, device, and readable storage medium for a robotic arm used for transplanting tissue culture seedlings. This method uses machine vision to identify the position of the tissue culture seedlings and the empty space in the plug tray. It defines a 28-dimensional state space and an 8-dimensional action space encompassing position, posture, joint angles, and gripping force. A multi-sub-item weighted reward function is designed that integrates successful transplanting, obstacle avoidance, posture matching, and time efficiency. A deep reinforcement learning network is constructed using the Soft Actor-Critic (SAC) algorithm. Through virtual simulation training, the robotic arm autonomously learns to generate a collision-free, smooth, and efficient grasping-transferring-transplanting path, adapting to dynamic environments such as the diverse morphology of tissue culture seedlings and varying plug tray sizes. This method eliminates the need for pre-set scenario parameters, significantly improving the system's generalization and automation capabilities, and providing a reliable solution for intelligent production of tissue culture seedling transplanting.

Description

Technical Field

[0001] The present invention relates to the field of mechanical engineering, and in particular to a method and device for robot arm path planning for tissue culture seedling transplanting, and a readable storage medium thereof, which integrates robot control, machine vision, intelligent algorithms (deep reinforcement learning) and agricultural biotechnology. Background Art

[0002] Transplanting tissue culture seedlings into plug trays is a core step in plant tissue culture production, and its mechanization and automation rely on robotic path planning technology. Traditional multi-DOF robotic path planning algorithms (such as A*, fast random trees, probabilistic roadmaps, and covariant Hamiltonian optimization) are based on mathematical modeling or heuristic rules, relying on known environmental information and deterministic calculations. These algorithms are only suitable for structured scenarios where the tissue culture seedlings have a fixed position and the layout of the plug tray remains unchanged.

[0003] However, in actual production, tissue culture seedlings exhibit diverse morphologies and random spatial positions, with dynamic changes in the number of available holes in the plug trays. Furthermore, complex factors such as seedling hardness and plug tray specifications must be considered. Traditional methods struggle to accurately model these dynamic, unstructured scenarios, leading to grasping failures and transplanting errors. Frequent manual adjustments to program parameters are required, resulting in insufficient system flexibility and generalizability, making them unable to meet the practical needs of automated tissue culture seedling transplanting.

[0004] Therefore, there is an urgent need for a robot arm path planning method, device and readable storage medium for tissue culture seedling transplanting to solve the problems existing in the prior art. Summary of the Invention

[0005] The embodiments of the present invention provide a robotic arm path planning method, device and readable storage medium for tissue culture seedling transplanting. The traditional robotic arm path planning method in the current technology relies on preset environmental models and heuristic rules, which is difficult to adapt to the complex environment of tissue culture seedling transplanting scenarios with random seedling postures, dynamic changes in plug trays and multi-factor coupling, resulting in low planning accuracy, poor flexibility, high maintenance costs and other problems.

[0006] The core technology of this invention mainly relies on customized design of the state space, action space and reward function of the tissue culture seedling transplanting scenario, combined with the soft actor-critic (SAC) algorithm, to enable the robotic arm to autonomously learn and adapt to the dynamic environment and generate a collision-free, smooth and efficient grasping-transferring-transplanting path.

[0007] In a first aspect, the present invention provides a method for robot arm path planning for tissue culture seedling transplanting, the method comprising the following steps:

[0008] S1. Use machine vision technology to identify the position of the target tissue culture seedlings and the empty space in the plug tray;

[0009] S2. Define the state space including the position and posture of the tissue culture seedling and the plug tray, the joint angles of the robot arm, the position, posture and clamping force of the end effector;

[0010] S3. Define the motion space including the motion of each joint of the robot arm, the gripping motion of the end effector, and the pushing motion;

[0011] S4. Design a reward function, which includes a reward for successful transplanting, a reward for avoiding obstacles, a reward for approaching the target seedling, a reward for approaching the target position in the plug tray, a reward for matching the posture of the target seedling, a reward for matching the posture of the plug tray transplanting, and a reward for time efficiency, and perform a weighted summation of each sub-reward function;

[0012] S5. A deep reinforcement learning network is constructed using the soft actor-critic algorithm framework. The deep reinforcement learning network is trained through the state space, action space, and reward function, so that the deep reinforcement learning network learns to generate the motion path of the robotic arm from the initial position to the point where it grabs the target tissue culture seedling and transplants it to the empty space in the hole tray.

[0013] Furthermore, in step S2, the positions of the tissue culture seedlings and the plug tray are three-dimensional coordinates, and the postures are represented by quaternions;

[0014] The joint angle of the robotic arm is obtained through the built-in angle sensor of the robotic arm;

[0015] The position of the end effector is calculated by the forward kinematics of the robot arm, the posture is expressed by quaternion, and the gripping force is obtained by the thin film force sensor on the end effector gripper.

[0016] Furthermore, in step S3, the angle range of each joint movement of the robotic arm is [-90°, 90°];

[0017] The end effector's clamping action includes two states: release and clamping, and the pushing action includes two states: retraction and extension.

[0018] Furthermore, in step S4, the successful transplanting reward is: when the tissue culture seedling is planted into the target plug tray and the clamping force is 0.3-0.5N, a first preset reward value is obtained; when the tissue culture seedling is successfully grasped but not transplanted, a second preset reward value is obtained; otherwise, the reward is zero;

[0019] The obstacle avoidance reward is: a negative reward is obtained when the distance between the end effector and the obstacle is less than the safety threshold, otherwise it is zero;

[0020] The rewards for approaching the target seedling and the target position of the plug tray are generated based on the distance between the end effector and the target. The smaller the distance, the higher the reward.

[0021] The rewards for matching the target seedling posture and the target tray transplant posture are both generated based on the quaternion space distance between the end effector and the target. The smaller the distance, the higher the reward.

[0022] Time efficiency rewards are generated based on the time it takes to complete a transplanting task. The shorter the time, the higher the reward.

[0023] Furthermore, the reward function is obtained by multiplying each sub-reward function by the corresponding weight and then summing them up. The weights are set according to expert experience. The weight of the successful transplanting reward is 300, the weight of the obstacle avoidance reward is 200, the weight of the reward for approaching the target seedling and the reward for approaching the target position of the plug tray are both 5, the weight of the target seedling posture matching reward and the plug tray transplanting posture matching reward are both 4, and the weight of the time efficiency reward is 1.

[0024] Furthermore, in step S5, the soft actor-critic algorithm framework includes an actor network and two critic networks. The actor network inputs the state space and outputs the action probability distribution, and the critic network inputs the state and action and outputs the state value.

[0025] The Actor network consists of three fully connected layers with 128, 256, and 128 hidden units, respectively. The first two layers use the ReLU activation function, and the output layer uses the tanh function.

[0026] In each Q network of the critic network, the state input is processed by four fully connected layers, with the number of hidden units being 256, 128, 256, and 128, respectively. The action input is processed by three fully connected layers, with the number of hidden units being 128, 256, and 128, respectively. The two are merged before the third hidden layer to output a single-dimensional state value.

[0027] Furthermore, in step S5, when grasping the target tissue culture seedling, the clamping force output by the thin film force sensor is combined with the fuzzy PID algorithm to control the clamping force within a preset range, thereby achieving flexible grasping;

[0028] When transplanting to the plug tray, adjust the posture of the end effector to make the tissue culture seedlings upright, and complete the transplanting by pushing.

[0029] In a second aspect, the present invention provides a method and device for robot arm path planning for tissue culture seedling transplanting, comprising:

[0030] The posture acquisition module uses machine vision technology to identify the posture of the target tissue culture seedlings and the empty space in the plug tray;

[0031] Definition module, which defines the state space including the position and posture of the tissue culture seedling and the plug tray, the joint angles of the robot arm, the position, posture and clamping force of the end effector; and defines the action space including the motion of each joint of the robot arm, the clamping action and the pushing action of the end effector;

[0032] The reward design module designs reward functions, including rewards for successful transplanting, obstacle avoidance, proximity to the target seedling, proximity to the target location in the plug tray, target seedling posture matching, plug tray transplant posture matching, and time efficiency. The reward functions are then weighted and summed.

[0033] The model construction module uses the soft actor-critic algorithm framework to build a deep reinforcement learning network. The deep reinforcement learning network is trained through state space, action space and reward function, so that the deep reinforcement learning network can learn to generate the motion path of the robotic arm from the initial position to grasping the target tissue culture seedling and transplanting it to the empty space in the hole tray.

[0034] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the above-mentioned robotic arm path planning method for tissue culture seedling transplanting.

[0035] In a fourth aspect, the present invention provides a readable storage medium, in which a computer program is stored. The computer program includes a program code for controlling a process to execute a process, and the process includes the above-mentioned robotic arm path planning method for tissue culture seedling transplanting.

[0036] The main contributions and innovations of the present invention are as follows:

[0037] 1. Strong environmental adaptability: No need to preset scene parameters such as tissue culture seedling posture and plug tray specifications. Through deep reinforcement learning, the system can autonomously extract dynamic environmental features and adapt to unstructured scenes with diverse seedling shapes and randomly changing vacancies.

[0038] 2. Outstanding generalization capabilities: Based on data-driven end-to-end policy learning, it can automatically handle tissue culture seedlings of different types, sizes, and hardness, as well as plug trays of different specifications, without the need for remodeling or program adjustments for new scenarios, significantly reducing system maintenance costs.

[0039] 3. Planning performance optimization: A multi-dimensional reward function (including successful transplanting, obstacle avoidance, posture matching, and time efficiency) guides strategy optimization to achieve collision-free paths, precise posture matching (using quaternions to avoid Euler angle singularities), and efficient transplanting (dynamically controlling the clamping force to 0.3-0.5N).

[0040] 4. High algorithm stability: The SAC algorithm framework is used in combination with the dual-Q network to reduce value estimation bias and improve data utilization efficiency through experience replay, ensuring stable generation of optimal paths in complex dynamic environments.

[0041] 5. Improved automation: Integrating machine vision and sensor feedback (2D cameras, joint angle sensors, and thin-film force sensors) enables full automation of the entire process, from posture recognition to path planning, flexible grasping, and precise transplanting, promoting intelligent upgrades in tissue culture seedling production.

[0042] The details of one or more embodiments of the invention are set forth in the accompanying drawings and the description below so that other features, objects, and advantages of the invention are more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0044] Figure 1 is a flow chart of a robotic arm path planning method for tissue culture seedling transplanting according to an embodiment of the present invention;

[0045] Figure 2 This is a scene diagram of transplanting tissue culture seedlings in a plug tray according to an embodiment of the present invention;

[0046] Figure 3 is a framework diagram of a DRL algorithm for generating a robot arm path according to an embodiment of the present invention;

[0047] Figure 4 FIG. 4 is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The implementations described in the following exemplary embodiments are not intended to represent all implementations consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with certain aspects of one or more embodiments of this specification, as detailed in the appended claims.

[0049] It should be noted that in other embodiments, the steps of the corresponding method are not necessarily performed in the order shown and described in this specification. In some other embodiments, the method may include more or fewer steps than those described in this specification. In addition, a single step described in this specification may be broken down into multiple steps for description in other embodiments, and multiple steps described in this specification may be combined into a single step for description in other embodiments.

[0050] Example 1

[0051] The scene of transplanting tissue culture seedlings into plug trays is as follows: Figure 2 As shown, the system primarily includes a tissue culture seedling transport line, plug trays, and a robotic system. The robotic system primarily comprises actuators and a perception system. The actuator structure includes a multi-degree-of-freedom robotic arm and a gripper-pushing end effector, while the perception system comprises a 2D camera, a built-in joint angle sensor in the robotic arm, and a thin-film force sensor.

[0052] Before the task begins, the tissue culture seedlings are randomly laid out on the assembly line. The robotic system then uses a 2D camera to identify the position of the target tissue culture seedling and the empty space in the tray. A path is then designed to guide the robotic arm from its initial position to above the target tissue culture seedling, ensuring that the spatial position of the end effector aligns with the seedling. The end effector then uses the clamping force output by a thin-film force sensor combined with a fuzzy PID algorithm to maintain a clamping force within the range of 0.3-0.5N, achieving flexible grasping. Finally, a path is used to guide the robotic arm to above the target tray, ensuring that the tissue culture seedling is upright, and the push function of the end effector is used to transplant the seedling.

[0053] In order to achieve this action, the present invention aims to propose a robot arm path planning method for tissue culture seedling transplanting, specifically, referring to Figure 1 , the method comprising:

[0054] S1. Use machine vision technology to identify the position of the target tissue culture seedlings and the empty space in the plug tray;

[0055] In this embodiment, a 2D camera is used to identify the position of the target tissue culture seedlings and the empty hole in the plug tray. For example, the 2D camera is used to capture images, and the pixel coordinates of the target (tissue culture seedlings / empty hole in the plug tray) are extracted using a general image processing algorithm (such as edge detection and feature matching). Then, the pixel coordinates are converted into three-dimensional world coordinates through camera calibration (internal reference / external reference calibration) to obtain 、 The 2D camera first extracts the Euler angle posture of the target (rotation angle around the x / y / z axis, such as pitch angle, yaw angle, roll angle), and then converts the Euler angle to quaternion form through the Euler angle to quaternion conversion formula (based on trigonometric function calculation). and , the specific conversion algorithm is the existing mature technology (such as Rodrigus transform).

[0056] S2. Define the state space including the position and posture of the tissue culture seedling and the plug tray, the joint angles of the robot arm, the position, posture and clamping force of the end effector;

[0057] In this embodiment, the state space is defined as follows:

[0058] Status of tissue culture seedlings and plug trays: including the positions of tissue culture seedlings and plug trays, recorded as and , where x, y, and z are the horizontal (x-axis), transverse (y-axis), and vertical (z-axis) coordinates of the tissue culture seedling in space, respectively, which are used to locate the spatial position of the seedling on the assembly line; the spatial posture of the tissue culture seedling and the plug tray expressed in quaternion form and . 、 The spatial posture of the tissue culture seedlings and plug trays expressed in the form of Euler angles can be extracted from the image data collected by the 2D camera through a general image processing algorithm, and further calculated by Euler angle conversion. and .

[0059] Among them, quaternion is a mathematical tool that describes the rotation in three-dimensional space through four components (three imaginary parts a, b, c and one real part w). It can avoid the "gimbal lock" singularity problem represented by Euler angles (such as multiple rotations around the same axis resulting in loss of freedom) and provide unambiguous gradient guidance for the precise posture adjustment of the end effector of the robotic arm.

[0060] Robotic arm state: including the angles of each joint of the robotic arm. Taking a 6-DOF robotic arm as an example, it is recorded as This parameter is directly obtained from the built-in angle sensor of the robot arm.

[0061] End effector state: is the spatial position of the end effector, recorded as , the spatial posture expressed in the form of quaternion, recorded as ; Clamping force, denoted as . The spatial posture of the end effector expressed in the form of Euler angles can be calculated by combining the joint angles of the manipulator and the length of the robot link with forward kinematics, and further calculated by Euler angle conversion . Acquired by the thin film force sensor on the end effector gripper.

[0062] Therefore, the overall state space can be expressed as , the dimension is 28.

[0063] S3. Define the motion space including the motion of each joint of the robot arm, the gripping motion of the end effector, and the pushing motion;

[0064] In this embodiment, the action space is defined as follows:

[0065] Joint motion: The motion of each joint of the robotic arm is recorded as , the angle range is [-90°, 90°].

[0066] End effector motion: The clamping motion is one-dimensional, including two states: release and clamping; the pushing motion is one-dimensional, including two states: retraction and extension.

[0067] Therefore, the overall action space dimension is 8.

[0068] S4. Design a reward function, which includes a reward for successful transplanting, a reward for avoiding obstacles, a reward for approaching the target seedling, a reward for approaching the target position in the plug tray, a reward for matching the posture of the target seedling, a reward for matching the posture of the plug tray transplanting, and a reward for time efficiency, and perform a weighted summation of each sub-reward function;

[0069] In this embodiment, the specific design of the reward function is as follows:

[0070] Reward function design is key to implementing DRL (deep reinforcement learning) methods. A reasonable reward function can reduce the number of blind attempts before obtaining a path with target characteristics, significantly improving DRL training efficiency and success rate. According to the transplantation task requirements:

[0071] 1) Successful transplanting R1: If the seedling is planted into the target plug tray and the clamping force is 0.3-0.5N, +1; if the seedling is successfully grasped but not transplanted, +0.5; otherwise, 0.

[0072] 2) Obstacle avoidance R2: Set to -1 when the distance between the terminal and the obstacle is less than the safety threshold (5mm), otherwise 0.

[0073] 3) Approaching the target seedling R3: When the robot arm approaches the target seedling, it is rewarded according to the distance between it and the target seedling. Let the distance between the end effector of the robot arm and the target seedling be d1, and define Under the guidance of this sub-reward function, the closer the robotic arm gets to the target, the higher the reward it receives, which encourages the robotic arm to guide the end effector to quickly approach the target seedling during the grasping phase.

[0074] 4) Approaching the target position R4 of the plug tray: When the robot arm approaches the target position of the plug tray, the distance between it and the target position is set to d2, and the reward function is defined Under the guidance of this sub-reward function, the robotic arm guides the end effector to quickly approach the target position of the plug tray during the transplanting phase.

[0075] 5) Target seedling posture matching R5: During the grasping stage, the robotic arm needs to adjust the posture of the end effector in real time to adapt to the grasping of seedlings with disordered postures.

[0076] Define the posture reward function , with a value ranging from 0 to 1, where 1 indicates a perfect match. This function encourages learning to adjust the end-effector's posture to accurately grasp the tissue culture seedling. Using quaternion space distance to calculate the matching degree of the end-effector's current posture overcomes the singularity problem of Euler angle representation and provides precise gradient guidance for complex 3D posture adjustments.

[0077] 6) Posture matching of plug tray transplantation R6: When transplanting tissue culture seedlings to the plug tray, it is necessary to consider the posture matching between the end effector of the robot arm and the plug tray to ensure the uprightness of the seedling. The reward function is .

[0078] 7) Time efficiency R7: Let the time it takes to complete a transplant task be t. The reward function can be set to R7 = 10 / t. This incentivizes the robot arm to complete the task as quickly as possible while ensuring transplant accuracy.

[0079] Finally, the reward function can be calculated by the following formula:

[0080]

[0081] in Seven sub-reward functions (N=7) were designed for different key objectives in the transplant task, corresponding to core requirements such as successful transplantation, obstacle avoidance, approaching the target, posture matching, and time efficiency. is the weight of each sub-function. According to expert experience, The values ​​are 300, 200, 5, 5, 4, 4, 1 respectively.

[0082] S5. A deep reinforcement learning network is constructed using the soft actor-critic (SAC) algorithm framework. The deep reinforcement learning network is trained using the state space, action space, and reward function to enable the deep reinforcement learning network to learn to generate the motion path of the robotic arm from its initial position to grasping the target tissue culture seedling and transplanting it to an empty space in the plug tray.

[0083] In this embodiment, the soft actor-critic (SAC) algorithm is used as the DRL algorithm framework for generating the robot arm path. Figure 3As shown in Figure 2, the core algorithm consists of one Actor (policy) network and two Critic (evaluation) networks. The Actor network inputs the state defined in the aforementioned state space, and outputs the action probability distribution. Its network structure consists of three fully connected layers with (128, 256, 128) hidden units. All three layers use Reinforced Lu (ReLU) activation, and the output layer uses a tanh function. The Critic network inputs the state and action, and outputs the state value. The Critic network uses two Q networks (Q1 and Q2) to reduce overestimation bias and improve the algorithm's stability and robustness. The Q network has two input dimensions: state and action. The input state passes through four layers with (256, 128, 256, 128) hidden units; the input action passes through three hidden layers with (128, 256, 128) hidden units. Before entering the third hidden layer, the outputs of the two branches, the input state and action, are merged and passed through the final single-dimensional output layer to output the action-state value Q.

[0084] In each training process, first, a state s0 is randomly initialized. In this state, the control action a is sampled from the actor network. t , and obtain the reward r at this time t , then the state transitions to the next state s t+1 Next, s t 、a t 、r t and s t+1 This is stored as a sample in the experience replay pool. Finally, a small batch of samples is drawn from the experience replay pool to train the network, and the network is updated using the stochastic gradient method. In this way, after each training process, the updated actor network can output the optimal robot arm path.

[0085] Compared with traditional path planning methods based on mathematical modeling or heuristic rules, the method proposed in this invention can autonomously extract key features from complex dynamic environments and generate adaptive strategies through an end-to-end strategy learning mechanism, and naturally has the generalization advantage of handling unknown scenarios.

[0086] Traditional methods rely on pre-set environmental models and deterministic rules. When faced with new scenarios, such as randomly changing seedling positions and diverse plug tray sizes, manual parameter adjustments or even remodeling are required, limiting flexibility and generalization. The SAC algorithm framework employed in this paper, however, encodes multi-dimensional information, such as seedling position, plug tray status, and robotic arm joint angles, into a 28-dimensional state space, enabling the DRL network to capture common features and dynamic relationships across diverse scenarios. During training, the network optimizes its strategy through trial-and-error interaction with the virtual environment, aiming to maximize long-term cumulative rewards. This results in an abstract understanding of the entire "grasp-transfer-insert" process in the transplanting task. This data-driven learning approach eliminates the need for additional programming of the trained DRL network for new scenarios (such as varying plug tray spacing and densely packed seedlings). Instead, the trained DRL network can dynamically adjust joint motion and end-point posture based on real-time input from the state space, automatically adapting to the diverse randomness and uncertainty inherent in seedling transplanting scenarios. This significantly reduces the system's reliance on specific scenarios and provides a stable and reliable path planning solution for complex practical applications.

[0087] Example 2

[0088] Based on the same concept, the present invention also proposes a robot arm path planning method and device for tissue culture seedling transplanting, comprising:

[0089] The posture acquisition module uses machine vision technology to identify the posture of the target tissue culture seedlings and the empty space in the plug tray;

[0090] Definition module, which defines the state space including the position and posture of the tissue culture seedling and the plug tray, the joint angles of the robot arm, the position, posture and clamping force of the end effector; and defines the action space including the motion of each joint of the robot arm, the clamping action and the pushing action of the end effector;

[0091] The reward design module designs reward functions, including rewards for successful transplanting, obstacle avoidance, proximity to the target seedling, proximity to the target location in the plug tray, target seedling posture matching, plug tray transplant posture matching, and time efficiency. The reward functions are then weighted and summed.

[0092] The model construction module uses the soft actor-critic algorithm framework to build a deep reinforcement learning network. The deep reinforcement learning network is trained through state space, action space and reward function, so that the deep reinforcement learning network can learn to generate the motion path of the robotic arm from the initial position to grasping the target tissue culture seedling and transplanting it to the empty space in the hole tray.

[0093] Example 3

[0094] This embodiment also provides an electronic device, referring to Figure 4, includes a memory 404 and a processor 402, wherein the memory 404 stores a computer program, and the processor 402 is configured to run the computer program to perform the steps in any of the above method embodiments.

[0095] Specifically, the processor 402 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits for implementing the embodiments of the present invention.

[0096] Memory 404 may include a large-capacity memory 404 for data or instructions. By way of example, and not limitation, memory 404 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 404 may include removable or non-removable (or fixed) media. Where appropriate, memory 404 may be internal or external to the data processing device. In certain embodiments, memory 404 is non-volatile memory. In certain embodiments, memory 404 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM) or a flash memory (FLASH), or a combination of two or more of these. In appropriate circumstances, the RAM may be a static random access memory (SRAM) or a dynamic random access memory (DRAM), wherein the DRAM may be a fast page mode dynamic random access memory 404 (FPMDRAM), an extended data output dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0097] The memory 404 may be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor 402 .

[0098] The processor 402 reads and executes computer program instructions stored in the memory 404 to implement any one of the robot arm path planning methods for tissue culture seedling transplanting in the above embodiments.

[0099] Optionally, the electronic device may further include a transmission device 406 and an input / output device 408 , wherein the transmission device 406 is connected to the processor 402 , and the input / output device 408 is connected to the processor 402 .

[0100] Transmission device 406 can be used to receive or transmit data via a network. Specific examples of such networks may include wired or wireless networks provided by the electronic device's communications provider. In one embodiment, the transmission device includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, transmission device 406 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0101] The input / output device 408 is used to input or output information.

[0102] Example 4

[0103] This embodiment also provides a readable storage medium, in which a computer program is stored. The computer program includes a program code for controlling a process to execute a process. The process includes the robotic arm path planning method for tissue culture seedling transplanting according to the first embodiment.

[0104] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be repeated here.

[0105] In general, various embodiments may be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention may be implemented in hardware, while other aspects may be implemented in firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flow charts, or using some other graphical representation, it should be understood that, as non-limiting examples, the blocks, devices, systems, techniques, or methods described herein may be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or a controller or other computing device, or some combination thereof.

[0106] The embodiments of the present invention may be implemented by computer software that is executable by a data processor of a mobile device, such as in a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products) including software routines, applets and / or macros may be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. A computer program product may include one or more computer executable components that are configured to perform an embodiment when the program is run. One or more computer executable components may be at least one software code or a portion thereof. In addition, it should be noted at this point that, for example, Figure 1 Any block of the logic flow in the program may represent program steps, or interconnected logic circuits, blocks and functions, or a combination of program steps and logic circuits, blocks and functions. The software may be stored on physical media such as memory chips or memory blocks implemented within the processor, magnetic media such as hard disks or floppy disks, and optical media such as, for example, DVDs and their data variants, CDs, etc. Physical media are non-transitory media.

[0107] Those skilled in the art should understand that the technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0108] The above embodiments merely illustrate several embodiments of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of the present invention. Therefore, the scope of the present invention shall be determined by the appended claims.

Claims

1. A robot arm path planning method for tissue culture seedling transplanting, characterized in that: The following steps are involved: S1. Use machine vision technology to identify the position of the target tissue culture seedlings and the empty space in the plug tray; S2. Define the state space including the position and posture of the tissue culture seedling and the plug tray, the joint angles of the robot arm, the position, posture and clamping force of the end effector; S3. Define the motion space including the motion of each joint of the robot arm, the gripping motion of the end effector, and the pushing motion; S4. Design a reward function, wherein the reward function includes a successful transplanting reward, an obstacle avoidance reward, a target seedling approaching reward, a plug tray target position approaching reward, a target seedling posture matching reward, a plug tray transplanting posture matching reward, and a time efficiency reward, and perform a weighted summation of each sub-reward function; S5. A deep reinforcement learning network is constructed using a soft actor-critic algorithm framework. The deep reinforcement learning network is trained using the state space, action space, and reward function, so that the deep reinforcement learning network learns to generate a motion path for the robotic arm from an initial position to grasping the target tissue culture seedling and transplanting it to an empty space in the plug tray.

2. A robot arm path planning method for tissue culture seedling transplanting according to claim 1, characterized in that: In step S2, the positions of the tissue culture seedlings and the plug tray are three-dimensional coordinates, and the postures are represented by quaternions; The joint angle of the robotic arm is obtained by the built-in angle sensor of the robotic arm; The position of the end effector is calculated by the forward kinematics of the robot arm, the posture is expressed by quaternion, and the clamping force is obtained by the thin film force sensor on the clamping claw of the end effector.

3. A robot arm path planning method for tissue culture seedling transplanting according to claim 1, characterized in that: In step S3, the angle range of motion of each joint of the robotic arm is [-90°, 90°]; The clamping action of the end effector includes two states: release and clamping, and the pushing action includes two states: contraction and extension.

4. A robot arm path planning method for tissue culture seedling transplanting according to claim 1, characterized in that: In step S4, the successful transplanting reward is: when the tissue culture seedling is planted into the target plug tray and the clamping force is 0.3-0.5N, a first preset reward value is obtained; when the tissue culture seedling is successfully grasped but not transplanted, a second preset reward value is obtained; otherwise, the reward is zero; The obstacle avoidance reward is: a negative reward is obtained when the distance between the end effector and the obstacle is detected to be less than a safety threshold, otherwise it is zero; The rewards for being close to the target seedling and the target location of the plug tray are both generated based on the distance between the end effector and the target. The smaller the distance, the higher the reward. The target seedling posture matching reward and the plug tray transplanting posture matching reward are both generated based on the quaternion space distance between the end effector and the target. The smaller the distance, the higher the reward. The time efficiency reward is generated based on the time it takes to complete a transplanting task, and the shorter the time, the higher the reward.

5. A robot arm path planning method for tissue culture seedling transplanting according to claim 4, characterized in that: The reward function is obtained by multiplying each sub-reward function by the corresponding weight and summing them up. The weight is set according to expert experience, where the weight of the successful transplanting reward is 300, the weight of the obstacle avoidance reward is 200, the weight of the reward for approaching the target seedling and the reward for approaching the target position of the plug tray are both 5, the weight of the target seedling posture matching reward and the plug tray transplanting posture matching reward are both 4, and the weight of the time efficiency reward is 1.

6. A robot arm path planning method for tissue culture seedling transplanting according to claim 1, characterized in that: In step S5, the soft actor-critic algorithm framework includes an actor network and two critic networks. The actor network inputs the state space and outputs the action probability distribution, and the critic network inputs the state and action and outputs the state value. The Actor network consists of three fully connected layers with 128, 256, and 128 hidden units, respectively. The first two layers use the ReLU activation function, and the output layer uses the tanh function. In each Q network of the Critic network, the state input is processed by four fully connected layers, with the number of hidden units being 256, 128, 256, and 128 respectively. The action input is processed by three fully connected layers, with the number of hidden units being 128, 256, and 128 respectively. The two are merged before the third hidden layer to output a single-dimensional state value.

7. A robot arm path planning method for tissue culture seedling transplanting according to any one of claims 1 to 6, characterized in that: In step S5, when grasping the target tissue culture seedling, the clamping force output by the thin film force sensor is combined with the fuzzy PID algorithm to control the clamping force within a preset range to achieve flexible grasping; When transplanting to the plug tray, adjust the posture of the end effector to make the tissue culture seedlings upright, and complete the transplanting by pushing.

8. A method and device for robot arm path planning for tissue culture seedling transplanting, characterized in that: include: The posture acquisition module uses machine vision technology to identify the posture of the target tissue culture seedlings and the empty space in the plug tray; Definition module, which defines the state space including the position and posture of the tissue culture seedling and the plug tray, the joint angles of the robot arm, the position, posture and clamping force of the end effector; and defines the action space including the motion of each joint of the robot arm, the clamping action and the pushing action of the end effector; The reward design module designs reward functions, including rewards for successful transplanting, obstacle avoidance, proximity to the target seedling, proximity to the target location in the plug tray, target seedling posture matching, plug tray transplant posture matching, and time efficiency. The reward functions are then weighted and summed. The model construction module uses the soft actor-critic algorithm framework to build a deep reinforcement learning network. The deep reinforcement learning network is trained through state space, action space and reward function, so that the deep reinforcement learning network can learn to generate the motion path of the robotic arm from the initial position to grasping the target tissue culture seedling and transplanting it to the empty space in the hole tray.

9. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to execute the robotic arm path planning method for tissue culture seedling transplanting according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which includes a program code for controlling a process to execute a process, and the process includes the robotic arm path planning method for tissue culture seedling transplanting according to any one of claims 1 to 7.

Images