Magnetic micro-robot cluster task adaptive formation control method and system thereof

Abstract

The application discloses a kind of magnetic micro-robot cluster task adaptive formation control method and system, belong to micro-nano robot cluster control and tumor immunotherapy cross technical field.According to real-time imaging feedback, the current vascular environment characteristics are identified, and the current task phase is judged;Based on the parameter limitation of three-dimensional helmholtz coil, a composite magnetic field sequence is generated;A prediction model is established based on the composite magnetic field sequence;The formation success rate is evaluated;When the evaluated formation success rate is less than 90%, online parameter optimization is performed;The task adaptive formation control of magnetic micro-robot cluster is realized.The application is used to solve the problems of single control dimension, lack of formation capability, no quantitative performance index and mismatch of magnetic field system in the prior art, to realize the adaptive switching and precise control of three task formation modes of magnetic micro-robot in different stages of blood circulation system.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of micro-nano robot swarm control and tumor immunotherapy, specifically relating to a method and system for adaptive formation control of magnetic microrobot swarm tasks. Background Technology

[0002] CAR-T cell therapy has achieved significant results in the treatment of hematological malignancies, but its efficacy in treating solid tumors remains limited by the efficiency of targeted cell delivery and the infiltration capacity of the tumor microenvironment. The development of microrobot technology has provided a new approach to solving this bottleneck: combining CAR-T cells with magnetic nanoparticles to form "cell robots" that can achieve active targeted delivery driven by an external magnetic field.

[0003] Existing technologies, such as PCT / KR2023 / 021451, disclose a magnetically driven anti-cancer robot that binds magnetic nanoparticles to CAR-T cells via biotin-streptavidin to achieve monomeric magnetic actuation. However, this technology has the following limitations:

[0004] 1. Limited control dimension: It can only achieve directional movement of a single cell and cannot achieve coordinated control of multiple cells in a cluster;

[0005] 2. Lack of formation capability: It cannot adaptively switch formation patterns in complex vascular environments (such as rapid transport in large vessels, narrow passage through capillaries, and infiltration of tumor sites);

[0006] 3. Lack of quantitative performance indicators: The lack of a quantitative assessment method for formation success rate makes it difficult to guarantee the reliability of treatment effects;

[0007] 4. Magnetic field system mismatch: Most existing magnetic control systems are single-mode and cannot generate composite magnetic field sequences to drive multimodal cluster behavior. Summary of the Invention

[0008] This invention provides a method and system for adaptive formation control of magnetic microrobot cluster tasks, which solves the problems of single control dimension, lack of formation capability, lack of quantitative performance indicators and magnetic field system mismatch in the prior art. It realizes adaptive switching and precise control of three task formation modes of magnetic microrobots in different stages of the blood circulation system.

[0009] This invention is achieved through the following technical solution:

[0010] A task-adaptive formation control method for a magnetic microrobot swarm, the control method comprising the following steps:

[0011] Step 1: Based on real-time imaging feedback, identify the current vascular environment characteristics and determine the current task stage;

[0012] Step 2: Based on the task stage determined in Step 1, generate a composite magnetic field sequence based on the parameter constraints of the three-dimensional Helmholtz coil;

[0013] Step 3: Establish a prediction model based on the composite magnetic field sequence from Step 2;

[0014] Step 4: Based on the prediction model established in Step 3, evaluate the formation success rate;

[0015] Step 5: When the formation success rate assessed in Step 4 is below 90%, perform online parameter optimization;

[0016] Step 6: Implement task-adaptive formation control for the magnetic microrobot cluster.

[0017] Furthermore, the specific characteristics of the current vascular environment in step 1 are: vascular diameter, bifurcation status, and blood flow velocity.

[0018] The specific steps of step 1, which determine the current task stage, are as follows: if the task is in the major blood vessel stage, then dispersed formation is used.

[0019] If the vessel is in the capillary stage, a strip formation should be used;

[0020] If the tumor is in the infiltration stage, a strip formation should be used;

[0021] Step 2 specifically involves a dispersed formation magnetic field mode:

[0022] X-axis: Rotating magnetic field, frequency 8-12Hz, amplitude 80-100Gs;

[0023] Y-axis: Orthogonal to the X-axis, with a phase difference of 90°;

[0024] Z-axis: Low-frequency oscillation, frequency 1-3Hz, amplitude 20-40Gs;

[0025] Strip formation magnetic field pattern:

[0026] X-axis: Gradient magnetic field, with a gradient of 0.5-1 Gs / mm along the axial direction of the blood vessel;

[0027] Y-axis: Constraint field, preventing cells from deviating from the axis;

[0028] Z-axis: Closed or weakly constrained;

[0029] Strip formation magnetic field pattern:

[0030] X-axis: Oscillation field along the tangential direction of the blood vessel, frequency 5-10Hz;

[0031] Y-axis: The pushing field perpendicular to the blood vessel wall, with a gradient of 0.2-0.5 Gs / mm;

[0032] Z-axis: Low-frequency modulated field, enhancing rolling behavior.

[0033] Furthermore, step 3 includes the following steps:

[0034] Step 3.1: Establish a dynamic model of the magnetic microrobot under a magnetic field;

[0035] Step 3.2: Optimize the target using distributed model predictive control;

[0036] Step 3.3: Adjust the magnetic field frequency in real time using a PID controller to ensure that the formation status meets the target.

[0037] Furthermore, the dynamic model in step 3.1 is as follows:

[0038]

[0039] in For the first The location of each cell It is a composite magnetic field distribution. For blood flow velocity, For random perturbations;

[0040] The specific optimization of the objective in step 3.2 is as follows:

[0041]

[0042] st magnetic field constraint: ;

[0043] Step 3.3, which involves adjusting the magnetic field frequency in real time using a PID controller, specifically involves:

[0044]

[0045]

[0046]

[0047] in For formation state error, For the target formation state, The current formation status, , , These are the parameters for the PID controller.

[0048] Furthermore, the formation state is defined as follows:

[0049] Dispersed formation status : , For robots i andj The distance;

[0050] Strip formation status : , For robots i The azimuth angle;

[0051] Strip formation status : , For robots i The relative position.

[0052] Furthermore, step 4 specifically includes the following steps:

[0053] Step 4.1: Perform image enhancement on the camera feedback image;

[0054] Step 4.2: Perform cell segmentation on the enhanced image from Step 4.1;

[0055] Step 4.3: Locate the cells segmented in Step 4.2;

[0056] Step 4.4: After acquiring cell cluster images through the real-time imaging module, key feature parameters are extracted through image processing to evaluate the formation success rate.

[0057] Furthermore, step 4.4, extracting key feature parameters, specifically involves:

[0058] Cell location set:

[0059]

[0060] Where N is the total number of identifiable magnetic microrobots within the current field of view;

[0061] Cell distance set:

[0062]

[0063] The ratio of the main axis direction of the cluster to its major axis is specifically as follows:

[0064] First, calculate the covariance matrix of the coordinates of all cell centers:

[0065]

[0066] in As the cluster center;

[0067] Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues. and its corresponding eigenvectors ; Maximum eigenvalue Corresponding feature vector This is the main axis direction of the cluster;

[0068] Define the aspect ratio of the cluster :

[0069]

[0070] Calculate the angle between the principal axis and the vascular axis. :

[0071] .

[0072] Furthermore, the success rates of creating the three types of formations are defined as follows:

[0073] In dispersed formations, the proportion of intercellular spacing within the desired range;

[0074]

[0075] in The average radius of the magnetic microrobot; II( () is an indicator function: it takes the value 1 if the condition is true, and 0 otherwise; The total number of cell pairs;

[0076] In a strip formation, the proportion of cells with a length-to-diameter ratio ≥ 5:1 and an axial deviation ≤ 30°;

[0077]

[0078] In a strip formation, the proportion of cells that are in contact with the blood vessel wall and whose rolling speeds are matched;

[0079] For banded formations, cells are required to adhere closely to the vessel wall and roll along the wall surface; firstly, the vessel wall boundary needs to be segmented from the image; let the vessel wall region be... The outline region of cell i is Define the contact area ratio The proportion of the area where the cell outline overlaps with the blood vessel wall to the total projected area of ​​the cell:

[0080]

[0081] in Indicates area; the larger the contact area ratio, the more tightly the cell adheres to the blood vessel wall;

[0082] Furthermore, the banded formation requires cells to roll along the vessel wall at a speed that matches the local blood flow velocity; the tangential velocity of the cells along the vessel wall can be calculated using inter-frame difference or optical flow methods. Simultaneously, the local blood flow velocity measured by blood flow tracer particles The rolling speed matching error is defined as:

[0083]

[0084] in Minimum speed threshold;

[0085] Qualified cells in a banded formation must simultaneously meet the following requirements:

[0086] Contact area ratio:

[0087] Speed ​​matching error:

[0088] .

[0089] Furthermore, step 5, which employs reinforcement learning for parameter optimization, specifically involves:

[0090] The state space S consists of key variables that reflect the current formation environment and control effectiveness, including:

[0091] Formation type identifier : These represent the current actual forms of the dispersed, strip-shaped, and band-shaped formations, respectively;

[0092] Current formation success rate

[0093] Blood vessel diameter Obtained by image measurement;

[0094] Blood flow velocity , estimated by particle image velocimetry or a preset model;

[0095] Cell distribution entropy Used to describe the uniformity of cell spatial distribution, defined as

[0096]

[0097] The imaging region is divided into K A sub-region of equal volume To fall into the first k The proportion of cells in each sub-region;

[0098] The state vector is denoted as ;

[0099] The action space A is a combination vector of adjustable magnetic field parameters; due to the hardware limitations of the three-dimensional Helmholtz coil, each parameter has a physical boundary; the action is a continuous quantity or a discrete quantity after discretization. This invention adopts a continuous action space and outputs it through a Gaussian strategy.

[0100] The reward function R is designed to guide the agent to choose actions that improve the formation success rate, while avoiding violations of magnetic field constraints or dangerous behaviors.

[0101] Single-step reward is defined as:

[0102]

[0103] in Bonus for success rate:

[0104]

[0105] in To enforce penalties for violations:

[0106]

[0107] in Rewards for smoothness of motion:

[0108]

[0109] SAC is used as the reinforcement learning engine; a large number of state-action-reward samples are generated using a simulation environment to train the value network or policy network.

[0110] Decision-making during the online operation phase:

[0111] Obtain the current status from the real-time imaging module. ;

[0112] If the formation success rate is below 90%, the state is input into the policy network, and the action is output. ;

[0113] The action is mapped to a specific magnetic field control signal, which is generated and applied by an arbitrary waveform generator;

[0114] Waiting for the next control cycle, observing the new state. and rewards ;

[0115] Storage experience ( , , , Add it to the experience pool and update the network parameters regularly.

[0116] A task-adaptive formation control system for a magnetic microrobot swarm, the control system employing the task-adaptive formation control method for a magnetic microrobot swarm as described above, the control system comprising:

[0117] Task recognition module: Based on real-time imaging feedback, it identifies the current vascular environment characteristics and determines the current task stage;

[0118] Composite magnetic field sequence generation module: Based on the task stage determined by the task identification module, a composite magnetic field sequence is generated based on the parameter constraints of a three-dimensional Helmholtz coil.

[0119] Prediction model building module: Establishes a prediction model based on the composite magnetic field sequence generated by the composite magnetic field sequence generation module;

[0120] Formation success rate assessment module: Based on the prediction model built by the prediction model construction module, the formation success rate is assessed;

[0121] Parameter optimization module: When the formation success rate assessed by the formation success rate assessment module is lower than 90%, online parameter optimization is performed; to achieve task-adaptive formation control of the magnetic microrobot cluster.

[0122] Furthermore, to verify the "CAR-T cell robot swarm task adaptive formation control method" proposed in this invention patent, the following simulation experiments were conducted:

[0123] 1. Independent Formation Formation Simulation Experiment

[0124] The aim is to verify whether the designed magnetic field pattern can enable cell clusters to converge to the expected formation geometry and meet quantitative requirements under a single task phase.

[0125] The specific simulation content involves simulating three typical operating conditions: dispersed formation, strip formation, and belt formation, with each condition running independently for 2000 time steps.

[0126] The driving effect of the composite magnetic field is equivalently replaced by setting up virtual fields corresponding to the three magnetic field modes (such as intercellular repulsion in dispersed formations, Y-axis constraint in strip formations, and blood vessel wall attraction in ribbon formations).

[0127] The final output is the spatial distribution map of the three formations, and the corresponding quantitative evaluation indicators are calculated, such as... Figure 4 As shown in Figures 5 and 6.

[0128] 2. Formation Adaptive Switching Simulation Experiment

[0129] The aim is to verify that the system can adaptively switch formation patterns when the vascular environment changes (e.g., from large blood vessels to capillaries to tumor sites), and that the switching process is smooth and the formation indicators converge in the expected direction, thereby proving the effectiveness of the task adaptive control strategy.

[0130] The specific simulation content is to simulate the cell cluster going through three stages in the same simulation process: "dispersion → strip → band". Each stage lasts for 1,500 steps, with a total of 4,500 steps.

[0131] The field function automatically switches as the stage changes, simulating the adaptive adjustment of the composite magnetic field sequence.

[0132] Dynamically record the change curves of three formation indicators (dispersion indicator, strip indicator, and band indicator) over time throughout the entire process (e.g.) Figure 7 As shown), and snapshots of the spatial distribution of cells at multiple key moments were plotted (e.g. Figure 8 , 9 (As shown in 10, 11, and 12).

[0133] The beneficial effects of this invention are:

[0134] This invention combines a task-adaptive formation control method to achieve precise control of three formation patterns (dispersed formation, strip formation, and ribbon formation) of magnetic microrobots in different vascular stages.

[0135] This invention overcomes the limitations of traditional single-cell magnetic drive control, which is limited by its single-dimensional control and inability to coordinate, by driving multi-cell clusters to move in a coordinated manner through a composite magnetic field. It is expected to significantly enhance the infiltration and therapeutic effect of CAR-T cells on solid tumors.

[0136] This invention proposes a quantitative calculation method for the success rate of three formation types: dispersed, strip, and strip (such as cell spacing ratio, aspect ratio and axial deviation, contact area ratio and velocity matching error), providing repeatable and comparable objective indicators for control effect.

[0137] This invention combines a dynamic model and online parameter optimization (SAC algorithm) to adjust parameters such as magnetic field frequency and amplitude in real time, maintaining a formation success rate of over 90% and adapting to the dynamically changing vascular environment.

[0138] This invention has closed-loop real-time control capability. Through real-time imaging feedback, image processing and controller linkage, a "perception-decision-drive" closed loop is formed, which can dynamically correct formation deviations and ensure control stability. Attached Figure Description

[0139] Figure 1 This is an overall flowchart of the formation control method of the present invention.

[0140] Figure 2 This is a schematic diagram of the system composition and control signal flow of the present invention.

[0141] Figure 3 This is a flowchart of the online parameter optimization based on reinforcement learning according to the present invention.

[0142] Figure 4 This is a simulation diagram of the strip formation of the present invention.

[0143] Figure 5 This is a schematic diagram of the strip formation simulation of the present invention.

[0144] Figure 6 This is the true intention of the distributed formation of the present invention.

[0145] Figure 7 This is a graph showing the change of formation performance indicators of the present invention over time.

[0146] Figure 8 This is a snapshot of the spatial distribution of cell clusters at t=10s, as described in this invention.

[0147] Figure 9 This is a snapshot of the spatial distribution of cell clusters at t=30s, as presented in this invention.

[0148] Figure 10 This is a snapshot of the spatial distribution of cell clusters at t=45s, as presented in this invention.

[0149] Figure 11 This is a snapshot of the spatial distribution of cell clusters at t=60s, as presented in this invention.

[0150] Figure 12 This is a snapshot of the spatial distribution of cell clusters at t=75s, as presented in this invention. Detailed Implementation

[0151] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of this application with unnecessary detail.

[0152] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0153] It should also be understood that the terminology used in this application specification is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this application specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0154] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0155] Many specific details are set forth in the following description in order to provide a full understanding of this application. However, this application may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of this application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0156] Implementation Method 1

[0157] This embodiment provides a task-adaptive formation control method for a magnetic microrobot swarm, the control method comprising the following steps:

[0158] Step 1: Based on real-time imaging feedback, identify the current vascular environment characteristics and determine the current task stage;

[0159] Step 2: Based on the task stage determined in Step 1, generate a composite magnetic field sequence based on the parameter constraints of the three-dimensional Helmholtz coil;

[0160] Step 3: Establish a prediction model based on the composite magnetic field sequence from Step 2;

[0161] Step 4: Based on the prediction model established in Step 3, evaluate the formation success rate;

[0162] Step 5: When the formation success rate assessed in Step 4 is below 90%, perform online parameter optimization;

[0163] Step 6: Implement task-adaptive formation control for the magnetic microrobot cluster.

[0164] Furthermore, the specific characteristics of the current vascular environment in step 1 are: vascular diameter, bifurcation status, and blood flow velocity.

[0165] The specific steps of step 1, which determine the current task stage, are as follows: if the task is in the major blood vessel stage, then dispersed formation is used.

[0166] If the vessel is in the capillary stage, a strip formation should be used;

[0167] If the tumor is in the infiltration stage, a strip formation should be used;

[0168] Three mission formation definitions

[0169] Based on the migration process of magnetic microrobots within the body, this invention defines three task formation patterns (Table 1).

[0170]

[0171] Step 2 specifically involves designing three magnetic field modes based on the parameter limitations of the three-dimensional Helmholtz coil (maximum frequency 15Hz).

[0172] Dispersed formation magnetic field mode:

[0173] X-axis: Rotating magnetic field, frequency 8-12Hz, amplitude 80-100Gs;

[0174] Y-axis: Orthogonal to the X-axis, with a phase difference of 90°;

[0175] Z-axis: Low-frequency oscillation, frequency 1-3Hz, amplitude 20-40Gs;

[0176] Effect: Generates a three-dimensional vortex field, causing the cellular robots to disperse.

[0177] Strip formation magnetic field pattern:

[0178] X-axis: Gradient magnetic field, with a gradient of 0.5-1 Gs / mm along the axial direction of the blood vessel;

[0179] Y-axis: Constraint field, preventing cells from deviating from the axis;

[0180] Z-axis: Closed or weakly constrained;

[0181] Effect: Forms an axially guided linear arrangement

[0182] Strip formation magnetic field pattern:

[0183] X-axis: Oscillation field along the tangential direction of the blood vessel, frequency 5-10Hz;

[0184] Y-axis: The pushing field perpendicular to the blood vessel wall, with a gradient of 0.2-0.5 Gs / mm;

[0185] Z-axis: Low-frequency modulated field, enhancing rolling behavior.

[0186] Effect: Makes cells adhere to the blood vessel wall and roll.

[0187] Furthermore, step 3 includes the following steps:

[0188] Step 3.1: Establish a dynamic model of the magnetic microrobot under a magnetic field;

[0189] Step 3.2: Optimize the target using distributed model predictive control;

[0190] Step 3.3: Adjust the magnetic field frequency in real time using a PID controller to ensure that the formation status meets the target.

[0191] Furthermore, the dynamic model in step 3.1 is specifically as follows:

[0192]

[0193] in For the first The location of each cell It is a composite magnetic field distribution. For blood flow velocity, For random perturbations;

[0194] The specific optimization of the objective in step 3.2 is as follows:

[0195]

[0196] st magnetic field constraint: ;

[0197] Step 3.3, which involves adjusting the magnetic field frequency in real time using a PID controller, specifically involves:

[0198]

[0199]

[0200]

[0201] in For formation state error, For the target formation state, The current formation status, , , These are the parameters for the PID controller.

[0202] Furthermore, the formation state is defined as follows:

[0203] Dispersed formation status : , For robots i and j The distance;

[0204] Strip formation status : , For robots i The azimuth angle;

[0205] Strip formation status : , For robots i The relative position.

[0206] Furthermore, step 4 specifically includes the following steps:

[0207] Step 4.1: Enhance the image of the camera feedback image; use Adaptive Histogram Equalization (CLAHE) to enhance contrast; use Gaussian filtering to remove high-frequency noise (kernel size 3×3);

[0208] Step 4.2: Perform cell segmentation on the enhanced image from Step 4.1; extract candidate cell regions based on the Otsu algorithm, fill in the internal holes of cells using morphological closing operations, label each cell using connected component analysis, and remove artifacts with abnormal area (<5μm² or >50μm²);

[0209] Step 4.3: Localize the cells segmented in Step 4.2; calculate the centroid coordinates of each cell region. Extract cell contours and fit them to an ellipse to obtain the major axis direction angle. Major axis length minor axis length

[0210] Step 4.4: After acquiring cell cluster images through the real-time imaging module, key feature parameters are extracted through image processing to evaluate the formation success rate.

[0211] Furthermore, step 4.4, extracting key feature parameters, specifically involves:

[0212] Cell location set:

[0213]

[0214] Where N is the total number of identifiable magnetic microrobots within the current field of view;

[0215] Cell distance set:

[0216]

[0217] The ratio of the main axis direction of the cluster to its major axis is specifically as follows:

[0218] First, calculate the covariance matrix of the coordinates of all cell centers:

[0219]

[0220] in As the cluster center;

[0221] Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues. and its corresponding eigenvectors ; Maximum eigenvalue Corresponding feature vector This is the main axis direction of the cluster;

[0222] Define the aspect ratio of the cluster :

[0223]

[0224] To align with the vessel axis, it is also necessary to calculate the principal axis direction relative to the vessel axis (the direction of the vessel centerline measured by real-time imaging). The angle between ) :

[0225] .

[0226] Furthermore, the success rates of creating the three types of formations are defined as follows:

[0227] In dispersed formations, the proportion of intercellular spacing within the desired range;

[0228]

[0229] in The average radius of the magnetic microrobot (pre-calibrated); II ( () is an indicator function: it takes the value 1 if the condition is true, and 0 otherwise; The total number of cell pairs;

[0230] In a strip formation, the proportion of cells with a length-to-diameter ratio ≥ 5:1 and an axial deviation ≤ 30°;

[0231]

[0232] In a strip formation, the proportion of cells that are in contact with the blood vessel wall and whose rolling speeds are matched;

[0233] For banded formations, cells are required to adhere closely to the vessel wall and roll along the wall surface; firstly, the vessel wall boundary needs to be segmented from the image; let the vessel wall region be... The outline region of cell i is Define the contact area ratio The proportion of the area where the cell outline overlaps with the blood vessel wall to the total projected area of ​​the cell:

[0234]

[0235] in Indicates area; the larger the contact area ratio, the more tightly the cell adheres to the blood vessel wall;

[0236] Furthermore, the banded formation requires cells to roll along the vessel wall at a speed that matches the local blood flow velocity; the tangential velocity of the cells along the vessel wall can be calculated using inter-frame difference or optical flow methods. Simultaneously, the local blood flow velocity measured by blood flow tracer particles The rolling speed matching error is defined as:

[0237]

[0238] in Minimum speed threshold;

[0239] Qualified cells in a banded formation must simultaneously meet the following requirements:

[0240] Contact area ratio:

[0241] Speed ​​matching error:

[0242] .

[0243] Furthermore, step 5, which employs reinforcement learning for parameter optimization, specifically involves:

[0244] The state space S consists of key variables that reflect the current formation environment and control effectiveness, including:

[0245] Formation type identifier : These represent the current actual forms of the dispersed, strip-shaped, and band-shaped formations, respectively;

[0246] Current formation success rate

[0247] Blood vessel diameter (Unit: μm), obtained from image measurement;

[0248] Blood flow velocity (Unit: cm / s), estimated by particle image velocimetry or a preset model;

[0249] Cell distribution entropy Used to describe the uniformity of cell spatial distribution, defined as

[0250]

[0251] The imaging region is divided into K A sub-region of equal volume To fall into the first k The proportion of cells in each sub-region;

[0252] The state vector is denoted as ;

[0253] The action space A is a combination vector of adjustable magnetic field parameters; based on the hardware limitations of a three-dimensional Helmholtz coil (maximum magnetic field 100 Gs, maximum frequency 15 Hz), each parameter has a physical boundary; the action is a continuous quantity or a discrete quantity after discretization. This invention adopts a continuous action space and outputs it through a Gaussian strategy; wherein the action vector a include:

[0254] XYZ three-axis magnetic field amplitude: , , ;

[0255] XYZ triaxial magnetic field frequencies: , , ;

[0256] The reward function R is designed to guide the agent (controller) to choose actions that improve the success rate of formation while avoiding violations of magnetic field constraints or dangerous behaviors.

[0257] Single-step reward is defined as:

[0258]

[0259] in To reward success rates, this piecewise function encourages maintaining a success rate above 90% and provides a moderate positive incentive to improve it, while penalizing success rates below 50%.

[0260]

[0261] in To enforce penalties for violations, negative penalties are applied when actions exceed the hardware's limits or cause cells to detach from their vascular system (e.g., cells leaving the blood vessel area).

[0262]

[0263] in To reward smoothness of motion and prevent drastic parameter fluctuations from causing control instability, a small penalty is applied to the amount of motion variation:

[0264]

[0265] This invention uses SAC as the reinforcement learning engine; during the algorithm training phase, a large number of state-action-reward samples are generated using a simulation environment (based on a cell dynamics model) to train the value network or policy network.

[0266] During online operation, a decision is made every 200ms:

[0267] 1. Obtain the current status from the real-time imaging module. ;

[0268] 2. If the formation success rate is below 90%, input the state into the policy network and output the action. (Magnetic field parameters);

[0269] 3. The action is mapped to a specific magnetic field control signal, which is generated and applied by an arbitrary waveform generator;

[0270] 4. Wait for the next control cycle and observe the new state. and rewards ;

[0271] 5. Storage experience ( , , , Add it to the experience pool and update the network parameters regularly.

[0272] Through the aforementioned reinforcement learning closed-loop optimization, the system can adaptively adjust the magnetic field parameters, ensuring that the formation success rate remains above 90%, and can respond quickly even to dynamic changes in the vascular environment.

[0273] Implementation Method 2

[0274] This embodiment provides a task-adaptive formation control system for a magnetic microrobot swarm. The control system uses a task-adaptive formation control method for a magnetic microrobot swarm as described in Embodiment 1. The control system includes:

[0275] Task recognition module: Based on real-time imaging feedback, it identifies the current vascular environment characteristics and determines the current task stage;

[0276] Composite magnetic field sequence generation module: Based on the task stage determined by the task identification module, a composite magnetic field sequence is generated based on the parameter constraints of a three-dimensional Helmholtz coil.

[0277] Prediction model building module: Establishes a prediction model based on the composite magnetic field sequence generated by the composite magnetic field sequence generation module;

[0278] Formation success rate assessment module: Based on the prediction model built by the prediction model construction module, the formation success rate is assessed;

[0279] Parameter optimization module: When the formation success rate assessed by the formation success rate assessment module is lower than 90%, online parameter optimization is performed; to achieve task-adaptive formation control of the magnetic microrobot cluster.

Claims

1. A task-adaptive formation control method for a magnetic microrobot swarm, characterized in that, The control method includes the following steps: Step 1: Based on real-time imaging feedback, identify the current vascular environment characteristics and determine the current task stage; Step 2: Based on the task stage determined in Step 1, generate a composite magnetic field sequence based on the parameter constraints of the three-dimensional Helmholtz coil; Step 3: Establish a prediction model based on the composite magnetic field sequence from Step 2; Step 4: Based on the prediction model established in Step 3, evaluate the formation success rate; Step 5: When the formation success rate assessed in Step 4 is below 90%, perform online parameter optimization; Step 6: Implement task-adaptive formation control for the magnetic microrobot swarm; Step 4 specifically includes the following steps: Step 4.1: Perform image enhancement on the camera feedback image; Step 4.2: Perform cell segmentation on the enhanced image from Step 4.1; Step 4.3: Locate the cells segmented in Step 4.2; Step 4.4: After acquiring cell cluster images through the real-time imaging module, key feature parameters are extracted through image processing to evaluate the formation success rate; Step 4.4, extracting key feature parameters, specifically involves: Cell location set: Where N is the total number of identifiable magnetic microrobots within the current field of view; Cell distance set: The ratio of the main axis direction of the cluster to its major axis is specifically as follows: First, calculate the covariance matrix of the coordinates of all cell centers: in As the cluster center; Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues. and its corresponding eigenvectors ; Maximum eigenvalue corresponding feature vector This is the main axis direction of the cluster; Define the aspect ratio of the cluster : Calculate the angle between the principal axis and the vascular axis. : 。 2. The task adaptive formation control method according to claim 1, characterized in that, The specific characteristics of the current vascular environment in step 1 are: vascular diameter, bifurcation, and blood flow velocity. The specific steps of step 1, which determine the current task stage, are as follows: if the task is in the major blood vessel stage, then dispersed formation is used. If the vessel is in the capillary stage, a strip formation should be used; If the tumor is in the infiltration stage, a strip formation should be used; Step 2 specifically involves a dispersed formation magnetic field mode: X-axis: Rotating magnetic field, frequency 8-12Hz, amplitude 80-100Gs; Y-axis: Orthogonal to the X-axis, with a phase difference of 90°; Z-axis: Low-frequency oscillation, frequency 1-3Hz, amplitude 20-40Gs; Strip formation magnetic field pattern: X-axis: Gradient magnetic field, with a gradient of 0.5-1 Gs / mm along the axial direction of the blood vessel; Y-axis: Constraint field, preventing cells from deviating from the axis; Z-axis: Closed or weakly constrained; Strip formation magnetic field pattern: X-axis: Oscillation field along the tangential direction of the blood vessel, frequency 5-10Hz; Y-axis: The pushing field perpendicular to the blood vessel wall, with a gradient of 0.2-0.5 Gs / mm; Z-axis: Low-frequency modulated field, enhancing rolling behavior.

3. The task adaptive formation control method according to claim 2, characterized in that, Step 3 includes the following steps: Step 3.1: Establish a dynamic model of the magnetic microrobot under a magnetic field; Step 3.2: Optimize the target using distributed model predictive control; Step 3.3: Adjust the magnetic field frequency in real time using a PID controller to ensure that the formation status meets the target.

4. The task adaptive formation control method according to claim 3, characterized in that, Step 3.1 The dynamic model is as follows: in For the first The location of each cell It is a composite magnetic field distribution. For blood flow velocity, For random perturbations; The specific optimization of the objective in step 3.2 is as follows: st magnetic field constraint: ; Step 3.3, which involves adjusting the magnetic field frequency in real time using a PID controller, specifically involves: in For formation state error, For the target formation state, The current formation status, , , These are the parameters for the PID controller.

5. The task adaptive formation control method according to claim 4, characterized in that, The formation state is defined as follows: Dispersed formation status : , For robots i and j The distance; Strip formation status : , For robots i The azimuth angle; Strip formation status : , For robots i The relative position.

6. The task adaptive formation control method according to claim 1, characterized in that, The success rates for creating three types of formations are defined as follows: In dispersed formations, the proportion of intercellular spacing within the desired range; in is the average radius of the magnetic microrobot; II() is an indicator function: it takes 1 if the condition is true, otherwise it takes 0; The total number of cell pairs; In strip formations, the proportion of clusters with a length-to-diameter ratio ≥ 5:1 and an axial deviation ≤ 30°; In a strip formation, the proportion of cells that are in contact with the blood vessel wall and whose rolling speeds are matched; For banded formations, cells are required to adhere closely to the vessel wall and roll along the wall surface; firstly, the vessel wall boundary needs to be segmented from the image; let the vessel wall region be... ,cell i The outline region is Define the contact area ratio The proportion of the area where the cell outline overlaps with the blood vessel wall to the total projected area of ​​the cell: in Indicates area; The larger the contact area ratio, the more tightly the cells adhere to the blood vessel wall; Furthermore, the banded formation requires cells to roll along the vessel wall at a speed that matches the local blood flow velocity; the tangential velocity of the cells along the vessel wall can be calculated using inter-frame difference or optical flow methods. Simultaneously, the local blood flow velocity measured by blood flow tracer particles The rolling speed matching error is defined as: in Minimum speed threshold; Qualified cells in a banded formation must simultaneously meet the following requirements: Contact area ratio: Speed ​​matching error: 。 7. The task adaptive formation control method according to claim 2, characterized in that, Step 5, which uses reinforcement learning to optimize parameters, specifically involves... The state space S consists of key variables that reflect the current formation environment and control effectiveness, including: Formation type identifier : These represent the current actual forms of the dispersed, strip-shaped, and band-shaped formations, respectively; Current formation success rate Blood vessel diameter Obtained by image measurement; Blood flow velocity , estimated by particle image velocimetry or a preset model; Cell distribution entropy Used to describe the uniformity of cell spatial distribution, defined as The imaging region is divided into K A sub-region of equal volume To fall into the first k The proportion of cells in each sub-region; The state vector is denoted as ; The action space A is a combination vector of adjustable magnetic field parameters; based on the hardware constraints of the three-dimensional Helmholtz coil, each parameter has a physical boundary; the action is a continuous quantity or a discrete quantity after discretization, and a continuous action space is adopted and output through a Gaussian strategy; The reward function R is designed to guide the agent to choose actions that improve the formation success rate, while avoiding violations of magnetic field constraints or dangerous behaviors. Single-step reward is defined as: in Bonus for success rate: in To constrain violations and penalties: in Rewards for smooth motion: SAC is used as the reinforcement learning engine; a large number of state-action-reward samples are generated using a simulation environment to train the value network or policy network. Decision-making during the online operation phase: Obtain the current status from the real-time imaging module. ; If the formation success rate is below 90%, the state is input into the policy network, and the action is output. ; The action is mapped to a specific magnetic field control signal, which is generated and applied by an arbitrary waveform generator; Waiting for the next control cycle, observing the new state. and rewards ; Storage experience ( , , , Add it to the experience pool and update the network parameters regularly.

8. A task-adaptive formation control system for a magnetic microrobot swarm, characterized in that, The control system uses a task-adaptive formation control method for a magnetic microrobot swarm as described in any one of claims 1-7, and the control system includes: Task recognition module: Based on real-time imaging feedback, it identifies the current vascular environment characteristics and determines the current task stage; Composite magnetic field sequence generation module: Based on the task stage determined by the task identification module, a composite magnetic field sequence is generated based on the parameter constraints of a three-dimensional Helmholtz coil. Prediction model building module: Establishes a prediction model based on the composite magnetic field sequence generated by the composite magnetic field sequence generation module; Formation success rate assessment module: Based on the prediction model built by the prediction model construction module, the formation success rate is assessed; Parameter optimization module: When the formation success rate assessed by the formation success rate assessment module is lower than 90%, online parameter optimization is performed; to achieve task-adaptive formation control of the magnetic microrobot cluster.

Images