Evolutionary nano-robot design method based on biological information feedback

Abstract

The invention relates to an evolutionary nanorobot design method based on biological information feedback, and belongs to the crossing field of soft robots, biomedical engineering and evolutionary computation. The objective of the invention is to solve the problem that an early-stage tiny tumor cannot be automatically recognized and targeted due to the fact that a nano robot is complex in structure / function coupling, depends on tumor prior information and is difficult to dynamically adapt to a biological environment. A multi-objective evolutionary algorithm AFPO is adopted to optimize the structure behavior of the nano-robot, and minimization of parameters such as motion performance is taken as a target; generating a network CPPN coding genotype by using a composite mode, mapping the three-dimensional coordinates to material distribution, and generating a regular structure; a robust filter is designed, and individuals with stable performance are screened through Gaussian noise disturbance; constructing a filter to screen the design conforming to the manufacturing constraint; establishing a reaction flow-dilute substance coupling model based on the tumor microenvironment to generate a three-dimensional time-varying biological gradient field BGF; in the VoxCAD environment, the BGF is used for dynamically guiding the movement adjustment of the nano-robot, so that the autonomous targeting of the early tumor is realized.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of soft robotics, biomedical engineering, and evolutionary computation, and specifically relates to a design method for evolutionary nanorobots based on biological information feedback. Background Technology

[0002] With the rapid development of nanotechnology and biomedical engineering, nanorobots have shown broad application prospects in the field of disease diagnosis and treatment. Since Nobel laureate Richard Feynman's famous "There's Plenty of Room at the Bottom" speech in 1959, the development of nanotechnology and nanomedicine has provided an intelligent means to solve traditional medical diagnosis and treatment (such as early tumor detection).

[0003] However, despite the innovative tools that nanotechnology and nanomedicine have brought to cancer diagnosis and treatment, traditional nanoparticle methods still face key bottlenecks. Statistics show that the targeting efficiency of nanoparticles is only around 0.7%, and has not significantly improved in the past decade. Therefore, nanorobots have been rapidly developed to improve tumor targeting efficiency. Alapan's team designed a therapeutic payload microrobot capable of navigating dynamically flowing blood at a rapid rolling speed of 600 μm / s. Wang developed a smart tumor microenvironment-responsive polymer peptide nanorobot that can effectively deliver CpG payloads to Toll-like receptor 9 (TLR9) positive tumors.

[0004] Current research on nanorobots mainly focuses on biomimetic structural design and system control, falling under the category of molecular biomimetics. Because of the complex coupling relationships between the robot's physical structure, sensing units, power modules, and performance, its design is extremely challenging, and every detail requires extensive and iterative conceptualization, prototyping, and testing before finalization. Therefore, it is difficult to achieve standardized, mass-produced designs of nanorobots with specific functions. Existing nanorobot systems, due to their reliance on prior information about tumor location, cannot achieve early microscopic tumor targeting.

[0005] Current nanorobot systems generally face the following problems:

[0006] 1. Strong coupling between structure and control: There are complex interactions between the physical configuration, sensing unit, power module and execution capability of nanorobots, making their design and optimization process extremely complicated.

[0007] 2. Reliance on prior information: Most existing methods require prior knowledge of the target tumor's location, making it difficult to achieve early detection and targeting of unknown micro-lesions.

[0008] 3. Lack of dynamic adaptability: In complex and variable biological environments (such as blood flow, pH gradient, etc.), traditional nanorobots cannot adjust their behavior according to real-time environmental changes.

[0009] To address the aforementioned issues, this invention employs a "de novo design" research paradigm, simulating the process of natural selection to discover and optimize efficient robot structures and behaviors. Evolutionary algorithms are used to automatically design candidate solutions for nanorobots that meet specific functions in a computer. The biological gradient field (BGF) induced by tumor lesions provides auxiliary knowledge for the design of the nanorobot's morphology, function, and adaptive behavior in complex biological environments. Furthermore, the BGF is used to dynamically guide the evolutionary process by embedding environmental feedback into gene coding as auxiliary information. Summary of the Invention

[0010] The purpose of this invention is to solve the problem that early micro-tumor lesions cannot be autonomously identified and targeted due to the complex physical structure / functional coupling of nanorobots, their reliance on prior information about tumor location, and their difficulty in adapting to the biological environment in real time.

[0011] To achieve the above objectives, the present invention employs the following technical means:

[0012] This invention provides a method for designing evolutionary nanorobots based on biological information feedback, comprising the following steps:

[0013] Step 1: The structure and behavior of the nanorobot are optimized using the multi-objective evolutionary algorithm AFPO, which aims to minimize motion performance, age, and the number of power modules.

[0014] Step 2: Use the Composite Pattern Generation Network (CPPN) to encode the genotype of the nanorobots, map the spatial coordinates of the three-dimensional Cartesian lattice to a specific numerical value to represent the distribution of physical materials, and generate regularized structures through activation functions;

[0015] Step 3: Design robust filters to screen individuals with stable performance under driving phase noise. The robustness of the design is evaluated by Gaussian noise perturbation.

[0016] Step 4: Design a construction filter to screen designs that meet manufacturing constraints. The construction filter includes the following conditions: single type of outer tissue, geometric indentation width ≥ 2 voxels, and passive tissue ratio > 50%.

[0017] Step 5: Establish a reactive flow-dilute matter coupling model based on the tumor microenvironment to generate a three-dimensional time-varying biological gradient field (BGF);

[0018] Step 6: In the VoxCAD environment, BGF is fed back as auxiliary information to the evolution process and dynamically guides the adjustment of the nanorobot's motion state.

[0019] In the above scheme, in order to solve the problem that the complex coupling relationship between the physical structure, sensing unit, power module and performance of nanorobots makes it difficult to mass design, a multi-objective evolutionary algorithm AFPO is adopted. The algorithm optimizes the motion performance, age and number of power modules with the goal of minimizing them. Starting from a random population, the algorithm iteratively optimizes the design through gene encoding, selection, crossover and mutation.

[0020] In the above scheme, the Composite Pattern Generation Network (CPPN) is used for genotype encoding, which maps the spatial coordinates of the three-dimensional rectangular lattice to the material distribution, thereby realizing the automated generation and optimization of complex configurations.

[0021] in:

[0022] The input parameters include: rectangular coordinates (x, y, z), radial distance to the center of the lattice workspace, and an offset term (value 1);

[0023] Activation functions include ±sin(), ±abs(), ±square(), and ±sqrt(abs()), which are used to generate regularized spatial structures.

[0024] The output determines whether a voxel contains material and its type based on a threshold, including passive voxels or contractile voxels.

[0025] Network mutation operations include: adding / deleting / modifying nodes or edges, adjusting weights, and repeating attempts if the mutation is neutral until a non-neutral mutation or the preset number of attempts is exceeded.

[0026] In the above scheme, a driving phase noise perturbation is applied to the design of high-performance nanorobots generated through evolution, and individuals with the least performance degradation under the perturbation are selected; specifically, this includes:

[0027] The 1.00 highest-performing designs are ranked according to their robustness to random perturbations, resulting in a design sequence of A1, A2, ..., A 100 A1 is the most robust, A 100 The least robust;

[0028] Make 20 copies of each design;

[0029] For each copy, the driving phase shift of all contracted voxels is independently abruptly mutated by applying Gaussian noise perturbation (using a random seed of numbers 1-20 randomly drawn from a normal distribution), with the driving phase shift ranging from [value missing].

[0030] To prevent parameters from disturbing the original phase offset values ​​and pushing all abrupt changes towards The boundary, therefore, the noise satisfies the following distribution:

[0031] N(μ=0, σ=0.4π)

[0032] μ represents the noise mean, used to ensure that the disturbance has no systematic shift;

[0033] σ represents the noise standard deviation, used to control the disturbance intensity;

[0034] Calculate the average motion performance of each design under disturbance, including net displacement;

[0035] The design with the least performance degradation under phase noise perturbation is selected and ranked by robustness.

[0036] The above scheme selects nanorobot designs based on actual manufacturing constraints, specifically including the following conditions:

[0037] The nanorobot has a single type of outermost organization in at least one spatial direction, and the single type is achieved by three-dimensional spatial rotation.

[0038] The geometric indentation width of the nanorobot satisfies:

[0039] w≥2 voxels

[0040] in:

[0041] w represents the geometric recess width (unit: voxel), used to prevent structural failure;

[0042] The passive tissue volume ratio of the nanorobots satisfies:

[0043] r passlive >50%

[0044] in:

[0045] r passive This represents the percentage of passive tissue in the total volume of the nanorobot, used to reduce metabolic costs and reserve space for functional expansion.

[0046] In the above scheme, a reaction flow-dilute matter coupling model of the tumor microenvironment is established using multiphysics simulation tools to generate a three-dimensional time-varying biological gradient field (BGF), specifically including:

[0047] Establish a three-dimensional geometric structure that includes the tumor core and surrounding tissue;

[0048] Simulation of hemodynamics and hydrogen ion diffusion using a bidirectionally coupled physics field module:

[0049] Laminar flow module: Hemodynamics is described using the Navier-Stokes equations.

[0050]

[0051] u: Blood flow velocity field, driving the convective transport of substances;

[0052] p: hydrostatic pressure, which drives blood flow;

[0053] ρ: Blood density;

[0054] μ: dynamic viscosity;

[0055] F: Volume force;

[0056] Dilution Species Transport Module: Hydrogen ion transport is described using a convection-diffusion-reaction equation.

[0057]

[0058] in.

[0059] c i Hydrogen ion concentration;

[0060] D i : Diffusion coefficient;

[0061] R i : Reaction term, describing metabolic rate;

[0062] The hydrogen ion concentration distribution was mapped to a three-dimensional time-varying BGF based on pH value, serving as a guiding signal for the motion behavior of nanorobots.

[0063] In the above scheme, the behavior of nanorobots is evolved by setting different objective functions in the VoxCAD environment, specifically including:

[0064] Evolution of motor behavior:

[0065] Multiple independent evolutionary experiments were conducted, with each experiment using a randomized initial design of 50 populations.

[0066] The objective function is to maximize the net displacement Δd within a 10-second cycle in the tumor microenvironment, with the direction of motion dynamically guided by the local gradient value of the BGF.

[0067] Optimal performance design is selected and constructed using robust filters;

[0068] Evolution of obstacle avoidance behavior:

[0069] Multiple independent evolutionary experiments were conducted, with each experiment involving a population size of 10 designs.

[0070] A 2×2×2 voxel obstacle is set in the work area and moves along the negative y-axis driven by blood flow.

[0071] The objective function is to minimize the number of collisions between obstacles and the robot, and to maximize the surface area to volume ratio to prevent clever design tricks involving holes;

[0072] Obstacle avoidance driven by fluid dynamics is achieved by varying the stiffness of voxels.

[0073] Evolution of rejection behavior:

[0074] Multiple independent evolutionary experiments were conducted, with each experiment involving a population size of 50.

[0075] A 2×2×2 voxel repellent was placed above the work area;

[0076] The objective function is to maximize the distance of the thrown repulsive object and maximize the surface area to volume ratio to ensure the evolution of throwing limbs;

[0077] Reduce fluid resistance through cross-sectional perforation design.

[0078] Because the present invention employs the above-mentioned technical means, it has the following beneficial effects:

[0079] 1. This invention addresses the challenge of mass design caused by the strong coupling between the physical structure, sensing units, and power modules of nanorobots by employing the multi-objective evolutionary algorithm AFPO to optimize the nanorobot structure (step 1) and combining it with the composite pattern generation network CPPN for genotyping (step 2). Through a regularized coding mechanism that maps three-dimensional spatial coordinates to material distribution, the invention achieves automated generation and optimization of complex configurations, significantly reducing design complexity and avoiding repeated manual trial and error. This provides a scalable technical path for the efficient generation of nanorobots with specific functions.

[0080] 2. This invention generates a three-dimensional time-varying biological gradient field (BGF) by establishing a reaction flow-dilute matter coupling model of the tumor microenvironment (step 5), and dynamically guides the adjustment of motion state using the BGF in the VoxCAD environment (step 6). This solves the key shortcomings of existing methods that rely on prior information about tumor location and cannot autonomously identify early microscopic lesions. By embedding environmental feedback (such as pH gradient) into gene encoding as a real-time guidance signal, the nanorobot can autonomously adjust its motion direction and behavior pattern according to the local BGF gradient, achieving early tumor targeting even under conditions of unknown lesion location, and significantly improving the active detection capability of microscopic lesions.

[0081] 3. This invention addresses the problems of poor adaptability and unmanufacturability of traditional simulation designs in real biological environments by designing robust filters to apply driving phase noise perturbations (step 3) and constructing filters to apply manufacturing constraints (step 4). Specifically, it screens for driving robust individuals through Gaussian noise perturbation and combines this with the homogenization of outer tissues, geometric concave width limitations, and passive tissue proportion constraints to ensure that the evolutionary design possesses both environmental resistance and physical feasibility. This achieves a triple effect of simultaneously improving the stability of nanorobots in dynamic blood flow environments, preventing structural failure, and reserving space for functional expansion. Attached Figure Description

[0082] Figure 1 : Flowchart of the implementation of this invention;

[0083] Figure 2 : Flowchart of the filter design of this invention;

[0084] Figure 3 : A three-dimensional geometric structure diagram of the tumor microenvironment of this invention;

[0085] Figure 4 : Time-varying BGF snapshot image of the present invention;

[0086] Figure 5 : Experimental results of the present invention. Detailed Implementation

[0087] The embodiments of the present invention will be described in detail below. Although the present invention will be described and illustrated in conjunction with some specific embodiments, it should be noted that the present invention is not limited to these embodiments. On the contrary, any modifications or equivalent substitutions made to the present invention should be covered within the scope of the claims of the present invention.

[0088] Furthermore, to better illustrate the present invention, numerous specific details are set forth in the following detailed embodiments. Those skilled in the art will understand that the present invention can be practiced without these specific details.

[0089] The purpose of this invention is to provide an evolutionary nanorobot design method based on bioinformation feedback. This method uses evolutionary computation technology to design nanorobots with corresponding functions in specific biological environments. The resulting nanorobots can continuously adjust their motion state (e.g., expansion or contraction) and direction based on changes in the biogenic globulin (BGF) at the location of the tumor lesion, even without knowing the exact location of the tumor, gradually approaching the tumor lesion to achieve specific functions (such as early detection of tumor lesions).

[0090] The objective of this invention is achieved through the following technical solution:

[0091] A bio-information feedback-based evolutionary nanorobot design method includes:

[0092] A multi-objective evolutionary algorithm (AFPO) is employed to optimize performance (e.g., net displacement), age (to maintain population diversity), and the minimization of the number of dynamic modules (to reserve space for functional expansion). The algorithm starts with a random population and iteratively optimizes the design through gene encoding, selection, crossover, and mutation.

[0093] Genotype encoding is performed using a Composite Pattern Generating Network (CPPN). CPPN maps three-dimensional spatial coordinates (x, y, z) to material distributions (such as passive voxels and contractile voxels), generates regularized structures through activation functions, and supports arbitrary resolution.

[0094] Design robust filters to apply drive phase noise (Gaussian distribution) to high-performance designs, select those with the least performance degradation under disturbance, and ensure robustness to drive signal uncertainty;

[0095] The filter was designed and constructed based on actual manufacturing constraints: single type of outer tissue (which can be satisfied by rotation), geometric indentation width ≥ 2 voxels (to prevent structural failure), and passive tissue ratio > 50% (to reduce metabolic costs and reserve space for sensors / drugs).

[0096] A reactive flow-dilute matter coupling model of the tumor microenvironment (including the Navier-Stokes equations and convection-diffusion-reaction equations) was established using COMSOL to simulate blood flow and hydrogen ion diffusion, generating a three-dimensional time-varying biological gradient field (BGF, such as pH distribution). The BGF is key to the nanorobots' environmental perception and driving behavior (voxels generate motion based on the expansion / contraction of local BGFs).

[0097] In the VoxCAD environment, three core behaviors evolved by setting different objective functions: motion, obstacle avoidance, and repulsion. Three experiments verified the effectiveness and scalability of the proposed method.

[0098] Example 1

[0099] (1) Simulation environment:

[0100] The simulation environment of this invention is based on the VoxCAD simulator and the COMSOL multiphysics simulation tool.

[0101] VoxCAD focuses on voxel-based (3D pixel) modeling and is commonly used for rapid prototyping in fields such as soft robotics and flexible material deformation. Its built-in physics engine supports basic mechanical simulations (tension, compression, and collision response) and allows adjustment of parameters such as material elastic modulus and damping. Its core functions include mechanical simulation of voxel structures and visualization of dynamic behavior.

[0102] COMSOL is a general-purpose simulation platform based on the finite element method, supporting coupled analysis of more than 30 physics modules, including electromagnetics, structural mechanics, fluid dynamics, heat transfer, acoustics, and chemical reactions. Users can freely combine modules through the "physics interface" to simulate complex interactive phenomena in the real world. COMSOL can perform multiphysics coupling modeling of reactive flow and dilute matter in the tumor microenvironment.

[0103] (2) Evolutionary design:

[0104] Evolutionary robotics offers a novel approach to nanorobot design by employing computer simulations using evolutionary algorithms. Unlike traditional robot design methods that design robot structure and control systems independently, evolutionary robotics, using a "de novo" paradigm, simulates natural selection to discover and optimize efficient robot structures and behaviors. It then uses evolutionary algorithms to iteratively optimize the robot as a whole, calculating the optimal solution for specific environments (such as tumor microenvironments) and functions as the output. The specific structure of the "de novo" design paradigm is as follows: Figure 1 As shown.

[0105] The evolutionary process begins with a randomly initialized, diverse population of individuals, each whose characteristics are genetically encoded. These genes can evolve through selection, crossover, mutation, and other operations. Then, the motion of each nanorobot is simulated in a specific simulation environment, and its performance is automatically scored. Poor-performing designs are deleted and replaced with randomly modified copies of better-performing designs. Repeating this process yields a large number of high-performance and diverse designs.

[0106] (3) Encoding method:

[0107] This invention employs a Composite Pattern Generating Network (CPPN) for genotype encoding. This design maps the spatial coordinates of a three-dimensional Cartesian lattice to a specific numerical value to represent the distribution of physical materials, and the encoding method is scale-independent, enabling mapping to coordinate spaces of arbitrary resolution. The coordinates of each constituent unit (voxel) are determined by its Cartesian coordinates (x, y, z) and its radial distance to the center of the lattice working space. These four coordinates (along with a bias itemset equal to 1) are used as the network input.

[0108] ±sin(), ±abs(), ±square(), and ±sqrt(abs()) are used as activation functions to generate regularized spatial structures. A threshold is used to determine whether a voxel contains material and its type, which is then output. Network crossover and mutation methods include adding, deleting, and modifying nodes or edges, with weight adjustments (within ±1). If the mutation is neutral (phenotype unchanged), the process is repeated until a non-neutral mutation or exceeding the limit (1500 times). The CPPN genotype encoding, through regularized function mapping and a dynamic mutation strategy, achieves scalable morphology generation, which is crucial in the evolution of nanorobots.

[0109] (4) Algorithm design:

[0110] This invention employs a multi-objective evolutionary algorithm, Age-Fitness-Pareto Optimization (AFPO), which promotes population diversity while optimizing a robot to satisfy multiple objectives. AFPO utilizes the concept of Pareto dominance and the objective of "age" (in addition to performance) to promote diversity in candidate designs and prevent premature convergence. Besides performance and age, AFPO also uses a third objective: minimizing the number of dynamic modules. This additional objective aims to ensure that the organism has sufficient non-functional volume to accommodate more extended functions (such as payload, sensing, etc.).

[0111] (5) Filter design:

[0112] like Figure 2 As shown, due to the numerous differences between the simulated environment and the real physical environment, a robust filter is used to select individuals that still perform well in noisy environments to ensure that the generated nanorobots are suitable for real-world environments (such as the high-risk tissue microenvironment of the human body). The selected individuals are then processed by constructing a filter to remove individuals that cannot be instantiated using existing fabrication methods and those with poor scalability.

[0113] For robust filters, this stage evaluates the design's tolerance to driving perturbations. From each evolution, 100 high-performance designs are ranked according to their robustness to random perturbations in the drive. Twenty copies of each design are made, and Gaussian noise (mean 0, standard deviation 0.4π) is independently applied to the driving phase shift of all contracted voxels in each copy to perturb the design. The average motion performance (e.g., net displacement) under the perturbation is calculated for each design. The design with the least performance degradation under phase noise is selected and ranked by robustness before being sent to the manufacturing screening stage, ensuring that the selected designs are robust to uncertainties in the driving signal.

[0114] For filter construction, this stage involves screening designs based on actual manufacturing constraints and includes three core criteria:

[0115] ① The outermost tissue in a certain spatial direction (such as dorsal or anterior-posterior) must be of a single type (pure passive or pure contractile voxel), which can be satisfied by 3D rotation;

[0116] ② The width of the geometric indentation (gap) must be at least 2 voxels (approximately 25% of the total length) to prevent cell adhesion from causing structural failure (experiments show that indentations with a width of approximately 12.5% ​​of a single voxel are unstable).

[0117] ③ Passive tissue must account for more than 50% to reduce the high metabolic costs of shrinking tissue, extend its lifespan, and reserve space for future integration of functional modules such as sensors and payloads. Designs selected through this process possess manufacturability, structural stability, and potential for functional expansion.

[0118] (6) Biological environment characterization:

[0119] Nanorobots, as biosensors, can sense the BGF at their location (e.g., the nanorobots in this invention sense the pH value of an acidic environment as reactants) and generate movement (the expansion or contraction of voxels).

[0120] In the robot design phase, this method uses the tumor-induced biogeometry (BGF) as auxiliary knowledge to automatically design the optimal solution under the tumor microenvironment and function. During tumor targeting, the method incorporates the local gradient value of the BGF into the fitness assessment, providing guidance for the robot's motion. The designed nanorobot continuously adjusts its motion state (e.g., expansion or contraction) and orientation based on changes in the BGF at its location, gradually approaching the target even without knowing the location of the tumor lesion, thus achieving early tumor lesion detection. The above discussion shows that the BGF is the objective function for the adaptive assessment of the computational agent, exhibiting time-varying characteristics, and that the local gradient value of the BGF provides an observable measure of the nanorobot's spatial location based on its perceptual information. Therefore, the dynamic in vivo computation problem can be represented by multiphysics modeling of the BGF.

[0121] This invention uses COMSOL Multiphysics to perform reactive flow-dilute matter multiphysics simulation modeling of the tumor microenvironment to study the pH distribution in a three-dimensional time-varying tumor microenvironment (BGF). The three-dimensional geometry of the model is as follows: Figure 3 As shown, the tumor consists of two parts: the tumor core and the surrounding tissue. The 1mm diameter sphere at the center of the image represents the tumor core; the 10mm × 10mm × 10mm cube represents the surrounding tissue, i.e., the search space for the nanorobots. Figure 4As shown, the simulation model established in this invention simultaneously studies fluid flow and hydrogen ion mass transfer in the tumor microenvironment, using the hydrogen ion concentration distribution as a three-dimensional time-varying BGF representation based on pH. We selected the "laminar flow" module in the physics field to describe fluid motion (Navier-Stokes equations), and the "dilution species transport" module to simulate hydrogen ion transport. These two modules are bidirectionally coupled through convection terms to achieve a closed-loop simulation of matter-blood flow-metabolism in the tumor microenvironment.

[0122] The "Laminar Flow" module determines the motion of fluid in laminar flow. The Navier-Stokes equations are as follows:

[0123]

[0124] Where u is the blood flow velocity field, describing the distribution of blood flow velocity within the blood vessel and driving the convective transport of substances; p is the hydrostatic pressure, the main driving force for blood flow; ρ is the density of blood, which may be altered by hypoxia in the tumor area; μ is the dynamic viscosity, reflecting the viscosity of blood; and F is the volume force (such as gravity).

[0125] The Navier-Stokes equations provide a framework for fluid motion, upon which the convection-diffusion-reaction equations quantify biochemical processes. The convection-diffusion-reaction equations are as follows:

[0126]

[0127] Where c i The concentration of a substance (such as hydrogen ions) characterizes the spatial distribution of oxygen, drugs, or growth factors, and determines the metabolic state of a tumor; D i R is the diffusion coefficient, describing the rate of random molecular motion in the interstitial space; i The reaction term is a core variable that describes the rate of metabolism, synthesis, or consumption of substances.

[0128] (7) Experimental analysis:

[0129] The nanorobots of this invention are evolved within VoxCad through the reconfigurable aggregation of passive and contractile voxels. The nanorobots' movement is generated by the expansion or contraction of voxels according to the BGF (pH field) at their location. The reconfigurable nanorobots exhibit three distinct behaviors—movement, obstacle avoidance, and repulsion—through evolution with different objective functions. These behaviors are all realized in the same simulated environment. This simulated environment consists of an infinite plane completely immersed in blood, and first-order hydrodynamics is introduced to ensure the accuracy of the simulation. The fluid density is assumed to be blood (ρ = 1060 kg / m³). 3 By reducing the acceleration due to gravity to -0.1 m / s² 2The simulation environment is approximated by adding buoyancy, applying drag (drag coefficient C = 1.5) to each voxel surface of the design surface, and applying thrust to each voxel surface of the surface along the negative y-axis (simulating blood flow).

[0130] The results of the experiment are as follows Figure 5 As shown.

[0131] For the movement behavior, this invention conducted 60 independent evolutionary algorithm trials, each starting with a different initial randomized design (population size 50 individuals). In each trial, the design was selected based on its net displacement achieved within a 10-second period in the tumor microenvironment. The direction of movement was guided by the local gradient value of the BGF. At the end of each trial, the design with the highest fitness was extracted and processed through robust filters and construct filters.

[0132] For obstacle avoidance behavior, this invention conducted 200 evolutionary algorithm experiments (population size 10). In each experiment, a 2×2×2 voxel object (obstacle) was placed in the middle of the rightmost side of the working area. Driven by blood flow, the obstacle moved towards the robot along the negative y-axis. The design objective was to minimize the number of collisions between the obstacle voxel and the robot voxel. In addition to the number of collisions, this application introduced an additional objective into the objective function: maximizing the surface area to volume ratio. This objective was introduced to apply direct selection pressure to the limbs, ensuring that holes are not directly evolved to avoid obstacles during the design process, thus making the design objective more focused on the robot's obstacle avoidance behavior. The designed objective function is:

[0133]

[0134] Where CollisionCounter represents the cumulative number of collisions recorded by the collision counter; SA / V represents the maximum surface area to volume ratio; SA = 6·N-2·E represents the total surface area, where N is the number of voxels and E is the number of neighbor connections; V = N represents the volume; α represents the weight coefficient of the collision penalty term; and β represents the weight coefficient of the morphology optimization term.

[0135] Figure 5 (b) The yellow voxels represent voxel stiffness. The additional objective of maximizing the surface area to volume ratio enables the robot to evolve a jellyfish-like swimming posture, achieved through the contraction and expansion of its limbs, thus evolving a unique hydrodynamic mechanism.

[0136] For the rejection behavior, this invention conducted 50 experiments, with each iteration generating a population of 50 individuals. In each experiment, a 2×2×2 voxel object (rejector) was placed one voxel above the 8×8×7 working area, and a non-motion-based objective was introduced into the objective function: maximizing the distance the object was thrown. In addition to maximizing the throwing distance, we also introduced maximizing the surface area to volume ratio. This objective was introduced to ensure the evolution of the limbs required for throwing the object. The specific objective function is designed as follows: Where D is the maximum distance the object is thrown. In the fourth evolutionary experiment, the computer reduced blood flow drag by designing a perforation in the center of its cross-section and threw objects through the connection above the perforation.

[0137] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An evolutionary nanorobot design method based on bio-information feedback, characterized by, The method comprises the following steps: Step 1: a multi-objective evolutionary algorithm AFPO is used to optimize the structure and behavior of the nanobot, and the AFPO takes motion performance, age and minimum number of power modules as the target; Step 2: a composite pattern generation network CPPN is used to encode the genotype of the nanobot, and a space coordinate of a three-dimensional rectangular lattice is mapped to a specific numerical value to represent the distribution of physical materials, and a regularization structure is generated through an activation function; Step 3: a robust filter is designed to screen individuals with stable performance under driving phase noise, and the robust filter evaluates the robustness of the design through Gaussian noise disturbance; Step 4: a construction filter is designed to screen designs meeting manufacturing constraints, and the construction filter includes conditions of outer layer organization single type, geometric recess width ≥ 2 voxels and passive organization volume ratio > 50%; Step 5: a reaction flow-dilute substance coupling model is established based on a tumor microenvironment to generate a three-dimensional time-varying biological gradient field BGF; Step 6: in the VoxCAD environment, the BGF is fed back to the evolution as auxiliary information, and the motion state of the nanobot is dynamically guided.

2. The method of claim 1, wherein: A multi-objective evolutionary algorithm AFPO is used to optimize the structure and behavior of the nanobot, and the AFPO takes motion performance, age and minimum number of power modules as the target, starting from a random population, and iteratively optimizing the design through gene encoding, selection, crossover and mutation.

3. The method of claim 1, wherein: A composite pattern generation network CPPN is used for genotype encoding, and a space coordinate of a three-dimensional rectangular lattice is mapped to a material distribution for realizing automatic generation and optimization of a complex configuration. Wherein: The input parameters include: Cartesian coordinates , radial distance to the center of the lattice workspace, and a bias term (value of 1); The activation function includes: , , , for generating regularized spatial structure; The output is a threshold judgment on whether a voxel exists and its type, including passive voxels or shrinkage voxels. Network mutation operations include: adding / deleting / modifying nodes or edges, weight adjustment, and if neutral mutation, repeated attempts until non-neutral mutation or exceeding a preset number of times.

4. The method of claim 1, wherein: A driving phase noise disturbance is applied to the high-performance nanobot design generated by evolution to screen individuals with the least performance decline under the disturbance, specifically including: The n best performing designs are ranked according to their robustness to random perturbations, the ranked sequence of designs being , , , wherein is the most robust, is the least robust; Each design is replicated 20 copies; Mutations were performed by independently applying a Gaussian noise perturbation to the driving phase offset of all shrunken voxels in each copy, in the range ; To prevent the parameter from pushing all the discontinuities towards the boundaries after disturbing the original phase offset values, the noise therefore satisfies the distribution: represents the noise mean, to ensure that the disturbance has no systematic offset; represents the noise standard deviation, used to control the perturbation intensity; The average motion performance of each design under disturbance is calculated, and the motion performance includes net displacement; The design with the least performance decline under phase noise disturbance is screened, and the output is ranked according to robustness.

5. The method of claim 1, wherein: The nanobot design is screened according to actual manufacturing constraints, specifically including the following conditions: The outermost layer of the nanobot in at least one spatial direction is of a single type, and the single type is realized through three-dimensional space rotation; The geometric recess width of the nanobot satisfies: Wherein: Geometric indentation width (in voxels) to prevent structural failure; The passive organization volume ratio of the nanobot satisfies: Wherein: represents the percentage of passive tissue in the total volume of the nanorobot, used to reduce metabolic cost and reserve space for functional expansion.

6. The method of claim 1, wherein: A reaction flow-dilute substance coupling model of a tumor microenvironment is established using a multi-physical field simulation tool to generate a three-dimensional time-varying biological gradient field BGF, specifically including: A three-dimensional geometric structure including a tumor core and surrounding tissue is established; Blood flow dynamics and hydrogen ion diffusion are simulated through bidirectional coupled physical field modules: A laminar flow module: Navier-Stokes equation is used to describe blood flow dynamics: : blood flow velocity field, driving material convective transport; : fluid static pressure, drives blood flow; : blood density; : dynamic viscosity; : volume force; A dilute species transport module: convection-diffusion-reaction equation is used to describe hydrogen ion transport: Wherein: : hydrogen ion concentration; : diffusion coefficient; : reaction term, describes the rate of metabolism; The hydrogen ion concentration distribution is mapped to a three-dimensional time-varying BGF based on pH value as the guide signal for the motion behavior of the nanorobot.

7. The method of claim 1, wherein: The behavior of the nanorobot is evolved in the VoxCAD environment by setting different objective functions, specifically including: Motion behavior evolution: Multiple independent evolution experiments are performed, with a population size of 50 random initial designs for each experiment; The objective function is to maximize the net displacement in a 10 second period in the tumor microenvironment The direction of motion is dynamically guided by the local gradient values of the BGF. The best-performing design is selected and filtered through a robust filter and a construction filter; Obstacle avoidance behavior evolution: Multiple independent evolution experiments are performed, with a population size of 10 designs for each experiment; A 2x2x2 voxel obstacle is set in the workspace and is pushed by the blood flow to move in the negative y-axis direction; The objective function is to minimize the number of collisions between the obstacle and the robot, and to maximize the surface area to volume ratio to prevent hole cheating designs; Fluid dynamics-driven obstacle avoidance is achieved through changes in voxel stiffness; Repulsion behavior evolution: Multiple independent evolution experiments are performed, with a population size of 50 designs for each experiment; A 2x2x2 voxel repeller is set above the workspace; The objective function is to maximize the distance of throwing the repeller and to maximize the surface area to volume ratio to ensure that the throwing limb is evolved; Fluid resistance is reduced through cross-sectional perforation design.

Images