Motion control method and system for drug loading of magnetic control micro-nano robot in simulated environment

Abstract

The invention relates to the technical field of micro-nano robots, and provides a motion control method and system for a magnetic control micro-nano robot carrying medicine in a simulation environment, and the method comprises the steps: obtaining a motion image of a magnetic control silicon dioxide-based micro-nano robot cluster in the simulation environment in real time; according to a preset transportation task, the current state of the micro-nano robot cluster is judged; based on the current state of the micro-nano robot cluster, establishing a magnetic field generated by the micro-nano robot cluster and the three-dimensional Helmholtz coil, a fluid resistance field of a simulation environment and a dynamic model of a simulation environment channel boundary condition; and based on the motion image acquired in real time, using a deep learning model to identify real-time pose information of the micro-nano robot cluster and the simulated environment channel boundary information, and using a path planning algorithm to perform trajectory planning according to the real-time pose information and the simulated environment channel boundary information. And the magnetic field is adjusted to control the movement of the micro-nano robot cluster.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of micro-nano robots, in particular to a motion control method and system of a magnetic micro-nano robot for drug delivery in a simulation environment. BACKGROUND

[0002] Micro-nano robot clusters have great application potential in biomedical fields such as targeted drug delivery and minimally invasive surgery. Magnetic control, as a non-contact driving method, has become one of the mainstream technologies for driving micro-nano robots due to its good biocompatibility and penetration. Currently, some researches have used devices such as Helmholtz coils to generate magnetic fields to drive and control micro-nano robots. However, existing technologies mostly rely on open-loop control strategies, lack real-time perception and feedback of cluster state and environmental information, resulting in insufficient navigation accuracy and poor adaptability in complex and unknown simulation physiological environments. In addition, existing magnetic field control models are relatively single, making it difficult to achieve flexible and robust switching of multi-modal behaviors of the cluster (such as band-shaped transportation and vortex drug delivery). Meanwhile, traditional path planning algorithms do not fully consider fluid resistance, channel boundaries and cluster dynamics at the microscale, which can easily lead to problems such as adhesion of the motion path and unstable turning, and cannot meet the high requirements of precise drug delivery tasks for automated navigation and control. Therefore, there is an urgent need for a closed-loop control system that integrates real-time visual feedback, multi-modal magnetic field regulation and intelligent path planning to improve the autonomous motion and drug delivery capabilities of micro-nano robot clusters in simulation environments. SUMMARY

[0003] To overcome the above-mentioned defects, the present application aims to provide a motion control method and system of a magnetic micro-nano robot for drug delivery in a simulation environment, which aims to build a closed-loop control system integrating real-time visual feedback, multi-modal magnetic field accurate regulation and intelligent path planning, to realize autonomous, precise and robust navigation and drug delivery transportation of micro-nano robot clusters in complex and unknown simulation environments.

[0004] To achieve this purpose, the present application adopts the following technical solutions: The motion control method of a magnetic micro-nano robot for drug delivery in a simulation environment comprises the following steps: S1: Real-time acquisition of motion images of a magnetic silica-based micro-nano robot cluster in a simulation environment; S2: According to a pre-set transportation task, determine the current state of the micro-nano robot cluster, wherein the current state includes a band-shaped cluster state when no drug is obtained and a vortex-shaped cluster state when drug is obtained; S3: Based on the current state of the micro-nano robot cluster, establish a dynamics model of the micro-nano robot cluster, the magnetic field generated by a three-dimensional Helmholtz coil, the fluid resistance field of the simulation environment and the boundary conditions of the simulation environment channel, and associate a power amplifier matched with the three-dimensional Helmholtz coil to maintain the magnetic field; S4: Based on the real-time acquired motion images, a deep learning model is used to identify the real-time pose information of the micro-nano robot cluster and the channel boundary information of the simulated environment, and a path planning algorithm is used to plan the trajectory according to the real-time pose information and the channel boundary information of the simulated environment to generate control signals; S5: The control signal is converted into an analog signal through the control board, and then amplified by the power amplifier to drive the three-dimensional Helmholtz coil to adjust the magnetic field and thus control the movement of the micro-nano robot cluster.

[0005] Preferably, in step S3, if the magnetic field is a one-dimensional static magnetic field, it satisfies the following relationship: ; in, Indicates the amplitude of magnetic flux density. This represents the direction vector of the static magnetic field in three-dimensional space; If the magnetic field is a one-dimensional alternating magnetic field, it satisfies the following relationship: ; in, This represents the angular frequency of the alternating magnetic field. This represents the direction vector of the alternating magnetic field in three-dimensional space.

[0006] Preferably, in step S3, if the magnetic field is a two-dimensional magnetic field, the pose relationship between the magnetic field and the absolute coordinate system of the three-dimensional Helmholtz coil is first established, by confirming the rotation axis direction vector of the two-dimensional rotating field. The angle between the rotation field and the z-axis of the absolute coordinate system determines the direction and spatial coordinates of the two-dimensional rotation field. Wherein, when the origin of the rotation field coordinates is the origin of the absolute coordinate system, the rotation axis direction vector is defined. The angle between the absolute coordinate system and the positive z-axis is Define the direction vector of the rotation axis. The angle between the projection of the xoy plane in the absolute coordinate system and the positive x-axis direction is . And establish the coordinate space coordinates of the two-dimensional rotating field: ; ; ; If the magnetic field is a two-dimensional rotating field, the magnetic induction intensity vector of the two-dimensional rotating field... Satisfying the relation: ; ; in, This represents the magnetic field frequency of the two-dimensional rotating field. Indicates the magnitude of the magnetic field component along the x-axis. With the amplitude of the y-axis magnetic field component The ratio, This represents the movement speed control parameters of the cluster. Indicates the control parameters for the movement direction of the cluster. , and These represent the amplitudes of the magnetic field components along the x, y, and z axes, respectively.

[0007] In step S3, if the magnetic field is a two-dimensional magnetic field, the pose relationship between the magnetic field and the absolute coordinate system of the three-dimensional Helmholtz coil is first established by confirming the rotation axis direction vector of the two-dimensional rotating field. The angle between the rotation field and the z-axis of the absolute coordinate system determines the direction and spatial coordinates of the two-dimensional rotation field. Wherein, when the origin of the rotation field coordinates is the origin of the absolute coordinate system, the rotation axis direction vector is defined. The angle between the absolute coordinate system and the positive z-axis is Define the direction vector of the rotation axis. The angle between the projection of the xoy plane in the absolute coordinate system and the positive x-axis direction is . And establish the coordinate space coordinates of the two-dimensional rotating field: ; ; ; If the magnetic field is a two-dimensional rotating field, the magnetic induction intensity vector of the two-dimensional rotating field... Satisfying the relation: ; ; in, This represents the magnetic field frequency of the two-dimensional rotating field. Indicates the magnitude of the magnetic field component along the x-axis. With the amplitude of the y-axis magnetic field component The ratio, This represents the movement speed control parameters of the cluster. Indicates the control parameters for the movement direction of the cluster. , and These represent the amplitudes of the magnetic field components along the x, y, and z axes, respectively.

[0008] Preferably, if the magnetic field is a two-dimensional oscillating field, the magnetic induction intensity vector of the two-dimensional oscillating field is... Satisfying the relation: ; ; Wherein, rotation matrix , and They respectively satisfy the following relations: ; ; ; in, The magnetic field frequency represents the frequency of a two-dimensional oscillating field. This represents the magnetic flux density vector in the absolute coordinate system of the three-dimensional Helmholtz coil after rotational transformation. This represents the movement speed control parameters of the cluster. Indicates the control parameters for the movement direction of the cluster. The orientation control parameter of the xoy plane, representing the motion coordinate system of the oscillation field, is... and These represent the amplitudes of the magnetic field components along the x-axis and y-axis, respectively.

[0009] Preferably, in step S3, if the magnetic field is a three-dimensional dynamic segmented field, the magnetic induction intensity vector of the three-dimensional dynamic segmented field... Satisfying the relation: ; in, The magnetic field frequency represents the three-dimensional dynamic segmented field. Indicates time A mathematical function that rounds up. , and These represent the magnetic induction intensity vectors of the three-dimensional dynamic segmented field. Components in the x, y, and z axes.

[0010] Preferably, based on the real-time acquired motion images, using a deep learning model to identify the real-time pose information of the micro-nano robot cluster and the channel boundary information of the simulated environment includes: A cluster modal feature dataset is constructed based on the real-time acquired motion images. The cluster modal feature dataset includes the modal feature information of the cluster and the channel information of the simulation environment. The cluster modal feature dataset is trained using a target detection neural network based on the YOLOv5 architecture to identify and track the micro-nano robot cluster and obtain its real-time pose information. The YOLOv5 network structure includes a Backbone module, a Neck module, and a Head module. The YOLOv5 network extracts image features through the Backbone module, performs multi-scale feature fusion through the Neck module, and outputs detection results including the pose information of the micro-nano robot cluster through the Head module.

[0011] Preferably, trajectory planning using a path planning algorithm based on the real-time pose information and the channel boundary information of the simulated environment includes: Adopting traditional The algorithm is an optimized hybrid path planning algorithm that satisfies the following relation: ; ; in, This indicates the nodes passed from the starting point. The total cost evaluation function to the destination, Indicates the distance from the starting point to the node. The actual cost, Indicates from node Optimized estimated cost to reach the destination Represents the flow field evaluation function. This represents the distance evaluation function. Represents the curvature evaluation function. This represents the weighting coefficients of the comprehensive penalty function. This represents the weighting coefficients of the distance evaluation function. Represents the weighting coefficients of the curvature evaluation function; and Satisfying the relation: ; ; ; ; ; ; ; in, Represents a node Distance to the nearest obstacle Represents the set of obstacle nodes. Represents the set of walkable nodes. This represents the maximum distance from all passable nodes to the obstacle. Represents the distance transformation graph at the nodes Laplace response at the location, This represents the negative part of the Laplace response. This represents the maximum value of the negative part of the Laplace response for all traversable nodes. It is a very small constant set to avoid division by zero errors.

[0012] The motion control system for drug delivery in a simulated environment by a magnetically controlled micro-nano robot includes an ultra-depth-of-field optical microscope, a camera, a three-axis microscopic observation platform, a control terminal computer, a control board, a power amplifier, and a three-dimensional Helmholtz coil. The ultra-depth-of-field optical microscope is optically connected to the camera to observe the motion state of the micro-nano robot cluster in a simulated environment and transmit optical images to the camera; The camera is mechanically connected to the three-axis microscopic observation platform and is used to record image information observed by the ultra-depth-of-field optical microscope. The three-axis microscopic observation platform is equipped with a cluster simulation environment channel, which can be moved at the micrometer level by adjusting the knob to adjust the pose of the cluster simulation environment channel in the camera's field of view. The camera is connected to the control terminal computer and transmits the captured real-time video data stream to the control terminal computer. The control terminal computer is connected to the control board via USB. The control terminal computer is used to identify the real-time pose information and simulated environment channel boundary information of the micro-nano robot cluster based on the real-time video data stream using the deep learning YOLOv5 model, and uses a path planning algorithm to perform trajectory planning based on the real-time pose information and simulated environment channel boundary information to generate digital control signals. The control board is connected to the power amplifier via a BNC to DuPont wire, which is used to convert the digital control signals generated by the control terminal computer into analog circuit signals. The power amplifier is connected to the three-dimensional Helmholtz coil via a BNC connector to a banana plug, and is used to amplify the analog circuit signal to drive the three-dimensional Helmholtz coil. The three-dimensional Helmholtz coil is used to generate a corresponding magnetic field based on the amplified analog circuit signal, and the movement state of the micro-nano robot cluster is controlled by adjusting the magnetic field parameters.

[0013] One of the above technical solutions has the following advantages or beneficial effects: This invention overcomes the limitations of existing technologies that cannot perceive cluster status and environmental information by acquiring real-time motion images of micro-nano robot swarms. It then determines the appropriate banded or vortex state of the swarm based on transportation task requirements, enabling proactive identification and flexible switching of multimodal behavior. This overcomes the shortcomings of single magnetic field models, which are ill-suited to adapting to different stages of drug delivery. Furthermore, a comprehensive dynamic model encompassing magnetic field, fluid resistance, and boundary conditions is established and coupled with a power amplifier to maintain stable magnetic field output, ensuring the accuracy and continuity of multimodal magnetic field control. This transforms swarm morphology transitions from traditional extensive control to refined proactive management. Subsequently, a deep learning model is used to identify the swarm pose and channel boundaries in real time. Combined with an intelligent path planning algorithm, trajectory control signals are generated. By introducing environmental constraints and swarm dynamic characteristics to optimize the path generation logic, the path-hugging and turning instability phenomena that are common in traditional algorithms at the microscale are effectively avoided. The final control signal is converted and amplified by the board to drive the Helmholtz coil, forming a complete closed loop of "perception-decision-execution-re-perception". It can dynamically adjust the control strategy according to environmental changes, significantly improving the autonomous navigation capability, morphological switching robustness, and drug delivery stability and accuracy of micro-nano robot clusters in complex and unknown simulated environments, and realizing the leap from passive execution to autonomous intelligent control. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0015] Figure 1 This is a flowchart of the motion control method for a magnetically controlled micro / nano robot to deliver drugs in a simulated environment, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the YOLOv5 structure of the motion control method for drug delivery in a simulated environment provided in this embodiment of the invention. Detailed Implementation

[0016] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0017] In this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0018] A motion control method for drug delivery in a simulated environment using magnetically controlled micro / nano robots, such as Figure 1 As shown, it includes the following steps: S1: Real-time acquisition of motion images of a magnetically controlled silica-based micro / nano robot swarm in a simulated environment; It should be noted that a magnetically controlled silica-based micro / nano robot swarm refers to a group system composed of multiple individual robots with sizes on the micro / nano scale, made of magnetic silica material. This swarm can generate cooperative motion under the drive of an external magnetic field, and its morphology can transform between a dispersed state, a ribbon-like shape, or a vortex shape. The simulated environment refers to a microscopic fluid channel network constructed based on human physiological characteristics, used to simulate fluid resistance, boundary constraints, and drug delivery scenarios within blood vessels or interstitial spaces. Motion images refer to a visual data stream continuously acquired through optical imaging equipment, containing information on the swarm's outline, position, orientation, and dynamic behavior. The core technology employed in step S1 is a collaborative imaging architecture combining a super-depth-of-field optical microscope and a high-frame-rate camera. The super-depth-of-field optical microscope provides continuously adjustable magnification from 10x to 80x, ensuring clear imaging depth at the micro- and nano-scale. The camera features a 2600 (horizontal) × 2160 (vertical) pixel resolution, a 1 / 2-inch target surface size, and a 2.5 μm × 2.5 μm pixel size, transmitting image data at a rate of 70.3 frames per second. It supports dynamic configuration of camera parameters via direct SDK interface calls through a programming language, including real-time adjustments to attributes such as image height, width, exposure mode, exposure time, and acquisition frame rate. This ensures the capture of the cluster's instantaneous state without motion blur or information loss during high-speed cluster movement. The three-axis microscopic observation platform, as a mechanical support device, supports the simulated environment channel and enables micron-level precision spatial pose adjustment. By adjusting the platform knobs, the channel area is kept centered in the microscope's field of view, preventing image shift and target loss due to field-of-view jitter.

[0019] Understandably, the use of depth-of-field extension technology in optical microscopy overcomes the limitations of focal depth at the microscale. Combined with a high-sensitivity camera, it achieves a balance of spatiotemporal resolution, ensuring that each frame of the image at a high frame rate of 70.3fps clearly presents the morphological features and spatial location of the cluster. This avoids control command lag caused by imaging blur or delay. By visualizing invisible microscopic movements through an active sensing mechanism, the control system can form a real-time understanding of the current state of the cluster and the structure of the environment, laying a data foundation for building closed-loop feedback and improving the timeliness and accuracy of subsequent decisions.

[0020] Specifically, the implementation of step S1 is divided into two sub-stages: First, in the system deployment stage, the official SDK package provided by the camera manufacturer needs to be downloaded, and a test program needs to be run on the control terminal computer to verify the availability of the SDK interface. This ensures that the camera device can be successfully opened, the image acquisition mode can be set, and test images can be acquired through function calls. This stage requires confirming the compatibility of the camera driver with the operating system and the SDK's support for programming languages ​​such as Python and C++. Second, in the system integration stage, the dynamic link library of the camera SDK needs to be imported into an integrated development environment such as PyCharm. Code is written to instantiate the camera object, configure parameters, and register image callback functions. For example, the camera's SetExposureTime function is used to set the exposure time to match the cluster's movement speed, the SetFrameRate function is used to lock the acquisition frequency at 70.3fps, and the StartGrabbing function is used to start continuous acquisition. Image data is then acquired in the callback function and pushed to the subsequent processing queue. This implementation avoids the performance loss caused by third-party software intermediaries, achieving seamless integration from hardware acquisition to software processing. It ensures that the transmission latency of image data from the sensor to memory is less than 10 milliseconds, meeting the time constraints of real-time control.

[0021] S2: Determine the current state of the micro-nano robot cluster according to the pre-set transportation task, wherein the current state includes the strip cluster state when no drug has been obtained and the vortex cluster state when the drug has been obtained. It should be noted that the pre-set transportation task refers to the task script pre-loaded by the control system, which includes the coordinates of the drug loading point, the coordinates of the unloading point, and key nodes along the intermediate path. This script defines the macroscopic goals and state switching logic of the cluster movement. The ribbon-like cluster state refers to the linear aggregation form of the micro-nano robot cluster under magnetic field control, suitable for rapid movement in straight channels. The vortex-like cluster state refers to the ring-shaped rotating structure formed by the cluster under the action of a rotating magnetic field, which can generate a low-pressure zone in the central area to adsorb and encapsulate drug particles, suitable for stable drug loading and transport. The core technology of step S2 lies in the state machine logic judgment mechanism based on the task stage. Its judgment is not based on direct visual recognition, but on state transition control based on the task progress flag. When the system does not receive a drug loading completion signal, the cluster is in the ribbon-like state by default to facilitate rapid approach to the drug source. When loading is completed, the system automatically triggers the state transition command to switch the cluster to the vortex state to achieve centrifugal adsorption and encapsulation of the drug. The input for state judgment is the Boolean flag in the task management module, and the output is a discrete state identifier code (such as 0x01 representing the strip state and 0x02 representing the vortex state). This identifier code serves as the basis for selecting the mode for subsequent magnetic field parameter generation.

[0022] Understandably, the purpose of determining the current state is to achieve dynamic matching between the cluster configuration and task requirements. By decomposing the macro-transportation task into two sub-stages—"empty navigation" and "drug-loaded transportation"—and configuring the optimal cluster configuration for each stage, the stability of drug carrying is maximized while ensuring transportation efficiency. The principle is that the ribbon-like configuration has low fluid resistance and high maneuverability, making it suitable for long-distance, high-speed movement, while the vortex configuration generates centripetal force through a rotating flow field, effectively capturing and fixing drug particles and preventing them from falling off during transportation due to fluid disturbances or cluster vibrations. This state machine mechanism effectively avoids the contradiction between speed and stability that cannot be simultaneously achieved with a single configuration, realizing modal adaptive optimization in multi-task scenarios. This allows the control system to autonomously decide the cluster configuration based on task logic without manual intervention, improving the system's automation level and task execution efficiency.

[0023] S3: Based on the current state of the micro-nano robot cluster, establish a dynamic model of the magnetic field generated by the micro-nano robot cluster and the three-dimensional Helmholtz coil, the fluid resistance field of the simulated environment, and the channel boundary conditions of the simulated environment, and associate it with the power amplifier matched with the three-dimensional Helmholtz coil to maintain the magnetic field. It should be noted that a three-dimensional Helmholtz coil refers to a magnetic field generating device composed of three sets of mutually orthogonal circular coil pairs. When current is passed through each coil pair, a uniform magnetic field is generated in its central region. The synergistic excitation of the three sets of coils can synthesize a three-dimensional spatial magnetic field of arbitrary direction and intensity. A power amplifier is an electronic device that amplifies the power of the low-voltage analog signal output from the control board. Its output current range needs to match the impedance characteristics of the Helmholtz coil. Typically, the output current is required to be continuously adjustable from 0 to 5 amps, with a bandwidth of not less than 100Hz to ensure the fidelity of the dynamic magnetic field waveform. A dynamic model refers to a set of differential equations describing the motion of a cluster of micro- and nano-robots under the combined action of magnetic field driving force, fluid resistance, and boundary constraints. The magnetic field driving force is calculated by multiplying the magnetization intensity and the magnetic field gradient. The fluid resistance adopts the Stokes drag formula, and the boundary conditions are simulated by introducing a wall repulsion potential function to simulate the near-wall effect. The core of step S3 is to construct a multiphysics model that couples magnetic field, fluid, and boundary, and to establish a mapping relationship between the model output and the input current of the power amplifier. Specifically, the required current value is calculated by back-calculating the magnetic induction intensity of the three-dimensional Helmholtz coil, and then the current waveform parameters are encoded into a digital instruction sequence that can be recognized by the control board.

[0024] Understandably, the purpose of establishing a dynamic model is to transform the state requirements of step S2 into executable magnetic field parameter commands. By quantifying the magnetic field strength, gradient, and rate of change of time required for different cluster morphologies through physical modeling, and considering the attenuation effect of fluid resistance on cluster velocity and the limiting effect of channel boundaries on motion trajectory, the generated control signals can be accurately driven to move the cluster in the expected morphology. The principle is that the magnetically controlled silica-based micro-nano robots have superparamagnetism, and the magnetic torque they experience in the magnetic field aligns their long axis along the direction of the magnetic field. The rotating or oscillating magnetic field induces the cluster to generate collective behavior patterns through periodic torque. The dynamic model predicts the stable morphology and motion velocity of the cluster under given magnetic field parameters by solving the coupled system of the cluster's center of mass motion equation and individual orientation equation. This guides the power amplifier to output a matching current waveform to maintain the magnetic field. In the modeling process, the abstract state description is transformed into specific electromagnetic parameters, realizing a cross-domain mapping from the task layer to the execution layer. This ensures that the magnetic field generation matches the cluster's dynamic characteristics and avoids morphological instability or motion deviation caused by parameter mismatch.

[0025] Specifically, the implementation process unfolds in three modes based on the status flag output in step S2: When the status flag is a strip-shaped cluster, the system sets the magnetic field frequency to within the range of 10Hz to 20Hz, and the amplitude ratio coefficient to between 0.5 and 4. By adjusting the coefficient, the width-to-thickness ratio of the strip is controlled. The reciprocating magnetic torque of the oscillating field is used to make the robot linearly aligned along the direction of magnetic field vibration. At the same time, by jointly setting the motion direction control parameters and orientation control parameters, the long axis of the strip-shaped cluster is aligned with the tangent direction of the planned path to achieve efficient forward movement. When the status flag is a vortex-shaped cluster, the system calls the rotation field mode, sets the frequency to between 6Hz and 12Hz to form a stable vortex, uses the speed control parameters to adjust the cluster rotation angular velocity to match the fluid resistance, and uses the orientation control parameters to determine the spatial orientation of the vortex plane, making the vortex axis perpendicular to the channel axis to maximize the drug adsorption area. When the status flag is a three-dimensional dynamic segmented field, the system activates the segmented mode, sets the frequency to between 20Hz and 50Hz, and uses the round-up function to achieve a magnetic field mode jump every 0.2 seconds, splitting the cluster into multiple subgroups and driving them separately.

[0026] S4: Based on the real-time acquired motion images, a deep learning model is used to identify the real-time pose information of the micro-nano robot cluster and the channel boundary information of the simulated environment, and a path planning algorithm is used to plan the trajectory according to the real-time pose information and the channel boundary information of the simulated environment to generate control signals; It should be noted that deep learning models refer to machine learning models built on convolutional neural network architectures that can automatically extract multi-level features from raw images and perform object detection and classification. This embodiment uses the YOLOv5 architecture, with its Backbone module employing a CSPDarknet53 structure, achieving gradient splitting and feature reuse through cross-stage partial connections. The Neck module uses an FPN+PAN structure to achieve multi-scale feature fusion, and the Head module uses an anchor-based detection head to output bounding boxes, confidence scores, and class probabilities. Pose information refers to the position of the micro / nano robot cluster in the two-dimensional image plane. The centroid coordinates, principal axis orientation angles, and bounding box dimensions are used to determine the cluster's location, orientation angles, and bounding box dimensions. The centroid coordinates are used to locate the cluster, the orientation angles are used to determine the cluster's direction of movement, and the bounding box dimensions are used to assess the cluster's aggregation degree and morphological category. The simulated environment channel boundary information refers to the channel wall contour lines extracted from the background using an image segmentation algorithm. Its mathematical expression is a set of ordered points used to construct a cost map for navigation. The control signal refers to a digitally encoded frame containing magnetic field parameter commands, motion speed commands, and direction commands. Its format includes fields such as frame header, data length, command type, current setting values ​​for each axis coil, and CRC checksum, used to drive subsequent hardware execution.

[0027] Understandably, the purpose of using deep learning models to identify pose information is to overcome the instability of traditional image processing methods at the micro- and nano-scale due to low signal-to-noise ratio and variable target shape. YOLOv5's end-to-end training mechanism enables it to automatically learn robust feature representations of clusters under different modalities, and can still stably track targets even under partial occlusion, uneven lighting, or background noise interference, thereby providing continuous and accurate position input for path planning.

[0028] S5: The control signal is converted into an analog signal through the control board, and then amplified by the power amplifier to drive the three-dimensional Helmholtz coil to adjust the magnetic field and thus control the movement of the micro-nano robot cluster.

[0029] It should be noted that a control board is a hardware device used to convert digital control signals into analog voltage signals; analog signals refer to continuously changing electrical signals, typically voltage or current; a power amplifier is an electronic device used to amplify analog signals to drive a load, in this case referring to a Helmholtz coil; a three-dimensional Helmholtz coil consists of three sets of orthogonal coils, through which an amplified current is passed to generate a controllable magnetic field. Adjusting the magnetic field refers to changing the amplitude, direction, or frequency of the magnetic induction intensity to affect the movement of the cluster.

[0030] Understandably, the purpose of step S5 is to convert the planned control commands into physical actions, generate the target magnetic field by driving the Helmholtz coil, and thus precisely control the movement of the cluster. This achieves a closed loop from digital calculation to physical execution, ensuring that the cluster can move according to the planned path, solving the problems of response lag and inaccuracy in open-loop control, and improving the real-time performance and reliability of the entire system.

[0031] Specifically, a digital control signal can be generated by a control terminal computer and sent to a control board via a USB connection. The control board converts the digital signal into an analog voltage signal (e.g., 0-5V), which is then connected to a power amplifier via BNC-to-DuPont wires. The power amplifier converts the voltage signal into an amplified current signal (e.g., up to 50A), driving a three-dimensional Helmholtz coil. The coil generates a corresponding magnetic field based on the current, adjusting the movement state of the cluster to achieve precise control.

[0032] Preferably, in step S3, if the magnetic field is a one-dimensional static magnetic field, it satisfies the following relationship: ; in, Indicates the amplitude of magnetic flux density. This represents the direction vector of the static magnetic field in three-dimensional space; If the magnetic field is a one-dimensional alternating magnetic field, it satisfies the following relationship: ; in, This represents the angular frequency of the alternating magnetic field. This represents the direction vector of the alternating magnetic field in three-dimensional space.

[0033] It should be noted that a one-dimensional static magnetic field refers to a spatially uniform magnetic field whose magnetic field strength does not change with time and has only a fixed direction and amplitude. It is generated by independently energizing a set of coil pairs in a three-dimensional Helmholtz coil system, and its magnetic induction intensity vector... Maintain a constant orientation in space Amplitude Determined by the magnitude of the coil current, this magnetic field, through its magnetic torque, causes the magnetic moment direction of the micro / nano robot to be along... Arrangement is suitable for controlling the state of clusters requiring stable orientation or static state. A one-dimensional alternating magnetic field utilizes the continuous and smooth driving force provided by a sine wave, enabling the cluster to exhibit sustained dynamic behavior, with its instantaneous magnetic induction intensity... In direction vector Press up angular frequency Oscillation, in which, This indirectly represents the magnetic field frequency during the operation of the magnetic field, and is an important parameter affecting the current state of the cluster (generally speaking, whether the cluster is vortex-shaped, ribbon-shaped, or discrete, the cluster in different states has its corresponding frequency range. When the frequency is lower or higher than the frequency range of that state, the resulting cluster will not be a regular shape, or it may not be able to form a cluster of the corresponding shape at all). The amplitude remains [missing value]. angular frequency With frequency satisfy This magnetic field induces reciprocating oscillations or vibrations in micro / nano robots through periodically varying magnetic torque. It is suitable for dynamic control of swarms, such as excitation, dispersion, or prevention of sedimentation. In contrast, existing technologies employ pulsed one-dimensional alternating magnetic fields, which are discrete and instantaneous driving forces. These fields aim to elicit a step-like, discontinuous response from the swarm control described in this invention. (Direction vector) and All are unit vectors in three-dimensional space, and their components The orientation of the magnetic field in space is determined, and by adjusting this vector, the arrangement direction or vibration axis of the cluster can be controlled to align it with the axis of the simulation channel or the direction of motion required by the task. As an amplitude parameter of magnetic induction intensity, it directly determines the magnitude of the magnetic force acting on the micro-nano robot. Its value needs to be calculated comprehensively based on the robot's magnetization intensity, fluid viscosity, and required driving force, and is usually adjusted in the range of 0.5mT to 5mT.

[0034] Understandably, when a swarm needs to temporarily stop at a certain location to wait for drug loading or avoid obstacles, a one-dimensional static magnetic field can provide a stable directional torque, aligning the robot's long axis along a preset direction and maintaining spatial pose stability, preventing attitude inaccuracies caused by fluid disturbances. When the swarm needs to disperse from an aggregated state into individual robots to expand the search range, or needs to overcome static friction to initiate movement, a one-dimensional alternating magnetic field can disrupt the magnetic attraction balance between robots through periodic torque perturbations, inducing them to vibrate and separate. By clearly defining the expressions for two one-dimensional magnetic fields, the abstract magnetic field control requirements are transformed into specific electromagnetic parameter equations, enabling the control system to select the appropriate magnetic field mode according to the task stage and calculate accurate current waveforms. This achieves basic swarm control without relying on complex two-dimensional or three-dimensional magnetic fields, simplifying the control logic and hardware configuration, and reducing the system's computational load and energy consumption.

[0035] Preferably, in step S3, if the magnetic field is a two-dimensional magnetic field, the pose relationship between the magnetic field and the absolute coordinate system of the three-dimensional Helmholtz coil is first established, by confirming the rotation axis direction vector of the two-dimensional rotating field. The angle between the rotation field and the z-axis of the absolute coordinate system determines the direction and spatial coordinates of the two-dimensional rotation field. Wherein, when the origin of the rotation field coordinates is the origin of the absolute coordinate system, the rotation axis direction vector is defined. The angle between the absolute coordinate system and the positive z-axis is Define the direction vector of the rotation axis. The angle between the projection of the xoy plane in the absolute coordinate system and the positive x-axis direction is . And establish the coordinate space coordinates of the two-dimensional rotating field: ; ; ; If the magnetic field is a two-dimensional rotating field, the magnetic induction intensity vector of the two-dimensional rotating field... Satisfying the relation: ; ; in, This represents the magnetic field frequency of the two-dimensional rotating field. Indicates the magnitude of the magnetic field component along the x-axis. With the amplitude of the y-axis magnetic field component The ratio, This represents the movement speed control parameters of the cluster. Indicates the control parameters for the movement direction of the cluster. , and These represent the amplitudes of the magnetic field components along the x, y, and z axes, respectively.

[0036] It should be noted that a two-dimensional magnetic field refers to a time-varying magnetic field in which the magnetic induction intensity vector is constrained within a plane in three-dimensional space. It is generated by the coordinated excitation of two sets of coil pairs in a three-dimensional Helmholtz coil. Its magnetic field components remain zero or constantly biased in the third dimension. This magnetic field, through rotation or oscillation within the plane, causes micro-nano robot swarms to form collective behavioral patterns such as ribbons or vortices. The absolute coordinate system of the three-dimensional Helmholtz coil refers to a fixed Cartesian coordinate system established with the geometric center of the Helmholtz coil as the origin and the axes of the three sets of orthogonal coil pairs as the x, y, and z axes. This coordinate system is the sole reference for the magnetic field space, and all calculations and transformations of magnetic field parameters are based on this coordinate system. The rotation axis direction vector of the two-dimensional rotating field... This refers to the unit normal vector perpendicular to the plane of magnetic field rotation. Its spatial orientation determines the orientation of the plane containing the rotating field. This can be adjusted... This allows the plane of rotation to align with the curved plane of the simulated channel or the drug loading plane. (Included angle) yes The polar angle relative to the positive z-axis of the absolute coordinate system, with a range of values ​​of... This is used to describe the degree of inclination of the rotation axis in the vertical direction. When the rotation axis coincides with the z-axis, the rotation field is a rotation in the horizontal plane; when When the axis of rotation is horizontal, the rotational field is a rotation within a vertical plane. (Included angle) yes The azimuth angle projected onto the xoy plane and along the positive x-axis has a range of values. The coordinate axes are used to describe the orientation of the rotation axis in the horizontal plane, and together they constitute the spherical coordinate parameters that determine the spatial orientation of the rotation axis. , , These are the three orthogonal basis vectors of the local coordinate system of the rotating field, where With respect to the direction of rotation axis coincide, and Zhang Cheng's rotating plane, a local coordinate system, decouples the motion description of the rotating field from the absolute coordinate system, simplifying the mathematical expression of the rotating magnetic field. (Ratio) It is a rotating magnetic field in and The ratio of magnitudes in the direction, when The magnetic field vector traces a circular trajectory in the plane of rotation, forming a symmetrical vortex-like cluster; when By depicting elliptical trajectories, the cluster experiences uneven forces along its major and minor axes, which can form asymmetric banded or flattened vortex structures. The value range is typically from 0.5 to 4. Motion speed control parameters. The rotating magnetic field in the local coordinate system and - The pitch angle of the plane is adjusted by... The tilt of the magnetic field's rotation axis relative to the plane of cluster motion can be changed, thereby controlling the magnitude of the magnetic torque component in the forward direction. The larger the value, the greater the effective torque driving the cluster forward, and the faster the movement speed. In the experiment, values ​​ranging from 1 to 4 radians were used. Motion direction control parameters. It is a rotating magnetic field in - The rotational phase angle within the plane determines the initial direction of the magnetic field vector, which is adjusted in real time. It enables the instantaneous movement direction of the cluster to follow the tangent direction of the planned path, achieving precise tracking of curved paths.

[0037] It is understandable that by establishing the pose relationship between the two-dimensional rotation field and the absolute coordinate system, and defining two rotation angles... and This invention aims to determine the spatial orientation of the rotation axis, addressing the challenge of directional control of a rotating magnetic field in three-dimensional space. The actual motion of micro / nano robot swarms often occurs in non-horizontal, non-vertical inclined planes, such as at simulated blood vessel bifurcation points or in myocardial tissue gaps, where the rotation plane needs to match the channel curvature. Traditional methods fix the rotation axis on the z-axis, severely limiting the swarm's adaptability in three-dimensional environments. This invention addresses this by introducing... and With two degrees of freedom, the rotation plane can be arbitrarily oriented. The control system can calculate the best-matching rotation axis vector in real time based on the tangent direction of the waypoint and the normal direction of the channel provided by the path planning module. This generates a rotating magnetic field that is compatible with the geometric environment. Simultaneously, through... The parameters expand the circular rotation to an elliptical rotation, solving the problem of cluster shape adaptation under different channel widths: setting parameters in narrow channels... Extend the cluster along the channel axis to reduce the risk of lateral collisions; set up in open areas To improve drug loading stability, symmetrical vortices are formed. and The separate design enables independent control of both speed and direction of motion. The forward driving torque is controlled by adjusting the pitch angle of the magnetic field. By adjusting the rotation phase to control the motion vector angle, the two are decoupled, allowing the cluster to adjust synchronously without deceleration when cornering. Changing the direction of motion avoids turning instability or drug detachment caused by velocity-direction coupling. This claim separates the geometric parameters and dynamic control parameters of the rotating magnetic field, constructing a complete parameter system from spatial orientation and morphological shaping to motion control, significantly improving the flexibility and precision of magnetic field regulation.

[0038] Preferably, if the magnetic field is a two-dimensional oscillating field, the magnetic induction intensity vector of the two-dimensional oscillating field is... Satisfying the relation: ; ; Wherein, rotation matrix , and They respectively satisfy the following relations: ; ; ; in, The magnetic field frequency represents the frequency of a two-dimensional oscillating field. This represents the magnetic flux density vector in the absolute coordinate system of the three-dimensional Helmholtz coil after rotational transformation. This represents the movement speed control parameters of the cluster. Indicates the control parameters for the movement direction of the cluster. The orientation control parameter of the xoy plane, representing the motion coordinate system of the oscillation field, is... and These represent the amplitudes of the magnetic field components along the x-axis and y-axis, respectively.

[0039] It should be noted that a two-dimensional oscillating field refers to a time-varying magnetic field in which the magnetic induction intensity vector oscillates back and forth along a straight line in a plane in three-dimensional space. It is generated by the coordinated excitation of two sets of coils in a three-dimensional Helmholtz coil. The x-axis component changes according to a sinusoidal law, the y-axis component remains constant, and the z-axis component is zero. This magnetic field drives a cluster of micro-nano robots to form a chain-like or strip-like aggregate structure along the oscillation direction through periodic linear magnetic force, which is suitable for rapid transportation in straight channels. This represents a fundamental formula for a basic oscillating field, the oscillating magnetic field required to form a banded cluster. However, using only the basic formula is insufficient to enable the cluster to move; therefore, further processing is necessary. Transformation , This is the result of mapping the magnetic field vector in the local coordinate system to the absolute coordinate system of the three-dimensional Helmholtz coil through three rotation transformations. It can satisfy the function of individual cluster rotation and translation movement of strip-shaped clusters in an oscillating magnetic field. Rotation matrix , and There are three 3×3 orthogonal matrices, among which To achieve the rotation of the oscillation field coordinate system around the z-axis, which is used to adjust the initial deflection angle of the y-axis in the oscillation plane relative to the x-axis of the absolute coordinate system, that is, the angle between the major axis of the strip cluster and the x-axis of the absolute coordinate system; Rotation around the y-axis is achieved to adjust the pitch angle of the oscillation plane relative to the horizontal plane, thereby controlling the projection of the cluster's motion velocity in the absolute coordinate system; A secondary rotation around the z-axis is implemented to adjust the angle between the actual forward direction of the cluster and the horizontal axis of the absolute coordinate system. Parameters and Essentially, they are the same physical quantity, both representing the reference angle of the cluster's movement direction, but in the formula... Used with Subtraction eliminates the coupling caused by the initial coordinate system deflection, ensuring the final direction of motion. Control independence. Towards control parameters. It directly determines the orientation of the xoy plane of the oscillation field coordinate system in the absolute coordinate system, and its value ranges from 0 to... By adjusting This allows the long axis of the ribbon-like cluster to be aligned with the channel axis, preventing the cluster from colliding with the wall due to directional deviation within the channel.

[0040] It is understandable that, through the cascaded transformation of three rotation matrices, firstly by Set the initial orientation of the oscillation plane so that the long axis of the strip is aligned with the channel inlet direction; then... Adjusting the pitch angle of the oscillation plane ensures the cluster remains parallel to the pipe wall in the inclined channel, preventing vertical displacement; ultimately, by The system corrects the forward direction angle of the cluster in real time, ensuring that the ribbon structure always follows the tangential direction of the planned path during its movement. This transformation system decouples the motion direction, velocity, and orientation of the oscillating field into three independent control parameters, enabling precise control of the morphology, orientation, and velocity of the ribbon cluster in a complex three-dimensional channel. This overcomes the limitation of traditional oscillating magnetic fields, which can only drive along a fixed direction and cannot adapt to changes in spatial geometry.

[0041] In step S3, if the magnetic field is a three-dimensional dynamic segmented field, the magnetic induction intensity vector of the three-dimensional dynamic segmented field... Satisfying the relation: ; in, The magnetic field frequency represents the three-dimensional dynamic segmented field. Indicates time A mathematical function that rounds up. , and These represent the magnetic induction intensity vectors of the three-dimensional dynamic segmented field. Components in the x, y, and z axes.

[0042] Preferably, in step S3, if the magnetic field is a three-dimensional dynamic segmented field, the magnetic induction intensity vector of the three-dimensional dynamic segmented field... Satisfying the relation: ; in, The magnetic field frequency represents the three-dimensional dynamic segmented field. Indicates time A mathematical function that rounds up. , and These represent the magnetic induction intensity vectors of the three-dimensional dynamic segmented field. Components in the x, y, and z axes.

[0043] It should be noted that a three-dimensional dynamic segmented field refers to a magnetic field that exhibits a discrete segmented and jump-like pattern over time in three-dimensional space. Its core feature is that the magnetic field vector periodically switches between discrete direction sets with 120° intervals, while maintaining sinusoidal amplitude modulation in each direction. The function is a floor function that returns a value not less than 100%. The smallest integer, expression The coefficient 5 in the equation indicates that 5 rounding operations are triggered per second, making... The value increases by 1 every 0.2 seconds, and the corresponding 120° product term causes the magnetic field direction vector to rotate counterclockwise by 120° every 0.2 seconds, forming a direction jump period. Magnetic field frequency It is the sinusoidal modulation frequency of the magnetic field strength in each direction, typically ranging from 20Hz to 50Hz. This high-frequency oscillation generates rapid torque changes in the robot's magnetic moment, enabling it to maintain dynamic stability during directional transitions and preventing cluster disintegration caused by sudden directional changes. Components and In and This constitutes a direction selection mechanism, and the two are in The magnetic field direction remains constant within each 0.2-second interval, and the combination of these parameters determines the direction of the magnetic field in a plane. The factor causes the magnetic field to oscillate sinusoidally in that direction, generating local dynamic drive; component In Maintaining a 90° phase difference with the x and y components, the magnetic field vector traces a time evolution trajectory similar to a Lissajous figure in three-dimensional space, providing multi-directional excitation for the cluster in three-dimensional space. , , The amplitude parameters control the magnetic field strength along the three axes, and their proportional relationship determines the spatial orientation distribution characteristics of the segmented field. These parameters can be set... To maintain a constant amplitude of the synthetic magnetic field when switching between directions, and to prevent jumps in robot speed caused by amplitude fluctuations.

[0044] Understandably, the purpose of setting up a three-dimensional dynamic segmented field is to use a magnetic field to divide the particle chains in the cluster. The repulsive force between the magnetic chains of particles within the cluster is used to transform the cluster from its original aggregated state to a discrete state, thus forming a discrete cluster. The purpose of the three-dimensional dynamic segmented field is to release the captured drug and achieve targeted drug delivery when the cluster acquires the drug and transports it to its destination.

[0045] Preferably, based on the real-time acquired motion images, using a deep learning model to identify the real-time pose information of the micro-nano robot cluster and the channel boundary information of the simulated environment includes: A cluster modal feature dataset is constructed based on the real-time acquired motion images. The cluster modal feature dataset includes the modal feature information of the cluster and the channel information of the simulation environment. The cluster modal feature dataset is trained using a target detection neural network based on the YOLOv5 architecture to identify and track the micro-nano robot cluster and obtain its real-time pose information. The YOLOv5 network structure includes a Backbone module, a Neck module, and a Head module. The YOLOv5 network extracts image features through the Backbone module, performs multi-scale feature fusion through the Neck module, and outputs detection results including the pose information of the micro-nano robot cluster through the Head module.

[0046] It should be noted that the cluster modality feature dataset refers to a structured data set containing visual features of clusters in different morphologies, formed by manually annotating sequences of motion images of micro- and nano-robots acquired through cameras. The cluster modality feature dataset covers three typical modalities: strip-shaped, vortex-shaped, and dispersed. It includes image samples of each modality, and the annotations include the coordinates of the cluster's bounding box, centroid position, principal axis orientation angle, and the wall contour point set of the simulated environment channel. This dataset is used to train deep learning models to enable them to distinguish different modalities and accurately locate them. The YOLOv5 architecture is a single-stage object detection convolutional neural network. Its backbone module uses a CSPDarknet53 or ResNet structure, dividing the feature map into two parts through cross-stage partial connections, processing them separately, and then concatenating them. This preserves rich semantic information while reducing computational cost. This module is responsible for extracting multi-level features from the input image, ranging from low-level edges to high-level semantics. The Neck module employs a cascaded structure of FPN (Feature Pyramid Network) and PAN (Path Aggregation Network), achieving lateral connectivity and vertical fusion of multi-scale feature maps through upsampling and downsampling. This enables the model to simultaneously detect cluster targets of different sizes in an image, exhibiting robustness to scale scaling caused by depth variations. The Head module consists of three parallel detection heads, each corresponding to a different receptive field. Each detection head outputs a tensor containing bounding box coordinates, target confidence, and class probability. The Head module ultimately selects the optimal detection result through non-maximum suppression, completing the end-to-end mapping from image to cluster pose.

[0047] In this embodiment, the structure of YOLOv5 is as follows: Figure 2 As shown, the system utilizes feature maps extracted from the backbone network to detect the target bounding box, confidence, and class on each feature map. Within these three main parts are three sub-modules: Conv, C3, Concat, Upsample, SPFF, and Detect. The Conv module mainly consists of convolutional layers, batch normalization (BN) layers, and activation functions, used to extract local features and perform feature mapping. The C3 module reduces computation while maintaining gradient flow and includes some residual connections. The Concat module concatenates feature maps from different sources along the channel dimension to achieve multi-scale information fusion. The Upsample module upsamples low-resolution feature maps for fusion with high-resolution feature maps, enhancing small target detection capabilities. The SPFF module obtains more comprehensive spatial information through weighted fusion of global and local features. The Detect module is the final detection layer, outputting the target's bounding box position, target confidence, and class probability for each grid point.

[0048] Preferably, trajectory planning using a path planning algorithm based on the real-time pose information and the channel boundary information of the simulated environment includes: Adopting traditional The algorithm is an optimized hybrid path planning algorithm that satisfies the following relation: ; ; in, This indicates the nodes passed from the starting point. The total cost evaluation function to the destination, Indicates the distance from the starting point to the node. The actual cost, Indicates from node Optimized estimated cost to reach the destination Represents the flow field evaluation function. This represents the distance evaluation function. Represents the curvature evaluation function. This represents the weighting coefficients of the comprehensive penalty function. This represents the weighting coefficients of the distance evaluation function. Represents the weighting coefficients of the curvature evaluation function; and Satisfying the relation: ; ; ; ; ; ; ; in, Represents a node Distance to the nearest obstacle Represents the set of obstacle nodes. Represents the set of walkable nodes. This represents the maximum distance from all passable nodes to the obstacle. Represents the distance transformation graph at the nodes Laplace response at the location, This represents the negative part of the Laplace response. This represents the maximum value of the negative part of the Laplace response for all traversable nodes. It is a very small constant set to avoid division by zero errors.

[0049] It should be noted that hybrid path planning algorithms refer to algorithms that combine traditional path planning with hybrid path planning. Based on the algorithm framework, the heuristic function is reconstructed. An intelligent search algorithm incorporating various microscale physical constraints, with its search space being a discretized grid map, where each grid node... Including coordinates Value and parent node pointer. Actual cost. Using Euclidean distance calculation, the path length from the starting point to the current node increases by one grid cell. Accumulate the current grid side length (e.g., 1 micrometer) to ensure the shortest total path length. Optimize the estimated cost. It consists of three weighted terms, among which the flow field evaluation function The flow cost of each node is calculated by reverse Dijkstra's algorithm from the target point. Initially, the target point is set... The distance evaluation function updates the value of a node based on its eight neighboring nodes using Euclidean distance, iterating to all reachable nodes. A larger value indicates that the node is further away from the target flow direction, guiding the search towards the target. Pre-calculated using a distance transformation algorithm, for each free grid Traverse all obstacle grids Calculate Euclidean distance And take the minimum value Seeking all again The maximum value ,final Forming a range The potential field, where the function takes a value close to 1 in the near-wall region, penalizes nodes close to the wall, forcing the planned path away from the wall. Curvature evaluation function Through the Calculation by applying discrete Laplace operator Extract the negative part And normalized, the function is affected by the bends in the channel. Drastic curvature changes result in high values, penalizing sharp-turn nodes and prompting path selection to follow a smooth external tangent arc. Weighting coefficients. , , All are positive real numbers. Adjust the overall penalty intensity. Adjusting the wall-avoidance weight, Adjust the smoothing weights; the values ​​are calibrated through offline experiments. The penalty term significantly affects the search between 1.0 and 5.0. , Balance wall avoidance and smoothing effects between 0.2 and 0.5 to avoid excessive penalties that lead to excessively long paths.

[0050] Understandably, the heuristic function of traditional A Typically, Manhattan or Euclidean distances are used, considering only the geometric shortest path. This neglects physical factors at the microscale, such as velocity decay due to fluid resistance, additional resistance and boundary layer effects near walls, and swarm instability due to inertia during sharp turns. Therefore, while the generated path may be geometrically optimal, it is not kinetically feasible. This invention proposes… First, flow field assessment is introduced. First, the global target direction is used as a strong guiding factor to avoid the search getting stuck in local dead zones, especially in branch channels where priority is given to selecting tributaries leading to the main target. Second, distance assessment. A static potential field was constructed to allow the path to be automatically maintained. When the value exceeds a safety threshold (e.g., 30 micrometers), this mechanism simulates the actual need for robots to maintain a safety margin with the wall surface due to Brownian motion and fluid disturbances at the microscale, effectively avoiding robot adhesion or drug scratching losses caused by wall-attaching motion. Finally, curvature assessment. Detection by Laplace operator Local curvature changes impose additional costs at channel corners, guiding... The search employs an externally tangent circular arc path with continuous curvature, conforming to the turning dynamics constraints of the magnetically controlled cluster and preventing cluster disintegration or drug ejection due to abrupt changes in magnetic field direction during right-angle turns. Three penalties are applied through weighting. , , Linear combination, making During the search process, the optimality of path length, safety margin, and smoothness are comprehensively considered. The final generated path is short, far from the wall, and has continuous curvature, perfectly matching the kinematic characteristics of micro-nano robots.

[0051] A motion control system for a magnetically controlled micro / nano robot to deliver drugs in a simulated environment includes an ultra-depth-of-field optical microscope, a camera, a three-axis microscopic observation platform, a control terminal computer, a control board, a power amplifier, and a three-dimensional Helmholtz coil. The ultra-depth-of-field optical microscope is optically connected to the camera to observe the motion state of the micro-nano robot cluster in a simulated environment and transmit optical images to the camera; The camera is mechanically connected to the three-axis microscopic observation platform and is used to record image information observed by the ultra-depth-of-field optical microscope. The three-axis microscopic observation platform is equipped with a cluster simulation environment channel, which can be moved at the micrometer level by adjusting the knob to adjust the pose of the cluster simulation environment channel in the camera's field of view. The camera is connected to the control terminal computer and transmits the captured real-time video data stream to the control terminal computer. The control terminal computer is connected to the control board via USB. The control terminal computer is used to identify the real-time pose information and simulated environment channel boundary information of the micro-nano robot cluster based on the real-time video data stream using the deep learning YOLOv5 model, and uses a path planning algorithm to perform trajectory planning based on the real-time pose information and simulated environment channel boundary information to generate digital control signals. The control board is connected to the power amplifier via a BNC to DuPont wire, which is used to convert the digital control signals generated by the control terminal computer into analog circuit signals. The power amplifier is connected to the three-dimensional Helmholtz coil via a BNC connector to a banana plug, and is used to amplify the analog circuit signal to drive the three-dimensional Helmholtz coil. The three-dimensional Helmholtz coil is used to generate a corresponding magnetic field based on the amplified analog circuit signal, and the movement state of the micro-nano robot cluster is controlled by adjusting the magnetic field parameters.

[0052] In the data acquisition phase, a super-depth-of-field optical microscope observes the cluster motion within the simulated channel at 60x magnification. A camera acquires 2600×2160 resolution image data at a frame rate of 70.3fps, transmitting it in real-time to the control terminal computer via USB 3.0. The three-axis platform automatically maintains the optimal observation position according to a preset program. In the data processing phase, the control terminal computer's GPU processes the video stream in parallel. The YOLOv5 model, based on the PyTorch framework, performs real-time analysis of each frame, identifying the cluster's current position, motion state, and channel boundary information. The path planning module calculates the optimal motion trajectory based on the identification results and generates corresponding magnetic field control parameters. , , Used to control the cluster configuration. and The system controls the direction and speed of motion separately. During the control execution phase, the control board receives control commands containing these parameters via a USB interface, converts the digital signals into corresponding analog voltage signals, and transmits them to a power amplifier via a BNC-to-DuPont wire. The amplifier amplifies the signal into a drive current at a set gain, which drives a three-dimensional Helmholtz coil to generate the target magnetic field through a BNC-to-banana connector. The entire system employs a double-buffering mechanism to ensure real-time performance; the next frame is acquired while the current frame is being processed. Control parameters are linearly interpolated for smooth transitions, and hardware synchronization signals ensure timing consistency between components.

[0053] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0054] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for motion control of a drug-loaded magnetic micro-nano robot in a simulated environment, characterized in that, The method comprises the following steps: S1: Real-time acquisition of motion images of a magnetically controlled silicon dioxide-based micro-nano robot cluster in a simulation environment; S2: According to a pre-set transportation task, the current state of the micro-nano robot cluster is judged, wherein the current state includes a ribbon-shaped cluster state when no drug is obtained and a vortex-shaped cluster state when the drug is obtained; S3: Based on the current state of the micro-nano robot cluster, a dynamic model of the micro-nano robot cluster, a magnetic field generated by a three-dimensional Helmholtz coil, a fluid resistance field of the simulation environment and a boundary condition of the simulation environment channel is established, and a power amplifier matched with the three-dimensional Helmholtz coil is associated to maintain the magnetic field; S4: Based on the real-time acquisition of the motion images, a deep learning model is used to identify real-time pose information of the micro-nano robot cluster and boundary information of the simulation environment channel, and a path planning algorithm is used to plan a trajectory according to the real-time pose information and the boundary information of the simulation environment channel to generate a control signal; S5: The control signal is converted into an analog signal through a control board card, amplified through the power amplifier and used to drive the three-dimensional Helmholtz coil to adjust the magnetic field and control the motion of the micro-nano robot cluster.

2. The motion control method of the drug-loaded magnetic micro-nano robot in a simulated environment according to claim 1, wherein, In step S3, if the magnetic field is a one-dimensional static magnetic field, the relationship is satisfied: ; wherein denotes the amplitude of the magnetic induction, denotes the direction vector of the static magnetic field in three-dimensional space; If the magnetic field is a one-dimensional alternating magnetic field, the following relationship is satisfied: ; wherein denotes the angular frequency of the alternating magnetic field, denotes the direction vector of the alternating magnetic field in three-dimensional space.

3. The motion control method of the drug-loaded magnetic micro-nano robot in a simulated environment according to claim 1, wherein, In step S3, if the magnetic field is a two-dimensional magnetic field, first establish the pose relationship between the magnetic field and the three-dimensional Helmholtz coil absolute coordinate system, by confirming the rotation axis direction vector of the two-dimensional rotating field The angle between the absolute coordinate system z axis, determine the direction and spatial coordinates of the two-dimensional rotating field; wherein, in the case that the origin of the rotating field coordinates is the origin of the absolute coordinate system, the rotating axis direction vector is defined as the angle between the positive direction of the absolute coordinate system z-axis and the rotating axis direction vector is , the rotating axis direction vector is defined as the angle between the projection of the rotating axis direction vector on the absolute coordinate system xoy plane and the positive direction of the x-axis is and the spatial coordinates of the coordinate axis of the two-dimensional rotating field are established as ； ； ； If the magnetic field is a two-dimensional rotating field, the magnetic induction intensity vector of the two-dimensional rotating field satisfies the relationship: ； ； wherein denotes the magnetic field frequency of the two-dimensional rotating field, denotes the ratio of the amplitude of the x-axis magnetic field component to the amplitude of the y-axis magnetic field component , denotes a motion speed control parameter of the cluster, denotes a motion direction control parameter of the cluster, , and denote the amplitudes of the x-axis, y-axis and z-axis magnetic field components, respectively.

4. The motion control method of the drug-loaded magnetic micro-nano robot in a simulated environment according to claim 1, wherein, If the magnetic field is a two-dimensional oscillating field, the magnetic induction vector B of the two-dimensional oscillating field satisfies the relationship: ； ； where the rotation matrix , and respectively satisfy the relations: ； ； ； wherein, denotes the magnetic field frequency of the two-dimensional oscillating field, denotes the magnetic induction intensity vector in the three-dimensional Helmholtz coil absolute coordinate system after the rotation transformation, denotes the motion speed control parameter of the cluster, denotes the motion direction control parameter of the cluster, denotes the orientation control parameter of the motion coordinate system xoy plane of the oscillating field, and denote the amplitude of the x-axis and y-axis magnetic field components, respectively.

5. The method of claim 1, wherein the magnetic microrobot is a drug-loaded microrobot in a simulated environment. In step S3, if the magnetic field is a three-dimensional dynamic segmented field, the magnetic induction intensity vector of the three-dimensional dynamic segmented field is calculated according to the following formula: satisfies the relationship: ； wherein denotes the magnetic field frequency of the three-dimensional dynamic piecewise field, denotes a mathematical function that rounds up to the next whole number for time t, , and denote the components of the magnetic induction vector of the three-dimensional dynamic piecewise field in the x-, y- and z-axis directions, respectively.

6. The method of claim 1, wherein the magnetic microrobot is a drug-loaded microrobot in a simulated environment. Based on the real-time acquisition of the motion images, the real-time pose information of the micro-nano robot cluster and the boundary information of the simulation environment channel are identified using a deep learning model, which comprises: Based on the real-time acquisition of the motion images, a cluster modal feature dataset is constructed, which includes modal feature information of the cluster and simulation environment channel information; A target detection neural network based on a YOLOv5 architecture is used to train the cluster modal feature dataset to identify and track the micro-nano robot cluster and obtain its real-time pose information; The YOLOv5 network structure includes a Backbone module, a Neck module and a Head module, the YOLOv5 network extracts image features through the Backbone module, performs multi-scale feature fusion through the Neck module, and outputs detection results including the pose information of the micro-nano robot cluster through the Head module.

7. The method of claim 1, wherein the magnetic microrobot is a drug-loaded microrobot in a simulated environment. The path planning algorithm is used to plan a trajectory according to the real-time pose information and the boundary information of the simulation environment channel, which comprises: A hybrid path planning algorithm based on traditional algorithms is adopted, and the hybrid path planning algorithm satisfies a relationship expression: ； ； wherein, represents a total cost evaluation function from a start point through a node to an end point, represents an actual cost from the start point to the node , represents an optimized estimated cost from the node to the end point, represents a flow field evaluation function, represents a distance evaluation function, represents a curvature evaluation function, represents a weight coefficient of a comprehensive penalty function, represents a weight coefficient of the distance evaluation function, represents a weight coefficient of the curvature evaluation function; and satisfies the relationship: ； ； ； ； ； ； ； wherein, denotes a node denotes the distance to the nearest obstacle, denotes a set of obstacle nodes, denotes a set of traversable nodes, denotes the maximum of all traversable node to obstacle distances, denotes the Laplacian response of the distance transform map at node denotes the negative part of the Laplacian response, denotes the negative part of the Laplacian response, denotes the maximum of all traversable node Laplacian response negative parts, is a small constant set to avoid division by zero errors.

8. A motion control system of a magnetic micro-nano robot in a simulated drug-loaded environment, characterized in that, It comprises a super-depth optical microscope, a camera, a three-axis microscopic observation platform, a control terminal computer, a control board card, a power amplifier and a three-dimensional Helmholtz coil; The super-depth optical microscope is optically connected with the camera and is used to observe the motion state of the micro-nano robot cluster in the simulation environment and transmit optical images to the camera; The camera is mechanically connected with the three-axis microscopic observation platform and is used to record image information observed by the super-depth optical microscope, wherein the three-axis microscopic observation platform is equipped with a cluster simulation environment channel and can realize micron-level movement through an adjusting knob to adjust the pose of the cluster simulation environment channel in the field of view of the camera; The camera is data-connected with the control terminal computer and transmits real-time video data shot by the camera to the control terminal computer; The control terminal computer is connected with the control board card through a USB, and the control terminal computer is used for identifying real-time pose information and simulated environment channel boundary information of the micro-nano robot cluster based on real-time video data stream through a deep learning YOLOv5 model, and generating a digital control signal according to the real-time pose information and the simulated environment channel boundary information by using a path planning algorithm; The control board card is connected with the power amplifier through a BNC to Dupont line, and is used for converting the digital control signal generated by the control terminal computer into an analog circuit signal; The power amplifier is connected with the three-dimensional Helmholtz coil through a BNC joint to a banana head, and is used for driving the three-dimensional Helmholtz coil after amplifying the analog circuit signal; The three-dimensional Helmholtz coil is used for generating a corresponding magnetic field according to the amplified analog circuit signal, and controlling the motion state of the micro-nano robot cluster by adjusting the magnetic field parameters.

Images