Micro-robot motion control method and system

Abstract

The invention provides a micro robot motion control method and system, and the method comprises the steps: carrying out the synchronization and calibration processing of environment state information, and generating a comprehensive perception image; analyzing the comprehensive perception image to obtain a classification result, and predicting the environment state change influencing the motion track of the micro robot in a period of time in the future based on the classification result; generating a candidate path by using a fast random tree, and optimizing the candidate path in combination with an AI search algorithm to obtain a target path; constructing a three-dimensional dynamic probability graph, continuously updating the three-dimensional dynamic probability graph by using a reinforcement learning algorithm, and generating an instant decision map; and estimating the position of the micro-robot by using a Kalman filter to obtain the position state of the micro-robot, adjusting the speed and steering angle of a driving motor of the micro-robot, and generating trajectory tracking of the micro-robot. According to the technical scheme provided by the invention, the flexibility and intelligence of the micro robot are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of robot control, and in particular relates to a micro robot motion control method and system. Background Art

[0002] Micro robots, due to their small size and flexibility, have shown broad application prospects in many key areas. They realize precision operation and quality inspection in industrial automation; assist surgery, drug delivery and patient care in the medical field; perform crop monitoring and pollution detection in environmental monitoring; provide daily assistance such as cleaning and security in home services, significantly improving the efficiency and safety of various industries.

[0003] At present, the motion control of micro-robots mainly relies on traditional path planning algorithms and simple feedback control systems. Using predefined rules to generate paths is suitable for structured and static environments, but it is difficult to cope with complex and changing scenes. Basic position tracking is achieved through proportional-integral-derivative (PID) control, but its performance is limited by the accuracy of the model and the stability of the environment. It is used to describe the behavior pattern of the robot, but lacks the ability to adapt to environmental changes.

[0004] Although existing methods perform well under certain specific conditions, they still have significant shortcomings: traditional methods usually assume that the environment is static or slowly changing, and cannot effectively handle dynamically changing environments, causing the robot to lose its way or collide. Most solutions lack advanced cognitive capabilities and learning mechanisms, making it difficult to make optimal decisions based on actual conditions, especially in the face of emergencies. Existing path planning methods often focus on finding feasible paths rather than optimal paths, which may lead to inefficiency or waste of resources. In order to ensure real-time performance, many systems sacrifice the accuracy of position estimation; and the pursuit of high precision may cause calculation delays and affect real-time response. Summary of the invention

[0005] The present invention aims to provide a micro robot motion control method and system to solve the problems of poor flexibility and low intelligence of the micro robot in the prior art.

[0006] In a first aspect, an embodiment of the present invention provides a micro robot motion control method, comprising:

[0007] Using non-contact sensors to collect current environmental status information in real time, and synchronizing and calibrating the environmental status information to generate a comprehensive perception image;

[0008] Applying target detection and recognition technology to analyze the comprehensive perception image to determine the objects existing in the current environment and the categories of the objects, obtain a classification result, and construct an environment state change prediction model based on the classification result and the environment state information. The environment state change prediction model is used to predict the environment state changes that affect the motion trajectory of the micro robot in the future period of time through historical environment state information and the current environment state information, combined with the classification result;

[0009] According to the comprehensive perception image and the environmental state change predicted by the environmental state change prediction model, a candidate path is generated using a fast random tree, and the candidate path is optimized in combination with an AI search algorithm to obtain a target path;

[0010] Based on the target path, a three-dimensional dynamic probability graph is constructed using a Bayesian network, wherein the three-dimensional dynamic probability graph uses the environmental state at a specific location as a node and the transition probability between the environmental states as an edge, and the three-dimensional dynamic probability graph is continuously updated using a reinforcement learning algorithm to generate an instant decision map;

[0011] When a change in the environmental state is detected, based on the instant decision map, an emotional computing model is introduced to simulate the emotional factors in the human decision-making process, and the motion trajectory of the micro-robot is adjusted to generate an adaptive behavior rule set;

[0012] A Kalman filter is used to estimate the position of the microrobot to obtain the position state of the microrobot. Based on the position state and the adaptive behavior rule set, the speed and steering angle of the drive motor of the microrobot are adjusted to ensure that the actual motion trajectory conforms to the ideal motion trajectory, and the trajectory tracking of the microrobot is generated. The trajectory tracking refers to the process in which the microrobot moves according to a predetermined motion trajectory during the execution of a task, and adjusts its own speed and steering angle in real time to minimize the deviation between the actual motion trajectory and the ideal trajectory.

[0013] Optionally, it also includes:

[0014] After the micro-robot completes a path planning and performs the trajectory tracking, the deviation between the actual motion trajectory and the ideal motion trajectory is compared to obtain a comparison result;

[0015] Based on the comparison result, determining an effective behavior rule from the adaptive behavior rule set, wherein the effective behavior rule refers to a behavior rule that can reduce the deviation between the actual motion trajectory and the ideal trajectory;

[0016] The effective behavior rules are incorporated into the motion control method library of the micro robot, so that the effective behavior rules in the motion control method library can be used to realize the motion control of the micro robot in future tasks.

[0017] Optionally, based on the target path, a three-dimensional dynamic probability graph is constructed using a Bayesian network, the three-dimensional dynamic probability graph uses the environmental state at a specific location as a node, and the transition probability between the environmental states as an edge, and the three-dimensional dynamic probability graph is continuously updated using a reinforcement learning algorithm to generate an instant decision map, including:

[0018] Based on the comprehensive perception image and the environmental state change prediction model, the actual motion trajectory of the micro robot is planned and optimized to obtain a target path, and a fast random tree or variant algorithm combined with an AI search algorithm is used for path planning and optimization to further refine and optimize the target path to generate an optimal path;

[0019] According to the optimal path, the environmental state of the specific location and the transition probability are modeled using a Bayesian network to construct a three-dimensional dynamic probability map, and based on the three-dimensional dynamic probability map, the environmental change is predicted through spatiotemporal correlation analysis;

[0020] Based on the three-dimensional dynamic probability map, a reinforcement learning algorithm is used to continuously update and generate a real-time decision map, so as to learn and optimize the behavior strategy of the micro robot through the real-time decision map to improve the decision quality.

[0021] Optionally, the three-dimensional dynamic probability map is continuously updated and processed by using a reinforcement learning algorithm to generate an instant decision map, including:

[0022] Based on the initial motion mode of the micro-robot, selection processing is performed, the non-contact sensing data of the micro-robot is integrated, and the current environmental state is evaluated in combination with the environmental state change prediction model, and the weight of the non-contact sensing data is adjusted in real time to obtain an optimized initial environmental evaluation result;

[0023] Whenever a new comprehensive perception image is acquired, differential privacy protection technology is applied to protect the security of the non-contact perception data and effectively shield sensitive information, thereby achieving the security of the non-contact perception data without affecting the accuracy of the three-dimensional dynamic probability map, and combining the newly acquired comprehensive perception image with the three-dimensional dynamic probability map to update the three-dimensional dynamic probability map to generate an updated three-dimensional dynamic probability map, wherein the updated three-dimensional dynamic probability map is used to reflect the latest environmental status;

[0024] Based on the updated three-dimensional dynamic probability map, a decision-making framework combining deep reinforcement learning and evolutionary algorithm is used to generate a real-time decision map. The framework is used to combine integrated multi-agent reinforcement learning to enable multiple micro-robots to collaborate with each other to complete complex tasks. The complex tasks include group obstacle avoidance and collaborative transportation, and adversarial training elements are used to simulate opponents or unforeseen environmental factors to generate a real-time decision map. The real-time decision map is used to guide the micro-robots to make optimal decisions.

[0025] Optionally, based on the updated three-dimensional dynamic probability map, a decision framework combining deep reinforcement learning and evolutionary algorithm is used to generate an instant decision map, the framework is used to combine integrated multi-agent reinforcement learning to enable multiple micro-robots to collaborate with each other to complete complex tasks, the complex tasks include group obstacle avoidance and collaborative transportation, and simulate opponents or unforeseen environmental factors through adversarial training elements to generate an instant decision map, the instant decision map is used to guide the micro-robot to make the best decision, including:

[0026] Based on the updated three-dimensional dynamic probability map, the decision framework combining deep reinforcement learning and evolutionary algorithm is used to automatically adjust the parameters of the environmental state change prediction model to obtain an optimized decision framework;

[0027] Based on the optimized decision-making framework, multi-agent reinforcement learning is introduced to enable multiple micro-robots to collaborate with each other to complete complex tasks, including group obstacle avoidance and collaborative transportation. Each micro-robot not only considers its own behavior strategy, but also shares information with other micro-robots to jointly develop the best action plan to generate a collaborative behavior model.

[0028] Based on the collaborative behavior pattern, an adversarial network is generated using adversarial training elements to model and simulate opponents or unforeseen environmental factors, enhance the collaborative ability of the micro-robot, and generate a collaboratively enhanced model;

[0029] The instant decision map is generated according to the optimized decision framework and the collaboratively enhanced model.

[0030] Optionally, the position of the micro robot is estimated by using a Kalman filter to obtain the position state of the micro robot, and based on the position state and the adaptive behavior rule set, the speed and steering angle of the drive motor of the micro robot are adjusted to ensure that the actual motion trajectory conforms to the ideal motion trajectory, and the trajectory tracking of the micro robot is generated. The trajectory tracking refers to the process in which the micro robot moves according to a predetermined motion trajectory during the execution of a task, and adjusts its own speed and steering angle in real time to minimize the deviation between the actual motion trajectory and the ideal trajectory, including:

[0031] Using the Kalman filter to estimate the position of the environmental state of the micro robot, obtain the precise position state of the micro robot, and based on the precise position state, adjust the speed and steering angle of the drive motor to ensure that the actual motion trajectory conforms to the ideal motion trajectory, and obtain the adjustment result of the position state;

[0032] According to the adjustment result of the position state, visual servo control processing is performed on the micro robot to obtain the actual motion trajectory after real-time adjustment, to ensure that the ideal motion trajectory can be achieved even when the environmental state changes, and to obtain trajectory tracking data of the micro robot;

[0033] Based on the trajectory tracking data, the difference between the actual motion trajectory and the ideal motion trajectory is compared to generate an optimized motion pattern. The optimized motion pattern is used to improve the motion control method of the micro robot to obtain the trajectory tracking of the micro robot.

[0034] Optionally, the step of performing visual servo control processing on the micro robot according to the adjustment result of the position state to obtain the actual motion trajectory adjusted in real time, ensuring that the ideal motion trajectory can be achieved even when the environmental state changes, and obtaining trajectory tracking data of the micro robot includes:

[0035] Using the adjustment result of the position state, visual servo control processing is performed on the micro robot, and the actual motion trajectory is adjusted in real time by introducing image processing technology and computer vision algorithm to obtain the adjusted actual motion trajectory;

[0036] According to the adjusted actual motion trajectory, combined with multi-sensor fusion technology, the accuracy of position estimation is further improved, so that the micro robot can achieve sub-centimeter positioning accuracy in a complex environment, and based on the high-precision position information, the speed and steering angle of the drive motor are fine-tuned to generate optimized motion instructions;

[0037] Based on the optimized motion instructions, adaptive control theory and predictive control algorithms are applied to calculate and compensate in advance for factors that may affect the actual motion trajectory, so that the microrobot can move smoothly along the ideal motion trajectory, and obtain the trajectory tracking data of the microrobot. Based on the trajectory tracking data, combined with the target tracking technology in deep learning, the travel route of the microrobot is continuously monitored and corrected to ensure that the travel route meets the task requirements.

[0038] Optionally, based on the updated three-dimensional dynamic probability map, a decision framework combining deep reinforcement learning and evolutionary algorithm is used to generate an instant decision map, the framework is used to combine integrated multi-agent reinforcement learning to enable multiple micro-robots to collaborate with each other to complete complex tasks, the complex tasks include group obstacle avoidance and collaborative transportation, and simulate opponents or unforeseen environmental factors through adversarial training elements to generate an instant decision map, the instant decision map is used to guide the micro-robot to make the best decision, including:

[0039] Based on the updated three-dimensional dynamic probability map, the decision framework combining deep reinforcement learning and evolutionary algorithm is used to automatically adjust the parameters of the environmental state change prediction model to obtain an optimized decision framework;

[0040] The parameters of the environmental state change prediction model include a smoothing coefficient, a dynamically adjusted smoothing coefficient η t Calculated by the following formula:

[0041]

[0042] Among them, η t represents the dynamically adjusted smoothing coefficient, k represents the speed of controlling the change of the smoothing coefficient, Δs t represents the change in the environmental state between the current moment and the previous moment, and b represents the threshold value, which is used to determine whether the environmental change is significant;

[0043] Based on the optimized decision-making framework, multi-agent reinforcement learning is introduced to enable multiple micro-robots to collaborate with each other to complete complex tasks, including group obstacle avoidance and collaborative transportation. Each micro-robot not only considers its own behavior strategy, but also shares information with other micro-robots to jointly develop the best action plan to generate a collaborative behavior model.

[0044] Among them, the behavior strategy of the micro robot considering itself includes a parameter local value weight w local , and the information shared by the micro robot and other robots includes a parameter global value weight w global , local value weight w localand the global value weight w global Calculated by the following formula:

[0045]

[0046] Among them, w local Indicates the definition of local value weight, w global represents the global value weight, d(s,s0) represents the distance measure between the current state s and the initial state s0, T represents the total task duration, and t represents the current time;

[0047] Based on the collaborative behavior pattern, an adversarial network is generated by using adversarial training elements to model and simulate opponents or unforeseen environmental factors. An exponential factor is included in the simulated reference data to enhance the collaborative ability of the micro-robot and generate a collaboratively enhanced model. The exponential factor γ is calculated by the following formula:

[0048]

[0049] Among them, γ represents the introduction of exponential factor, α represents a constant that controls the base value of the exponential factor, β represents a constant that controls the growth rate of the exponential factor, C represents the current environment complexity assessment value, and C0 represents the benchmark complexity assessment value;

[0050] The instant decision map is generated according to the optimized decision framework and the collaboratively enhanced model; wherein the instant decision map is calculated by the following formula:

[0051] M t+1 (s,a)=(1-η t )M t (s,a)+η t ·(w local Q local (s,a)+w global Q global (s,a)) γ

[0052] Among them, M t+1 (s,a) represents the updated value of taking action a in state s in the instant decision map at time t+1, where s represents the state, a represents the action, and M t (s,a) represents the value of taking action a in state s in the instant decision map at time t, Q local (s,a) represents the local value function estimated by fast random tree or other local search algorithm, Q global (s,a) represents the global value function estimated by a Bayesian network or other global optimization algorithm.

[0053] Optionally, the step of performing visual servo control processing on the micro robot according to the adjustment result of the position state to obtain the actual motion trajectory adjusted in real time, ensuring that the ideal motion trajectory can be achieved even when the environmental state changes, and obtaining trajectory tracking data of the micro robot includes:

[0054] The adjustment result of the position state is used to perform visual servo control processing on the micro robot, and the actual motion trajectory is adjusted in real time by introducing image processing technology and computer vision algorithm to obtain the adjusted actual motion trajectory; wherein the visual servo control updates u t+1 Calculated by the following formula:

[0055]

[0056] Among them, u t+1 represents the control input of the visual servo control update at time t+1, u t represents the control input at time t, K v represents the visual servo controller gain matrix, which is used to adjust the control force, x d,t represents the desired target position or feature point, x t Represents the actual position or feature point at the current moment, α t Represents the adaptive learning rate, which is dynamically adjusted according to the rate of change of the environment. Represents the gradient of the loss function with respect to the control input, which is used to optimize the control strategy;

[0057] According to the adjusted actual motion trajectory, combined with multi-sensor fusion technology, the accuracy of position estimation is further improved, so that the micro robot can achieve sub-centimeter positioning accuracy in complex environments, and based on the high-precision position information, the speed and steering angle of the drive motor are fine-tuned to generate optimized motion instructions; wherein the fused high-precision position estimation p t and drive motor fine-tuning v t+1 Calculated by the following formula:

[0058] Among them, p t represents the high-precision position estimate after fusion, N s represents the number of sensors, w i (t) represents the dynamic weight of the i-th sensor, reflecting the credibility of the sensor data, which is dynamically adjusted according to the sensor error and environmental conditions. i,t represents the measurement value of the i-th sensor at time t, p t represents the high-precision position estimate after fusion, v trepresents the driving motor speed and steering angle of the driving motor at time t, v t+1 represents the driving motor speed and steering angle of the driving motor at time t+1, K p Represents the proportional control gain matrix, which is used to adjust the fine-tuning strength, p d,t represents the desired target position, p t represents the high-precision position estimate at the current moment, β t represents the compensation coefficient, which is dynamically adjusted according to task requirements, σ act,t and σ des,t Represents the standard deviation of actual and expected position uncertainty, used to compensate for position error;

[0059] Based on the optimized motion instructions, the adaptive control theory and predictive control algorithm are applied to calculate and compensate in advance the factors that may affect the actual motion trajectory, so that the micro robot can move smoothly along the ideal motion trajectory, and obtain the trajectory tracking data of the micro robot; wherein the adaptive control updates θ t+1 and predictive compensation c t+1 Calculated by the following formula:

[0060]

[0061] Among them, θ t represents the adaptive controller parameters at time t, θ t+1 represents the adaptive controller parameters at time t+1, α represents the learning rate, which controls the step size of each update, Represents the gradient of the loss function with respect to the parameters, which is used to guide the direction of parameter update, γ t represents the evolution factor, which is dynamically adjusted according to the task difficulty and environment complexity, Δθ t represents the change in the best hyperparameter combination searched by the evolutionary algorithm, c t represents the compensation amount at time t, c t+1 represents the compensation amount at time t+1, β represents the compensation coefficient, which controls the compensation strength, and f pred,t represents the predicted disturbance force, f act,t represents the actual interference force encountered, δ t represents an exponential factor, which is dynamically adjusted according to the complexity of the environment to enhance the focus on key interference;

[0062] Based on the trajectory tracking data, combined with the target tracking technology in deep learning, the route of the micro robot is continuously monitored and corrected to make the route meet the task requirements; wherein, the target tracking update and the incremental learning update are calculated by the following formula:

[0063]

[0064] Among them, x track,t+1 represents the tracking state vector of the target tracking update at time t+1, x track,t represents the tracking state vector at time t, K t represents the Kalman gain, which is used to adjust the correction strength of the tracking error, y t represents the observed value, h() represents the observation model, λ t Adaptive adjustment coefficient, dynamically adjusted according to the tracking error, maps the state vector to the observation space, ρ t represents the exponential decay factor, which is used to balance the impact of long-term and short-term tracking errors, e t represents the tracking error at the current moment, w t+1 represents the incremental learning update of the behavior rule weight at time t+1, w t represents the weight of the behavior rule at time t, η represents the learning rate, which controls the step size of each update, Represents the gradient of the loss function with respect to the weights, based on the current dataset Calculate μ t represents the incremental learning adjustment coefficient, which is dynamically adjusted according to the progress of the task, σ err,t and σ ref Represents the standard deviation of the error between the current and reference behavior rules, which is used to optimize the learning process.

[0065] In a second aspect, an embodiment of the present invention provides a micro robot motion control system, comprising:

[0066] A collection and processing module, used to collect current environmental status information in real time using a non-contact sensor, and synchronize and calibrate the environmental status information to generate a comprehensive perception image;

[0067] A classification prediction module, used to apply target detection and recognition technology to analyze the comprehensive perception image to determine the objects existing in the current environment and the categories of the objects, obtain classification results, and build an environment state change prediction model based on the classification results and the environment state information. The environment state change prediction model is used to predict the environment state changes that affect the motion trajectory of the micro robot in the future period of time through historical environment state information and the current environment state information, combined with the classification results;

[0068] An optimization processing module, used to generate candidate paths using a fast random tree according to the comprehensive perception image and the environmental state changes predicted by the environmental state change prediction model, and optimize the candidate paths in combination with an AI search algorithm to obtain a target path;

[0069] An update generation module is used to construct a three-dimensional dynamic probability map based on the target path using a Bayesian network, wherein the three-dimensional dynamic probability map uses the environmental state at a specific location as a node and the transition probability between the environmental states as an edge, and continuously updates the three-dimensional dynamic probability map using a reinforcement learning algorithm to generate an instant decision map;

[0070] A simulation adjustment module is used to adjust the motion trajectory of the micro robot when a change in the environmental state is detected, based on the instant decision map and introducing an emotional computing model to simulate emotional factors in the human decision-making process, so as to generate an adaptive behavior rule set;

[0071] The position adjustment module is used to use a Kalman filter to estimate the position of the micro robot, obtain the position state of the micro robot, and adjust the speed and steering angle of the drive motor of the micro robot based on the position state and the adaptive behavior rule set to ensure that the actual motion trajectory conforms to the ideal motion trajectory, and generate the trajectory tracking of the micro robot. The trajectory tracking refers to the process in which the micro robot moves according to a predetermined motion trajectory during the execution of a task, and adjusts its own speed and steering angle in real time to minimize the deviation between the actual motion trajectory and the ideal trajectory.

[0072] In an embodiment of the present invention, a non-contact sensor is used to collect current environmental state information in real time, and the environmental state information is synchronized and calibrated to generate a comprehensive perception image; target detection and recognition technology is used to analyze the comprehensive perception image to determine the objects existing in the current environment and the categories of the objects, obtain a classification result, and build an environmental state change prediction model based on the classification result and the environmental state information. The environmental state change prediction model is used to predict environmental state changes that affect the motion trajectory of the micro robot in the future through historical environmental state information and the current environmental state information, combined with the classification result; based on the comprehensive perception image and the environmental state changes predicted by the environmental state change prediction model, a fast random tree is used to generate candidate paths, and the candidate paths are optimized in combination with an AI search algorithm to obtain a target path; based on the target path, a three-dimensional dynamic probability map is constructed using a Bayesian network, and the three-dimensional dynamic probability map is characterized by The environmental state at a fixed position is a node, the transition probability between the environmental states is an edge, and the reinforcement learning algorithm is used to continuously update the three-dimensional dynamic probability map to generate an instant decision map; when a change in the environmental state is monitored, based on the instant decision map, the emotional computing model is introduced to simulate the emotional factors in the human decision-making process, and the motion trajectory of the micro robot is adjusted to generate an adaptive behavior rule set; the position of the micro robot is estimated by using a Kalman filter to obtain the position state of the micro robot, and based on the position state and the adaptive behavior rule set, the speed and steering angle of the drive motor of the micro robot are adjusted to ensure that the actual motion trajectory conforms to the ideal motion trajectory, and the trajectory tracking of the micro robot is generated. The trajectory tracking refers to the process in which the micro robot moves according to the predetermined motion trajectory during the execution of the task, and adjusts its own speed and steering angle in real time to minimize the deviation between the actual motion trajectory and the ideal trajectory.

[0073] The technical solution of the present invention has the following beneficial effects:

[0074] The present invention uses a non-contact sensor to collect current environmental state information in real time, ensuring the accuracy and real-time nature of the data. By synchronizing and calibrating the environmental state information, a comprehensive perception image is generated, ensuring the consistency and reliability of multi-source data. The target detection and recognition technology is applied to analyze the comprehensive perception image, determine the objects and their categories in the environment, and improve the ability to understand the environment. Based on the classification results and environmental state information, an environmental state change prediction model is constructed to predict environmental changes that affect the robot's motion trajectory in the future, thereby enhancing the forward-looking nature of the system. A fast random tree (RRT) is used to generate candidate paths, and the candidate paths are optimized in combination with an AI search algorithm to obtain the optimal target path. This method not only improves the speed of path planning, but also ensures the quality of the path. A three-dimensional dynamic probability map is constructed using a Bayesian network, and is continuously updated through a reinforcement learning algorithm to generate an instant decision map. This dynamic update mechanism enables the robot to flexibly respond to complex and changeable environments. When changes in the environmental state are detected, an emotional computing model is introduced to simulate the emotional factors in the human decision-making process, adjust the robot's motion trajectory, and generate an adaptive set of behavior rules. This increases the flexibility and adaptability of the robot, enabling it to make more reasonable decisions in an uncertain environment. The Kalman filter is used to estimate the position of the micro robot and obtain the accurate position state. Based on this, the speed and steering angle of the drive motor are adjusted to ensure that the actual motion trajectory conforms to the ideal motion trajectory. By adjusting its own speed and steering angle in real time, the deviation between the actual motion trajectory and the ideal trajectory is reduced, ensuring that the robot moves according to the predetermined motion trajectory, thereby improving the accuracy of task execution.

[0075] Furthermore, by comparing the deviation between the actual motion trajectory and the ideal motion trajectory, effective behavior rules are determined and incorporated into the motion control method library. This allows the robot to use these experiences in future tasks, reduce trajectory deviations, and improve the accuracy of path execution. The system can continuously optimize the set of behavior rules based on the results of each task, making the robot more adaptable when facing different environments and able to quickly adjust strategies to cope with new situations. Storing effective behavior rules in the method library provides a reference for subsequent tasks, which helps to continuously improve and optimize long-term performance.

[0076] Furthermore, the use of fast random tree or variant algorithms combined with AI search algorithms for path planning and optimization ensures that the target path is not only feasible but also optimal, improving the quality and efficiency of path planning. Based on the three-dimensional dynamic probability map, the environmental changes are predicted and processed through spatiotemporal correlation analysis, which enhances the robot's understanding and prediction of future environmental conditions. The three-dimensional dynamic probability map is continuously updated using the reinforcement learning algorithm to generate an instant decision map, helping the robot make fast and reasonable decisions in complex environments, improving the quality and flexibility of decision-making.

[0077] These and other aspects of the present invention will become more apparent from the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0079] Figure 1 A flowchart of a micro robot motion control method provided by an embodiment of the present invention;

[0080] Figure 2 A schematic structural diagram of a micro robot motion control system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0081] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0082] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0083] Figure 1 A flowchart of a micro robot motion control method is provided for an embodiment of the present invention, such as Figure 1 As shown, the method includes:

[0084] Based on this, the present invention provides a micro robot motion control method, such as Figure 1 ,include:

[0085] Step 101: Use a non-contact sensor to collect current environmental status information in real time, and synchronize and calibrate the environmental status information to generate a comprehensive perception image;

[0086] In this step, non-contact sensors include LiDAR, cameras, infrared sensors, etc. These sensors can obtain environmental data without directly contacting objects. For example, LiDAR determines the distance of an object by emitting a laser beam and measuring the reflection time; the camera captures the visual image.

[0087] Environmental status information: refers to the physical properties of objects in the environment, such as position, shape, color, temperature, and external factors such as lighting conditions and weather conditions.

[0088] Synchronization and calibration processing: Ensure that data from different sensors are consistent in time and space. This usually involves operations such as timestamp alignment and coordinate system conversion in order to fuse multi-source data into a unified perception picture.

[0089] In the embodiment of the present invention, in industrial automation, a micro robot can use a combination of laser radar and camera to scan obstacles and other robots in the working area in real time to ensure safe navigation. A home service robot can detect heat sources through infrared sensors and identify family members with cameras to provide personalized services.

[0090] Step 102: Analyze the comprehensive perception image using target detection and recognition technology to determine the objects existing in the current environment and the categories of the objects, obtain a classification result, and construct an environment state change prediction model based on the classification result and the environment state information. The environment state change prediction model is used to predict the environment state changes that affect the motion trajectory of the micro robot in the future through historical environment state information and the current environment state information in combination with the classification result;

[0091] In this step, object detection and recognition technology: computer vision algorithms (such as convolutional neural networks (CNNs)) are used to identify specific objects and their categories from the comprehensive perception images. For example, different types of objects such as people, furniture, and equipment can be distinguished.

[0092] Classification result: refers to the specific category label of each object obtained after target detection.

[0093] Environmental state change prediction model: Based on historical and current environmental state information, combined with classification results, it predicts possible environmental changes in the future. This helps plan routes or adjust behavior strategies in advance.

[0094] In the embodiments of the present invention, the medical surgery assisting robot can identify surgical tools, patient organs, etc. in the operating room and predict the doctor's operation actions, thereby optimizing its own position to avoid interference. The environmental monitoring robot can identify the health status of crops, predict the occurrence trend of pests and diseases, and provide support for precision agriculture.

[0095] Step 103: Generate candidate paths using a fast random tree according to the comprehensive perception image and the environmental state change predicted by the environmental state change prediction model, and optimize the candidate paths in combination with an AI search algorithm to obtain a target path;

[0096] In this step, Rapid Random Tree (RRT): An algorithm for path planning that can quickly find a feasible path in a complex environment. It does this by randomly sampling points in space and gradually expanding a tree structure.

[0097] AI search algorithms: Traditional search algorithms such as A*, Dijkstra, or more advanced deep learning methods are used to evaluate and select the best path.

[0098] Target path: The final path after optimization takes into account the information provided by the environmental change prediction model to ensure the safety and efficiency of the path.

[0099] In the embodiment of the present invention, the logistics distribution robot can use RRT to quickly generate multiple candidate paths in a dynamic human flow environment, and select the shortest and best path that avoids crowded areas through an AI search algorithm. The disaster rescue robot can adjust the path planning in real time according to the changes in the ruins structure to ensure that it can reach the destination safely.

[0100] Step 104: Based on the target path, a three-dimensional dynamic probability graph is constructed using a Bayesian network, wherein the three-dimensional dynamic probability graph uses the environmental state at a specific location as a node and the transition probability between the environmental states as an edge, and the three-dimensional dynamic probability graph is continuously updated using a reinforcement learning algorithm to generate an instant decision map;

[0101] In this step, a Bayesian network: a probabilistic graphical model is used to represent the dependencies between variables. Here, nodes represent the state of the environment at a specific location, and edges represent the transition probabilities between states.

[0102] Three-dimensional dynamic probability map: The environment is modeled as a series of interrelated states, each of which may change, forming a dynamic probability distribution map.

[0103] Real-time decision map: A map that guides the robot to make the best decision based on the current state of the environment and predicted changes.

[0104] In the embodiment of the present invention, the self-driving car can use the three-dimensional dynamic probability map to predict the behavior of other vehicles, so as to adjust the driving route in advance to avoid potential collisions. The household cleaning robot can update the instant decision map in real time according to the changes in the room layout to ensure efficient coverage of all areas that need to be cleaned.

[0105] Step 105: When a change in the environmental state is detected, based on the instant decision map, an emotional computing model is introduced to simulate emotional factors in the human decision-making process, and the motion trajectory of the micro robot is adjusted to generate an adaptive behavior rule set;

[0106] In this step, the instant decision map is created: a map that guides the robot to make the best decision based on the current environment state and predicted changes.

[0107] Affective computing model: Imitating human emotional response mechanisms, enabling robots to make decisions similar to humans when faced with emergencies, such as giving priority to safe paths in emergencies.

[0108] Adaptive behavior rule set: Action guidelines adjusted according to new environmental states and real-time decision maps, ensuring that the robot can flexibly respond to various changes.

[0109] In the embodiment of the present invention, the medical care robot can quickly adjust its behavior rules to prioritize the safety of the patient when it finds that the patient is suddenly unwell. The service robot can adjust its service route during busy hours in the restaurant to prioritize customer needs while avoiding crowded areas.

[0110] Step 106: Use a Kalman filter to estimate the position of the micro robot to obtain the position state of the micro robot, and based on the position state and the adaptive behavior rule set, adjust the speed and steering angle of the drive motor of the micro robot to ensure that the actual motion trajectory conforms to the ideal motion trajectory, and generate the trajectory tracking of the micro robot. The trajectory tracking refers to the process in which the micro robot moves according to the predetermined motion trajectory during the execution of the task, and adjusts its own speed and steering angle in real time to minimize the deviation between the actual motion trajectory and the ideal trajectory.

[0111] In this step, the Kalman filter: a recursive algorithm used to estimate the state of a system, especially in a noisy environment. It combines data from different sensors to provide a more accurate position estimate.

[0112] Position state: refers to the robot's current exact position and posture.

[0113] Trajectory tracking: Ensure that the robot moves along the predetermined trajectory and adjust the speed and steering angle in real time to minimize the deviation between the actual trajectory and the ideal trajectory.

[0114] In the embodiments of the present invention, industrial automation robots can accurately complete tasks on assembly lines and maintain stable trajectory tracking even when running at high speeds. Indoor delivery robots can use Kalman filters to accurately locate themselves, ensuring that items can be delivered to designated locations accurately every time.

[0115] In summary, it can be seen that the method of the present invention not only covers the entire process from environmental perception, target recognition, path planning to behavior adjustment, but also particularly emphasizes the importance of intelligent decision-making and adaptive capabilities, providing strong technical support for the efficient operation of micro-robots in complex and changing environments.

[0116] Optionally, the steps also include:

[0117] After the micro-robot completes a path planning and executes the trajectory tracking, the deviation between the actual motion trajectory and the ideal motion trajectory is compared to obtain a comparison result; based on the comparison result, an effective behavior rule is determined from the adaptive behavior rule set, and the effective behavior rule refers to a behavior rule that can reduce the deviation between the actual motion trajectory and the ideal trajectory; the effective behavior rule is incorporated into the motion control method library of the micro-robot, so that the effective behavior rules in the motion control method library can be used to realize the motion control of the micro-robot in future tasks.

[0118] In this step, the key concepts involved in this solution include: actual motion trajectory, ideal motion trajectory, deviation comparison results, adaptive behavior rule set, and motion control method library. The actual motion trajectory refers to the path that the micro robot actually travels during the execution of the task; the ideal motion trajectory is the pre-planned optimal path that the robot is expected to follow. The deviation comparison result is a quantitative analysis of the difference between the two, which is used to evaluate the execution accuracy of the robot. The adaptive behavior rule set is a set of dynamically adjusted behavioral guidelines that can automatically optimize the robot's action strategy according to environmental changes. The motion control method library is a database that stores effective behavior rules. These rules have been verified to reduce the deviation between the actual motion trajectory and the ideal trajectory, and can be called in future tasks to improve performance.

[0119] In an embodiment of the present invention, after the micro robot completes a path planning and performs trajectory tracking, the system will automatically compare the deviation between the actual motion trajectory and the ideal motion trajectory, and generate detailed comparison results. Based on these comparison results, the algorithm will filter out effective behavior rules that can significantly reduce the deviation from the adaptive behavior rule set. These rules are then incorporated into the robot's motion control method library so that they can be used as a reference in similar tasks in the future to help the robot perform path planning and trajectory tracking more accurately. By continuously accumulating and applying effective behavior rules, the robot can gradually optimize its motion control strategy and improve overall operating efficiency and accuracy.

[0120] In an application scenario of a smart logistics warehouse, a micro-carrying robot is responsible for transporting goods from one location to another. After each transportation task is completed, the system automatically records and analyzes the deviation between the path actually traveled by the robot (actual motion trajectory) and the preset ideal path (ideal motion trajectory). For example, if the robot deviates slightly from the predetermined route when encountering an obstacle, the system calculates the specific value of this deviation and compares it with the previously stored behavior rules. If a specific behavior rule has effectively reduced similar deviations in the past, then this rule will be marked as an "effective behavior rule" and added to the robot's motion control method library. When performing the same or similar tasks in the future, the robot can call these verified behavior rules to prevent possible deviations in advance, thereby ensuring more accurate and smooth path execution. Over time, this approach enables the robot to gradually learn and improve its behavior patterns in various complex environments, ultimately achieving higher task completion rates and lower operation errors.

[0121] Optionally, based on the target path described in step 104, a three-dimensional dynamic probability graph is constructed using a Bayesian network, the three-dimensional dynamic probability graph uses the environmental state at a specific location as a node, and the transition probability between the environmental states as an edge, and the three-dimensional dynamic probability graph is continuously updated using a reinforcement learning algorithm to generate an instant decision map, including:

[0122] Based on the comprehensive perception image and the environmental state change prediction model, the actual motion trajectory of the micro-robot is planned and optimized to obtain the target path. The path planning and optimization are performed using a fast random tree or variant algorithm combined with an AI search algorithm. The target path is further refined and optimized to generate an optimal path. According to the optimal path, the environmental state of the specific location and the transition probability are modeled using a Bayesian network to construct a three-dimensional dynamic probability map. Based on the three-dimensional dynamic probability map, the environmental changes are predicted through spatiotemporal correlation analysis. Based on the three-dimensional dynamic probability map, a reinforcement learning algorithm is used to continuously update and generate an instant decision map, so as to learn and optimize the behavior strategy of the micro-robot through the instant decision map to improve the decision quality.

[0123] Optionally, the three-dimensional dynamic probability map is continuously updated and processed by using a reinforcement learning algorithm to generate an instant decision map, including:

[0124] Based on the initial motion mode of the micro-robot, selection processing is performed, the non-contact perception data of the micro-robot is integrated, and the current environmental state is evaluated in combination with the environmental state change prediction model, and the weight of the non-contact perception data is adjusted in real time to obtain an optimized initial environmental evaluation result; whenever a new comprehensive perception image is obtained, the differential privacy protection technology is applied to protect the security of the non-contact perception data and effectively shield sensitive information, so that the security of the non-contact perception data is achieved without affecting the accuracy of the three-dimensional dynamic probability map, and the newly obtained comprehensive perception image is combined with the three-dimensional dynamic probability map to update the three-dimensional dynamic probability map to generate an updated three-dimensional dynamic probability map, and the updated three-dimensional dynamic probability map is used to reflect the latest environmental state; based on the updated three-dimensional dynamic probability map, a decision framework combining deep reinforcement learning with evolutionary algorithms is used to generate an instant decision map, and the framework is used to combine integrated multi-agent reinforcement learning to enable multiple micro-robots to cooperate with each other to complete complex tasks, and the complex tasks include group obstacle avoidance and collaborative transportation, and simulate opponents or unforeseen environmental factors through adversarial training elements to generate an instant decision map, and the instant decision map is used to guide the micro-robot to make the best decision.

[0125] Optionally, based on the updated three-dimensional dynamic probability map, a decision framework combining deep reinforcement learning and evolutionary algorithm is used to generate an instant decision map, the framework is used to combine integrated multi-agent reinforcement learning to enable multiple micro-robots to collaborate with each other to complete complex tasks, the complex tasks include group obstacle avoidance and collaborative transportation, and simulate opponents or unforeseen environmental factors through adversarial training elements to generate an instant decision map, the instant decision map is used to guide the micro-robot to make the best decision, including:

[0126] Based on the updated three-dimensional dynamic probability map, the decision framework combining deep reinforcement learning and evolutionary algorithm is used to automatically adjust the parameters of the environmental state change prediction model to obtain an optimized decision framework; based on the optimized decision framework, multi-agent reinforcement learning is introduced to enable multiple micro-robots to collaborate with each other to complete complex tasks, including group obstacle avoidance and collaborative transportation. Each micro-robot not only considers its own behavior strategy, but also shares information with other micro-robots to jointly formulate the best action plan to generate a collaborative behavior pattern; based on the collaborative behavior pattern, an adversarial network is generated using adversarial training elements to model and simulate opponents or unforeseen environmental factors, enhance the collaborative ability of the micro-robots, and generate a collaboratively enhanced model; based on the optimized decision framework and the collaboratively enhanced model, the real-time decision map is generated.

[0127] In this step, several key technical concepts involved in this solution include: three-dimensional dynamic probability graph, Bayesian network, reinforcement learning algorithm, rapid random tree (RRT), differential privacy protection technology, deep reinforcement learning, evolutionary algorithm, multi-agent reinforcement learning and adversarial training elements. The three-dimensional dynamic probability graph is a model used to represent the environmental state and its transition probability at different locations in the environment; the Bayesian network is used to model the conditional dependency between nodes, and here it is used to model the environmental state and transition probability at a specific location. The reinforcement learning algorithm learns the best behavior strategy by interacting with the environment; the rapid random tree is a path planning algorithm, which can further refine and optimize the path in combination with the AI ​​search algorithm. Differential privacy protection technology ensures the security and privacy protection of contactless perception data without affecting the accuracy of the three-dimensional dynamic probability graph. Deep reinforcement learning is a method that combines deep learning with reinforcement learning, while the evolutionary algorithm is an optimization algorithm based on natural selection and genetic mechanisms. Multi-agent reinforcement learning enables multiple robots to collaborate with each other to complete tasks; adversarial training elements are used to simulate opponents or unforeseen environmental factors to enhance the robot's adaptability.

[0128] In an embodiment of the present invention, firstly, according to the comprehensive perception image and the prediction model of the environmental state change, the actual motion trajectory of the micro-robot is planned and optimized using a fast random tree or its variant algorithm combined with an AI search algorithm to generate the optimal path. Then, the environmental state and transition probability of a specific location are modeled using a Bayesian network, a three-dimensional dynamic probability map is constructed, and environmental changes are predicted through spatiotemporal correlation analysis. Next, the three-dimensional dynamic probability map is continuously updated through a reinforcement learning algorithm to generate an instant decision map, thereby learning and optimizing the robot's behavior strategy and improving the quality of decision-making. In order to ensure data security, differential privacy protection technology is used to process non-contact perception data. Finally, through a framework combining deep reinforcement learning with evolutionary algorithms, combined with integrated multi-agent reinforcement learning, multiple micro-robots can collaborate with each other to complete complex tasks, such as group obstacle avoidance and collaborative transportation, and opponents or unforeseen environmental factors are simulated through adversarial training elements to generate an instant decision map to guide the robot to make the best decision.

[0129] In an intelligent logistics warehouse, a group of miniature handling robots equipped with sensors such as lidar and cameras need to efficiently complete the task of transporting goods. These robots first use a fast random tree algorithm combined with an A* search algorithm to plan an optimal path from the starting point to the end point, taking into account the possible existence of other robots. Subsequently, they use a Bayesian network to build a three-dimensional dynamic probability map, which not only reflects the current status of each location in the warehouse, but also predicts the trend of change in the future. Whenever a new perception image arrives, this probability map will be updated instantly, and the robot's behavior strategy will be continuously optimized through a reinforcement learning algorithm. In order to protect sensitive information, all data collected by sensors have been processed with differential privacy. In addition, through a framework that combines deep reinforcement learning with evolutionary algorithms, robots can share information and coordinate work, such as jointly deciding the best detour route when encountering obstacles, or cooperating to improve efficiency when handling large goods. In order to better cope with unforeseen situations, the system also introduces adversarial training elements to simulate possible problems, so that the robot team can maintain efficient operation in any situation.

[0130] Optionally, according to the position estimation of the micro robot using the Kalman filter as described in step 106, the position state of the micro robot is obtained, and based on the position state and the adaptive behavior rule set, the speed and steering angle of the drive motor of the micro robot are adjusted to ensure that the actual motion trajectory conforms to the ideal motion trajectory, and the trajectory tracking of the micro robot is generated. The trajectory tracking refers to the process in which the micro robot moves according to the predetermined motion trajectory during the execution of the task, and adjusts its own speed and steering angle in real time to minimize the deviation between the actual motion trajectory and the ideal trajectory, including:

[0131] The Kalman filter is used to estimate the position of the environmental state of the micro robot to obtain the precise position state of the micro robot, and based on the precise position state, the speed and steering angle of the drive motor are adjusted to ensure that the actual motion trajectory conforms to the ideal motion trajectory, and the adjustment result of the position state is obtained; according to the adjustment result of the position state, the micro robot is visually servoed and controlled to obtain the actual motion trajectory after real-time adjustment to ensure that the ideal motion trajectory can be achieved even when the environmental state changes, and the trajectory tracking data of the micro robot is obtained; based on the trajectory tracking data, the difference between the actual motion trajectory and the ideal motion trajectory is compared to generate an optimized motion pattern, and the optimized motion pattern is used to improve the motion control method of the micro robot to obtain the trajectory tracking of the micro robot.

[0132] Optionally, the visual servo control processing is performed on the micro robot according to the adjustment result of the position state to obtain the actual motion trajectory after real-time adjustment, to ensure that the ideal motion trajectory can be achieved even when the environmental state changes, and to obtain the trajectory tracking data of the micro robot, including:

[0133] The micro robot is subjected to visual servo control processing by using the adjustment result of the position state, and the actual motion trajectory is adjusted in real time by introducing image processing technology and computer vision algorithm to obtain the adjusted actual motion trajectory; according to the adjusted actual motion trajectory, combined with multi-sensor fusion technology, the accuracy of position estimation is further improved, so that the micro robot can achieve sub-centimeter positioning accuracy in a complex environment, and based on the high-precision position information, the speed and steering angle of the drive motor are fine-tuned to generate optimized motion instructions; based on the optimized motion instructions, adaptive control theory and predictive control algorithm are applied to calculate and compensate in advance the factors that may affect the actual motion trajectory, so that the micro robot can move smoothly along the ideal motion trajectory, and the trajectory tracking data of the micro robot is obtained, and based on the trajectory tracking data, combined with the target tracking technology in deep learning, the travel route of the micro robot is continuously monitored and corrected to make the travel route meet the task requirements.

[0134] In this step, the key concepts involved in this solution include: Kalman filter, position estimation, adaptive behavior rule set, visual servo control, multi-sensor fusion technology, adaptive control theory, predictive control algorithm and target tracking technology. Kalman filter is a recursive statistical estimation algorithm used to estimate the state of the system from a series of incomplete and noisy measurements; position estimation refers to determining the precise position of the robot in the environment by using sensor data; adaptive behavior rule set is a set of rules that can be automatically adjusted according to environmental changes to guide the robot's actions; visual servo control is the use of visual feedback to adjust the robot's movement in real time to achieve more accurate task execution; multi-sensor fusion technology is to combine data from different sensors to improve the quality and reliability of information; adaptive control theory allows the control system to automatically adjust its performance according to changes in the environment or system parameters; predictive control algorithm is a model-based control method that can calculate and compensate for factors that affect the trajectory of motion in advance; target tracking technology is part of the field of deep learning, which is specifically used to continuously monitor and correct the path of moving objects.

[0135] In the embodiment of the present invention, the Kalman filter is first used to accurately estimate the position of the micro robot to ensure that its position state is accurate. Then, based on the obtained position state and the preset adaptive behavior rule set, the speed and steering angle of the drive motor are dynamically adjusted to make the actual motion trajectory as close to the ideal trajectory as possible. Next, in order to ensure that the ideal motion trajectory can be achieved even when the environmental state changes, visual servo control processing is used to further optimize the actual motion trajectory and obtain trajectory tracking data. Finally, by analyzing the trajectory tracking data, the difference between the actual motion trajectory and the ideal trajectory is compared, and an optimized motion pattern is generated to improve the robot's motion control method. In this process, image processing technology and computer vision algorithms, multi-sensor fusion technology, adaptive control theory, and predictive control algorithms are also combined to ensure that the robot can maintain high-precision positioning and smooth movement in a complex environment, and continuously monitor and correct the route through target tracking technology.

[0136] In a smart agriculture scenario, micro robots are deployed in greenhouses to automatically navigate to designated locations for crop monitoring or irrigation tasks. These robots are equipped with a variety of sensors such as lidar, cameras, and inertial measurement units (IMUs). When the robot starts to move, it first applies a Kalman filter to integrate all sensor data to provide an accurate position estimate to ensure that the robot knows its exact position. Then, based on this position information and pre-programmed behavior rules, the robot adjusts the speed and steering angle of the drive motor to ensure that it moves along the predetermined ideal trajectory. If it encounters an obstacle or the ground conditions change, the robot uses visual servo control to instantly adjust its speed and direction with the visual feedback provided by the camera to maintain the correct path. In addition, the robot also uses multi-sensor fusion technology to combine information from different sensors to obtain higher positioning accuracy. Throughout the process, the robot continuously applies adaptive control theory and predictive control algorithms to respond in advance to various factors that may affect its path. For example, it slows down before approaching a turning point, or plans a detour route in advance when an obstacle is detected ahead. Ultimately, the robot uses target tracking technology to ensure that its route always meets the mission requirements, while continuously improving its movement efficiency and accuracy through continuous learning and optimization.

[0137] In the motion control of micro-robots, especially in complex and changing environments, ensuring that robots can efficiently collaborate to complete tasks (such as group obstacle avoidance and collaborative transportation) is a key challenge. To address this challenge, this solution introduces a framework that combines deep reinforcement learning with evolutionary algorithms, and combines it with an instant decision map generation mechanism to enable multiple robots to collaborate with each other and adapt to environmental changes. The framework optimizes the robot's behavior strategy by dynamically adjusting parameters such as smoothing coefficients, local and global value weights, and exponential factors to achieve optimal decisions.

[0138] Optionally, based on the updated three-dimensional dynamic probability map, a decision framework combining deep reinforcement learning and evolutionary algorithm is used to generate an instant decision map, the framework is used to combine integrated multi-agent reinforcement learning to enable multiple micro-robots to collaborate with each other to complete complex tasks, the complex tasks include group obstacle avoidance and collaborative transportation, and simulate opponents or unforeseen environmental factors through adversarial training elements to generate an instant decision map, the instant decision map is used to guide the micro-robot to make the best decision, including:

[0139] Based on the updated three-dimensional dynamic probability map, the decision framework combining deep reinforcement learning and evolutionary algorithm is used to automatically adjust the parameters of the environmental state change prediction model to obtain an optimized decision framework; wherein the parameters of the environmental state change prediction model include a smoothing coefficient, a dynamically adjusted smoothing coefficient η t Calculated by the following formula:

[0140]

[0141] Among them, η t represents the dynamically adjusted smoothing coefficient, k represents the speed of controlling the change of the smoothing coefficient, Δs t represents the change in the environmental state between the current moment and the previous moment, and b represents the threshold value, which is used to determine whether the environmental change is significant;

[0142] The smoothing coefficient η is calculated by the above t Adjustments are made and multi-agent reinforcement learning is introduced into the optimized decision-making framework, so that multiple micro-robots can collaborate with each other to complete complex tasks, including group obstacle avoidance and collaborative transportation. Each micro-robot not only considers its own behavior strategy, but also shares information with other micro-robots to jointly formulate the best action plan to generate a collaborative behavior pattern; wherein, the behavior strategy of the micro-robot considering itself includes a parameter local value weight w local , and the information shared by the micro robot and other robots includes a parameter global value weight w global , local value weight w localand the global value weight w global Calculated by the following formula:

[0143]

[0144] Among them, w local Indicates the definition of local value weight, w global represents the global value weight, d(s,s0) represents the distance measure between the current state s and the initial state s0, T represents the total task duration, and t represents the current time;

[0145] In the collaborative behavior model, an adversarial training element is used to generate an adversarial network, a model is simulated to simulate an opponent or unforeseen environmental factors, an exponential factor is included in the simulated reference data, the collaborative ability of the micro-robot is enhanced, and a collaboratively enhanced model is generated; wherein the exponential factor γ is calculated by the following formula:

[0146]

[0147] Among them, γ represents the introduction of exponential factor, α represents a constant that controls the base value of the exponential factor, β represents a constant that controls the growth rate of the exponential factor, C represents the current environment complexity assessment value, and C0 represents the benchmark complexity assessment value;

[0148] The instant decision map is generated according to the optimized decision framework and the collaboratively enhanced model; wherein the instant decision map is calculated by the following formula:

[0149] M t+1 (s,a)=(1-η t )M t (s,a)+η t ·(w local Q local (s,a)+w global Q global (s,a)) γ

[0150] Among them, M t+1 (s,a) represents the updated value of taking action a in state s in the instant decision map at time t+1, where s represents the state, a represents the action, and M t (s,a) represents the value of taking action a in state s in the instant decision map at time t, Q local (s,a) represents the local value function estimated by fast random tree or other local search algorithm, Q global (s,a) represents the global value function estimated by a Bayesian network or other global optimization algorithm.

[0151] The purpose of the overall formula is to create a flexible and efficient decision-making system that can automatically adjust its behavior strategy according to changes in the environment. Used to smoothly adjust model parameters to ensure that the decision-making framework can respond quickly to environmental changes without excessive fluctuations. Local and global value weights: Balance the relationship between local optimal solutions and global optimal solutions, so that each robot not only focuses on its own optimal path, but also considers the task progress of the entire team. Enhance sensitivity to environmental complexity to help robots better adapt to unforeseen situations. Combine all the above factors to generate the latest decision map to guide the robot to make the best decision.

[0152] The following is a brief explanation of the design reasons for each item of the formula:

[0153] Dynamically adjusted smoothing factor The change in the environmental state Δs_t reflects the speed of environmental change, and the change is converted into a smoothing coefficient η through the Sigmoid function t , which can effectively control the adjustment speed of model parameters. k controls the speed of change of the smoothing coefficient, and b is used as an intermediate value to determine whether the environmental change is significant.

[0154] Local value weight and global value weight The local value weight emphasizes the priority of actions within a short distance, while the global value weight focuses on long-term goals. The distance metric d(s,s0) measures the distance between the current state and the initial state. Represents the proportion of remaining time. These two indicators together determine the value of each action.

[0155] Exponential Factor The exponential factor increases the focus on highly complex environments. When the environment becomes more complex, γ increases, thereby increasing the importance of certain specific behaviors.

[0156] Instant Decision Map M t+1 (s,a)=(1-η t )M t (s,a)+η t ·(w local Q local (s,a)+w global Q global (s,a)) γ : By fusing local and global value functions and applying exponential factor γ weighting, a new instant decision map is generated to guide the robot to take optimal actions.

[0157] The following is a brief explanation of how to obtain each parameter of the formula:

[0158] Dynamically adjusted smoothing coefficient η t: k and b are hyperparameters that can be set experimentally or based on experience. t Calculated from sensor data, it represents the change in the environmental state between the current moment and the previous moment.

[0159] Local value weight w local and the global value weight w global :d(s,s0) is calculated by Euclidean distance or other appropriate distance metric. T and t are the total task duration and current time, respectively, which are directly obtained from task planning.

[0160] Exponential factor γ: α and β are hyperparameters, which are set through experiments or based on experience. C and C0 are the current environment complexity evaluation value and the baseline complexity evaluation value, respectively, which are usually calculated in real time by the environment perception module.

[0161] Instant Decision Map M t+1 (s,a):Q local (s,a) and Q global (s, a) are estimated by fast random tree and other local search algorithms, as well as global optimization algorithms such as Bayesian networks. t (s,a) is the instantaneous decision map of the previous period, stored in memory.

[0162] Parameter setting and input value:

[0163] Assume that in a smart logistics warehouse, there are 5 micro-carrying robots responsible for transporting goods from point A to point B. In this process, the robots need to avoid other moving objects and coordinate their work to improve efficiency. The following calculations are performed during this process:

[0164] In the parameter setting, set k = 2.0, b = 0.5 (determined based on historical data analysis), set α = 0.5, β = 0.2 (determined based on experimental results), set the total task duration T = 60 seconds, the current time t = 30 seconds, set the distance between the current state s and the initial state s0 d(s, s0) = 5 meters, set the current environment complexity assessment value C = 0.8, and the benchmark complexity assessment value C0 = 0.5.

[0165] For the smoothing coefficient η t :

[0166]

[0167] Assume Δs t =0.7, we get:

[0168]

[0169] For the local value weight w local and the global value weight wglobal Calculated:

[0170]

[0171] Calculating the exponential factor γ yields:

[0172]

[0173] Update instant decision map M t+1 (s,a):

[0174] Assume M t (s,a)=0.8,Q local (s,a)=0.9,Q global (s,a)=0.7

[0175] Calculated:

[0176] M t+1 (s,a)=(1-0.731)×0.8+0.731×(0.167×0.9+0.5×0.7) 0.684 ≈0.727Through the above calculation, the new value M in the instant decision map t+1 (s,a)≈0.727 shows that the value of taking a specific action in the current state has increased, which reflects that the robot optimizes its action strategy after considering environmental changes, local and global value weights, and environmental complexity.

[0177] In the field of robotics, especially for micro-robots, precise control of their motion trajectory is crucial. In order to ensure that the robot can accurately track the predetermined path in a dynamically changing environment and achieve high-precision position estimation and motion control, technologies such as visual servo control, multi-sensor fusion, adaptive control, and target tracking are widely used. These technologies use a series of formulas to describe how to adjust the robot's behavior in real time to adapt to environmental changes and optimize motion performance.

[0178] Optionally, the visual servo control processing is performed on the micro robot according to the adjustment result of the position state to obtain the actual motion trajectory after real-time adjustment, to ensure that the ideal motion trajectory can be achieved even when the environmental state changes, and to obtain the trajectory tracking data of the micro robot, including:

[0179] The adjustment result of the position state is used to perform visual servo control processing on the micro robot. By introducing image processing technology and computer vision algorithm, the actual motion trajectory is adjusted in real time to obtain the adjusted actual motion trajectory; wherein the visual servo control control at time t+1 is expressed as a visual servo control control update, and the visual servo control control update ut+1 Calculated by the following formula:

[0180]

[0181] Among them, u t+1 represents the control input of the visual servo control update at time t+1, u t represents the control input at time t, K v represents the visual servo controller gain matrix, which is used to adjust the control force, x d,t represents the desired target position or feature point, x t Represents the actual position or feature point at the current moment, α t Represents the adaptive learning rate, which is dynamically adjusted according to the rate of change of the environment. Represents the gradient of the loss function with respect to the control input, which is used to optimize the control strategy;

[0182] According to the actual motion trajectory after adjustment, combined with multi-sensor fusion technology, the accuracy of position estimation is further improved, so that the micro robot can achieve sub-centimeter positioning accuracy in complex environments. Based on the high-precision position information, the speed and steering angle of the drive motor are fine-tuned, and the data of the fine-tuning process are integrated into the drive motor fine-tuning to generate optimized motion instructions; wherein, the fused high-precision position estimation p t and drive motor fine-tuning v t+1 Calculated by the following formula:

[0183]

[0184] Among them, p t represents the high-precision position estimate after fusion, N s represents the number of sensors, w i (t) represents the dynamic weight of the i-th sensor, reflecting the credibility of the sensor data, which is dynamically adjusted according to the sensor error and environmental conditions. i,t represents the measurement value of the i-th sensor at time t, p t represents the high-precision position estimate after fusion, v t represents the driving motor speed and steering angle of the driving motor at time t, v t+1 represents the driving motor speed and steering angle of the driving motor at time t+1, K p Represents the proportional control gain matrix, which is used to adjust the fine-tuning strength, p d,t represents the desired target position, p t represents the high-precision position estimate at the current moment, β t represents the compensation coefficient, which is dynamically adjusted according to task requirements, σ act,t and σ des,tRepresents the standard deviation of actual and expected position uncertainty, used to compensate for position error;

[0185] Based on the optimized motion instructions, adaptive control theory and predictive control algorithm are applied, the adaptive control theory includes adaptive control update, the factors that will affect the actual motion trajectory are calculated and compensated in advance, the calculated and compensated data are expressed as predictive compensation, so that the micro robot can move smoothly along the ideal motion trajectory, and the trajectory tracking data of the micro robot is obtained; wherein, the adaptive control θ t+1 and predictive compensation c t+1 Calculated by the following formula:

[0186]

[0187] Among them, θ t represents the adaptive controller parameters at time t, θ t+1 represents the adaptive controller parameters at time t+1, α represents the learning rate, which controls the step size of each update, Represents the gradient of the loss function with respect to the parameters, which is used to guide the direction of parameter update, γ t represents the evolution factor, which is dynamically adjusted according to the task difficulty and environment complexity, Δθ t represents the change in the best hyperparameter combination searched by the evolutionary algorithm, c t represents the compensation amount at time t, c t+1 represents the compensation amount at time t+1, β represents the compensation coefficient, which controls the compensation strength, and f pred,t represents the predicted disturbance force, f act,t represents the actual interference force encountered, δ t represents an exponential factor, which is dynamically adjusted according to the complexity of the environment to enhance the focus on key interference;

[0188] Based on the trajectory tracking data, combined with the target tracking technology in deep learning, the target tracking technology includes target tracking update, continuously monitoring and correcting the route of the micro robot so that the route meets the task requirements. In the monitoring and correction process, the received constantly changing route data is integrated into incremental learning updates; wherein, the target tracking update x track,t+1 and incremental learning updates w t+1 Calculated by the following formula:

[0189]

[0190] Among them, x track,t+1 represents the tracking state vector of the target tracking update at time t+1, x track,t represents the tracking state vector at time t, Kt represents the Kalman gain, which is used to adjust the correction strength of the tracking error, y t represents the observed value, h() represents the observation model, λ t Adaptive adjustment coefficient, dynamically adjusted according to the tracking error, maps the state vector to the observation space, ρ t represents the exponential decay factor, which is used to balance the impact of long-term and short-term tracking errors, e t represents the tracking error at the current moment, w t+1 represents the incremental learning update of the behavior rule weight at time t+1, w t represents the weight of the behavior rule at time t, η represents the learning rate, which controls the step size of each update, Represents the gradient of the loss function with respect to the weights, based on the current dataset Calculate μ t represents the incremental learning adjustment coefficient, which is dynamically adjusted according to the progress of the task, σ err,t and σ ref Represents the standard deviation of the error between the current and reference behavior rules, which is used to optimize the learning process.

[0191] The purpose of the overall formula is to create an intelligent control system that can adapt to complex environmental changes, so that the micro robot can maintain good motion performance even under uncertain conditions. It is used to adjust the actual motion trajectory of the robot in real time based on image processing technology and computer vision algorithms to make it as close to the ideal motion trajectory as possible. Improve the accuracy of position estimation and ensure sub-centimeter positioning accuracy, thereby providing precise fine-tuning instructions for the speed and steering angle of the drive motor. Calculate factors that may affect the motion trajectory in advance and compensate for them so that the robot can move smoothly along the ideal trajectory. Continuously monitor and correct the route to ensure that the task requirements are met, while continuously optimizing the behavior rules through incremental learning.

[0192] The following is a brief explanation of the design reasons for each item of the formula:

[0193] Visual Server Control Update This formula aims to adjust the control input in combination with visual feedback information to minimize the desired position x d,t and the current position x t The difference between them is calculated and the control strategy is optimized by gradient descent method.

[0194] High-precision position estimation and drive motor fine-tuning By fusing data from multiple sensors, the accuracy of position estimation is improved, and the control of the drive motor is fine-tuned accordingly to achieve more precise movements.

[0195] Adaptive Control Update and predictive compensation These two formulas are designed to enable the robot to automatically adjust its own parameters according to environmental changes and make predictive compensation for potential interference factors to ensure stability and response speed.

[0196] Target tracking update x track,t+1 =x track,t +K t (y t -h(x track,t ))+λ t ·exp(-ρ t ||e t ||) and incremental learning updates These two formulas are used to continuously monitor the robot's path and continuously improve its behavior rules through incremental learning to better complete the task.

[0197] The following is a brief explanation of how to obtain each parameter of the formula:

[0198] Visual Server Control Update t+1 :K v is the visual servo controller gain matrix, usually determined experimentally. t The adaptive learning rate can be dynamically adjusted according to the rate of change of the environment. The gradient of the loss function with respect to the control input can be obtained through the back-propagation algorithm.

[0199] High-precision position estimation t and drive motor fine-tuning v t+1 :w i (t) Dynamic weight reflects the credibility of each sensor data and is dynamically adjusted according to sensor error and environmental conditions. i,t is the sensor measurement value, read directly from the sensor. K p is the proportional control gain matrix, which is used to adjust the fine-tuning force and is also set based on experiments or experience. t The compensation coefficient is dynamically adjusted according to the task requirements, and σ act,t and σ des,t represent the actual and expected position uncertainty standard deviations, respectively, to compensate for the position error.

[0200] Adaptive control updates θ t+1 and predictive compensation c t+1 : α The learning rate controls the step size of each update, The gradient of the loss function with respect to the parameters, γ t Evolutionary factors must be based on experimental settings. pred,t and f act,tare the predicted and actual disturbance forces, δ t The exponential factor enhances the focus on key disturbances, which are usually parameters from the environment perception module.

[0201] Target tracking update x track,t+1 and incremental learning updates w t+1 :K t Karl burst gain, λ t Adaptive adjustment coefficient, ρ t Exponential decay factor, η learning rate, μ t Incremental learning adjustment coefficients, all of which are hyperparameters that can be set experimentally or based on experience. t The tracking error at the current moment, y t Observation value, h() observation model, The gradient of the loss function with respect to the weights, based on the current dataset D t Calculate,σ err,t and σ ref The standard deviation of the behavioral rule error, used to optimize the learning process.

[0202] Consider a scenario where a fleet of five drones is performing a reconnaissance mission, during which the drones need to maintain specific relative positions and deal with sudden obstacles or other unforeseen situations.

[0203] Parameter setting and value substitution

[0204] Set K in the parameter setting v =[0.2,0.2],α t =0.1 (based on the rate of environmental change), set N s =3 (using three different types of sensors), w i (t) = [0.4, 0.3, 0.3] (based on sensor credibility), assuming z i,t =[2.5,3.0,2.8]m (each sensor only measured value), set K p =[0.5,0.5],β t =0.05,σ act,t =0.2,σ des,t =0.1, set α=0.01, γ t =0.1,Δθ t =[0.05,-0.03], set f pred,t =0.5N (predicted disturbance force), f act,t =0.6N (actual interference force),δ t =1.5, set K t =[0.7,0.7],λ t =0.2,ρt =0.1,η=0.001,μ t =0.1,σ err,t =0.3,σ ref =0.2.

[0205] Update for visual service control t+1 :

[0206] Assume u t =[0.5,0.5],x d,t =[3.0,3.0],x t =[2.8,2.9],

[0207] Calculated:

[0208] u t+1 =[0.5,0.5]+[0.2,0.2]×([3.0,3.0]-[2.8,2.9])+0.1×[0.1,0.1]

[0209] u t+1 ≈[0.54,0.53]

[0210] For high-precision position estimation, t and drive motor fine-tuning v t+1 :

[0211] Among them, calculate the high-precision position estimate p t =0.4×2.5+0.3×3.0+0.3×2.8=2.77

[0212] Assume v t =[0.6,0.6],p d,t =3.0

[0213] Calculated:

[0214]

[0215] v t+1 ≈[0.67,0.67]

[0216] For adaptive control update θ t+1 and predictive compensation c t+1 :

[0217] Assume θ t =[0.4,0.4],

[0218] Calculated:

[0219] θ t+1=[0.4,0.4]+0.01×[0.05,-0.05]+0.1×[0.05,-0.03]

[0220] θ t+1 ≈[0.405,0.397]

[0221] Assumption c t =0.1

[0222] Calculated:

[0223] c t+1 =0.1+0.05×(0.5-0.6) 1.5

[0224] c t+1 ≈0.097

[0225] For target tracking update x track,t+1 and incremental learning updates w t+1 :

[0226] Assume x track,t =[2.8,2.9],y t =[3.0,3.1],h(x track,t )=[2.9,3.0],e t =[0.1,0.1]

[0227] calculate:

[0228] x track,t+1 =[2.8,2.9]+[0.7,0.7]×([3.0,3.1]-[2.9,3.0])+0.2·exp(-0.1×

[0229] ∥[[0.1,0.1])

[0230] get:

[0231] x track,t+1 ≈[2.87,2.97]

[0232] Assumptions

[0233] calculate:

[0234] get:

[0235] w t+1 ≈[0.514,0.486]

[0236] Through the above calculations, the navigation capability and mission execution efficiency of the drone formation in complex environments are greatly enhanced. It not only ensures the coordination and consistency among the drone groups, but also improves the ability of individual drones to respond to emergencies, thereby ensuring that the entire formation can complete the reconnaissance mission safely and efficiently.

[0237] Figure 2 A schematic diagram of a micro robot motion control system is provided in an embodiment of the present invention. Figure 2 As shown, the system includes:

[0238] The collection and processing module 21 is used to collect the current environment status information in real time using a non-contact sensor, and synchronize and calibrate the environment status information to generate a comprehensive perception image;

[0239] A classification prediction module 22 is used to apply target detection and recognition technology to analyze the comprehensive perception image to determine the objects existing in the current environment and the categories of the objects, obtain classification results, and build an environment state change prediction model based on the classification results and the environment state information. The environment state change prediction model is used to predict the environment state changes that affect the motion trajectory of the micro robot in the future period of time through historical environment state information and the current environment state information, combined with the classification results;

[0240] An optimization processing module 23 is used to generate candidate paths using a fast random tree according to the comprehensive perception image and the environmental state change predicted by the environmental state change prediction model, and optimize the candidate paths in combination with an AI search algorithm to obtain a target path;

[0241] An update generation module 24 is used to construct a three-dimensional dynamic probability map based on the target path using a Bayesian network, wherein the three-dimensional dynamic probability map uses the environmental state at a specific location as a node and the transition probability between the environmental states as an edge, and continuously updates the three-dimensional dynamic probability map using a reinforcement learning algorithm to generate an instant decision map;

[0242] The simulation adjustment module 25 is used to adjust the motion trajectory of the micro robot based on the instant decision map and introduce the emotional computing model to simulate the emotional factors in the human decision-making process when the environmental state changes, so as to generate an adaptive behavior rule set;

[0243] The position adjustment module 26 is used to use a Kalman filter to estimate the position of the micro robot, obtain the position state of the micro robot, and adjust the speed and steering angle of the drive motor of the micro robot based on the position state and the adaptive behavior rule set to ensure that the actual motion trajectory conforms to the ideal motion trajectory, and generate the trajectory tracking of the micro robot. The trajectory tracking refers to the process in which the micro robot moves according to a predetermined motion trajectory during the execution of a task, and adjusts its own speed and steering angle in real time to minimize the deviation between the actual motion trajectory and the ideal trajectory.

[0244] Figure 2 The micro robot motion control system can execute Figure 1 The implementation principle and technical effect of the micro-robot motion control method described in the embodiment are not described in detail. The specific way in which each module and unit performs operations in the micro-robot motion control system in the above embodiment has been described in detail in the embodiment of the method, and will not be described in detail here.

[0245] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A micro robot motion control method, characterized in that: include: Using non-contact sensors to collect current environmental status information in real time, and synchronizing and calibrating the environmental status information to generate a comprehensive perception image; Applying target detection and recognition technology to analyze the comprehensive perception image to determine the objects existing in the current environment and the categories of the objects, obtain a classification result, and construct an environment state change prediction model based on the classification result and the environment state information. The environment state change prediction model is used to predict the environment state changes that affect the motion trajectory of the micro robot in the future period of time through historical environment state information and the current environment state information, combined with the classification result; According to the comprehensive perception image and the environmental state change predicted by the environmental state change prediction model, a candidate path is generated using a fast random tree, and the candidate path is optimized in combination with an AI search algorithm to obtain a target path; Based on the target path, a three-dimensional dynamic probability graph is constructed using a Bayesian network, wherein the three-dimensional dynamic probability graph uses the environmental state at a specific location as a node and the transition probability between the environmental states as an edge, and the three-dimensional dynamic probability graph is continuously updated using a reinforcement learning algorithm to generate an instant decision map; When a change in the environmental state is detected, based on the instant decision map, an emotional computing model is introduced to simulate the emotional factors in the human decision-making process, and the motion trajectory of the micro-robot is adjusted to generate an adaptive behavior rule set; A Kalman filter is used to estimate the position of the microrobot to obtain the position state of the microrobot. Based on the position state and the adaptive behavior rule set, the speed and steering angle of the drive motor of the microrobot are adjusted to ensure that the actual motion trajectory conforms to the ideal motion trajectory, and the trajectory tracking of the microrobot is generated. The trajectory tracking refers to the process in which the microrobot moves according to a predetermined motion trajectory during the execution of a task, and adjusts its own speed and steering angle in real time to minimize the deviation between the actual motion trajectory and the ideal trajectory.

2. The method according to claim 1, characterized in that Also includes: After the micro-robot completes a path planning and performs the trajectory tracking, the deviation between the actual motion trajectory and the ideal motion trajectory is compared to obtain a comparison result; Based on the comparison result, determining an effective behavior rule from the adaptive behavior rule set, wherein the effective behavior rule refers to a behavior rule that can reduce the deviation between the actual motion trajectory and the ideal trajectory; The effective behavior rules are incorporated into the motion control method library of the micro robot, so that the effective behavior rules in the motion control method library can be used to realize the motion control of the micro robot in future tasks.

3. The method according to claim 1, characterized in that Based on the target path, a three-dimensional dynamic probability graph is constructed using a Bayesian network, wherein the three-dimensional dynamic probability graph uses the environmental state at a specific location as a node and the transition probability between the environmental states as an edge, and the three-dimensional dynamic probability graph is continuously updated using a reinforcement learning algorithm to generate an instant decision map, including: Based on the comprehensive perception image and the environmental state change prediction model, the actual motion trajectory of the micro robot is planned and optimized to obtain a target path, and a fast random tree or variant algorithm combined with an AI search algorithm is used for path planning and optimization to further refine and optimize the target path to generate an optimal path; According to the optimal path, the environmental state of the specific location and the transition probability are modeled using a Bayesian network to construct a three-dimensional dynamic probability map, and based on the three-dimensional dynamic probability map, the environmental change is predicted through spatiotemporal correlation analysis; Based on the three-dimensional dynamic probability map, a reinforcement learning algorithm is used to continuously update and generate a real-time decision map, so as to learn and optimize the behavior strategy of the micro robot through the real-time decision map to improve the decision quality.

4. The method according to claim 3, characterized in that Based on the three-dimensional dynamic probability map, the reinforcement learning algorithm is used to continuously update and generate an instant decision map, including: Based on the initial motion mode of the micro-robot, selection processing is performed, the non-contact sensing data of the micro-robot is integrated, and the current environmental state is evaluated in combination with the environmental state change prediction model, and the weight of the non-contact sensing data is adjusted in real time to obtain an optimized initial environmental evaluation result; Whenever a new comprehensive perception image is acquired, differential privacy protection technology is applied to protect the security of the non-contact perception data and effectively shield sensitive information, thereby achieving the security of the non-contact perception data without affecting the accuracy of the three-dimensional dynamic probability map, and combining the newly acquired comprehensive perception image with the three-dimensional dynamic probability map to update the three-dimensional dynamic probability map to generate an updated three-dimensional dynamic probability map, wherein the updated three-dimensional dynamic probability map is used to reflect the latest environmental status; Based on the updated three-dimensional dynamic probability map, a decision-making framework combining deep reinforcement learning and evolutionary algorithm is used to generate a real-time decision map. The framework is used to combine integrated multi-agent reinforcement learning to enable multiple micro-robots to collaborate with each other to complete complex tasks. The complex tasks include group obstacle avoidance and collaborative transportation, and adversarial training elements are used to simulate opponents or unforeseen environmental factors to generate a real-time decision map. The real-time decision map is used to guide the micro-robots to make optimal decisions.

5. The method according to claim 4, characterized in that The method generates an instant decision map based on the updated three-dimensional dynamic probability map by using a decision framework combining deep reinforcement learning and evolutionary algorithm, including: Based on the updated three-dimensional dynamic probability map, the decision framework combining deep reinforcement learning and evolutionary algorithm is used to automatically adjust the parameters of the environmental state change prediction model to obtain an optimized decision framework; Based on the optimized decision-making framework, multi-agent reinforcement learning is introduced to enable multiple micro-robots to collaborate with each other to complete complex tasks, including group obstacle avoidance and collaborative transportation. Each micro-robot not only considers its own behavior strategy, but also shares information with other micro-robots to jointly develop the best action plan to generate a collaborative behavior model. Based on the collaborative behavior pattern, an adversarial network is generated using adversarial training elements to model and simulate opponents or unforeseen environmental factors, enhance the collaborative ability of the micro-robot, and generate a collaboratively enhanced model; The instant decision map is generated according to the optimized decision framework and the collaboratively enhanced model.

6. The method according to claim 1, characterized in that The method uses a Kalman filter to estimate the position of the micro robot, obtains the position state of the micro robot, adjusts the speed and steering angle of the drive motor of the micro robot based on the position state and the adaptive behavior rule set, ensures that the actual motion trajectory conforms to the ideal motion trajectory, and generates the trajectory tracking of the micro robot. The trajectory tracking refers to the process in which the micro robot moves according to the predetermined motion trajectory during the execution of the task, and adjusts its own speed and steering angle in real time to minimize the deviation between the actual motion trajectory and the ideal trajectory, including: Using the Kalman filter to estimate the position of the environmental state of the micro robot, obtain the precise position state of the micro robot, and based on the precise position state, adjust the speed and steering angle of the drive motor to ensure that the actual motion trajectory conforms to the ideal motion trajectory, and obtain the adjustment result of the position state; According to the adjustment result of the position state, visual servo control processing is performed on the micro robot to obtain the actual motion trajectory after real-time adjustment, to ensure that the ideal motion trajectory can be achieved even when the environmental state changes, and to obtain trajectory tracking data of the micro robot; Based on the trajectory tracking data, the difference between the actual motion trajectory and the ideal motion trajectory is compared to generate an optimized motion pattern. The optimized motion pattern is used to improve the motion control method of the micro robot to obtain the trajectory tracking of the micro robot.

7. The method according to claim 6, characterized in that The method of performing visual servo control processing on the micro robot according to the adjustment result of the position state to obtain the actual motion trajectory adjusted in real time, ensuring that the ideal motion trajectory can be achieved even when the environmental state changes, and obtaining the trajectory tracking data of the micro robot includes: Using the adjustment result of the position state, visual servo control processing is performed on the micro robot, and the actual motion trajectory is adjusted in real time by introducing image processing technology and computer vision algorithm to obtain the adjusted actual motion trajectory; According to the adjusted actual motion trajectory, combined with multi-sensor fusion technology, the accuracy of position estimation is further improved, so that the micro robot can achieve sub-centimeter positioning accuracy in a complex environment, and based on the high-precision position information, the speed and steering angle of the drive motor are fine-tuned to generate optimized motion instructions; Based on the optimized motion instructions, adaptive control theory and predictive control algorithms are applied to calculate and compensate in advance for factors that may affect the actual motion trajectory, so that the microrobot can move smoothly along the ideal motion trajectory, and obtain the trajectory tracking data of the microrobot. Based on the trajectory tracking data, combined with the target tracking technology in deep learning, the travel route of the microrobot is continuously monitored and corrected to ensure that the travel route meets the task requirements.

8. The method according to claim 4, characterized in that Based on the updated three-dimensional dynamic probability map, a decision framework combining deep reinforcement learning and evolutionary algorithm is used to generate an instant decision map. The framework is used to combine integrated multi-agent reinforcement learning to enable multiple micro-robots to collaborate with each other to complete complex tasks. The complex tasks include group obstacle avoidance and collaborative transportation, and adversarial training elements are used to simulate opponents or unforeseen environmental factors to generate an instant decision map. The instant decision map is used to guide the micro-robot to make the best decision, including: Based on the updated three-dimensional dynamic probability map, the decision framework combining deep reinforcement learning and evolutionary algorithm is used to automatically adjust the parameters of the environmental state change prediction model to obtain an optimized decision framework; The parameters of the environmental state change prediction model include a smoothing coefficient, a dynamically adjusted smoothing coefficient η t Calculated by the following formula: Among them, η t represents the dynamically adjusted smoothing coefficient, k represents the speed of controlling the change of the smoothing coefficient, Δs t represents the change in the environmental state between the current moment and the previous moment, and b represents the threshold value, which is used to determine whether the environmental change is significant; Based on the optimized decision-making framework, multi-agent reinforcement learning is introduced to enable multiple micro-robots to collaborate with each other to complete complex tasks, including group obstacle avoidance and collaborative transportation. Each micro-robot not only considers its own behavior strategy, but also shares information with other micro-robots to jointly develop the best action plan to generate a collaborative behavior model. Among them, the behavior strategy of the micro robot considering itself includes a parameter local value weight w local , and the information shared by the micro robot and other robots includes a parameter global value weight w global , local value weight w local and the global value weight w global Calculated by the following formula: Among them, w local Indicates the definition of local value weight, w global represents the global value weight, d(s,s0) represents the distance measure between the current state s and the initial state s0, T represents the total task duration, and t represents the current time; Based on the collaborative behavior pattern, an adversarial network is generated by using adversarial training elements to model and simulate opponents or unforeseen environmental factors. An exponential factor is included in the simulated reference data to enhance the collaborative ability of the micro-robot and generate a collaboratively enhanced model. The exponential factor γ is calculated by the following formula: Among them, γ represents the introduction of exponential factor, α represents a constant that controls the base value of the exponential factor, β represents a constant that controls the growth rate of the exponential factor, C represents the current environment complexity assessment value, and C0 represents the benchmark complexity assessment value; The instant decision map is generated according to the optimized decision framework and the collaboratively enhanced model; wherein the instant decision map is calculated by the following formula: M t+1 (s,a)=(1-η t )M t (s,a)+η t ·(w local Q local (s,a)+w global Q global (s,a)) γ Among them, M t+1 (s,a) represents the updated value of taking action a in state s in the instant decision map at time t+1, where s represents the state, a represents the action, and M t (s,a) represents the value of taking action a in state s in the instant decision map at time t, Q local (s,a) represents the local value function estimated by fast random tree or other local search algorithm, Q global (s,a) represents the global value function estimated by a Bayesian network or other global optimization algorithm.

9. The method according to claim 6, characterized in that The method of performing visual servo control processing on the micro robot according to the adjustment result of the position state to obtain the actual motion trajectory adjusted in real time, ensuring that the ideal motion trajectory can be achieved even when the environmental state changes, and obtaining the trajectory tracking data of the micro robot includes: The adjustment result of the position state is used to perform visual servo control processing on the micro robot, and the actual motion trajectory is adjusted in real time by introducing image processing technology and computer vision algorithm to obtain the adjusted actual motion trajectory; wherein the visual servo control updates u t+1 Calculated by the following formula: Among them, u t+1 represents the control input of the visual servo control update at time t+1, u t represents the control input at time t, K v represents the visual servo controller gain matrix, which is used to adjust the control force, x d,t represents the desired target position or feature point, x t Represents the actual position or feature point at the current moment, α t Represents the adaptive learning rate, which is dynamically adjusted according to the rate of change of the environment. Represents the gradient of the loss function with respect to the control input, which is used to optimize the control strategy; According to the adjusted actual motion trajectory, combined with multi-sensor fusion technology, the accuracy of position estimation is further improved, so that the micro robot can achieve sub-centimeter positioning accuracy in complex environments, and based on the high-precision position information, the speed and steering angle of the drive motor are fine-tuned to generate optimized motion instructions; wherein the fused high-precision position estimation p t and drive motor fine-tuning v t+1 Calculated by the following formula: Among them, p t represents the high-precision position estimate after fusion, N s represents the number of sensors, w i (t) represents the dynamic weight of the i-th sensor, reflecting the credibility of the sensor data, which is dynamically adjusted according to the sensor error and environmental conditions. i,t represents the measurement value of the i-th sensor at time t, p t represents the high-precision position estimate after fusion, v t represents the driving motor speed and steering angle of the driving motor at time t, v t+1 represents the driving motor speed and steering angle of the driving motor at time t+1, K p Represents the proportional control gain matrix, which is used to adjust the fine-tuning strength, p d,t represents the desired target position, p t represents the high-precision position estimate at the current moment, β t represents the compensation coefficient, which is dynamically adjusted according to task requirements, σ act,t and σ des,t Represents the standard deviation of actual and expected position uncertainty, used to compensate for position error; Based on the optimized motion instructions, the adaptive control theory and predictive control algorithm are applied to calculate and compensate in advance the factors that may affect the actual motion trajectory, so that the micro robot can move smoothly along the ideal motion trajectory, and obtain the trajectory tracking data of the micro robot; wherein the adaptive control updates θ t+1 and predictive compensation c t+1 Calculated by the following formula: Among them, θ t represents the adaptive controller parameters at time t, θ t+1 represents the adaptive controller parameters at time t+1, α represents the learning rate, which controls the step size of each update, Represents the gradient of the loss function with respect to the parameters, which is used to guide the direction of parameter update, γ t represents the evolution factor, which is dynamically adjusted according to the task difficulty and environment complexity, Δθ t represents the change in the best hyperparameter combination searched by the evolutionary algorithm, c t represents the compensation amount at time t, c t+1 represents the compensation amount at time t+1, β represents the compensation coefficient, which controls the compensation strength, and f pred,t represents the predicted disturbance force, f act,t represents the actual interference force encountered, δ t represents an exponential factor, which is dynamically adjusted according to the complexity of the environment to enhance the focus on key interference; Based on the trajectory tracking data, combined with the target tracking technology in deep learning, the route of the micro robot is continuously monitored and corrected to make the route meet the task requirements; wherein, the target tracking update and the incremental learning update are calculated by the following formula: Among them, x track,t+1 represents the tracking state vector of the target tracking update at time t+1, x track,t represents the tracking state vector at time t, K t Represents the Kalman gain, which is used to adjust the correction strength of the tracking error, y t represents the observed value, h() represents the observation model, λ t Adaptive adjustment coefficient, dynamically adjusted according to the tracking error, maps the state vector to the observation space, ρ t represents the exponential decay factor, which is used to balance the impact of long-term and short-term tracking errors, e t represents the tracking error at the current moment, w t+1 represents the incremental learning update of the behavior rule weight at time t+1, w t represents the weight of the behavior rule at time t, η represents the learning rate, which controls the step size of each update, Represents the gradient of the loss function with respect to the weights, based on the current dataset Calculate μ t represents the incremental learning adjustment coefficient, which is dynamically adjusted according to the progress of the task, σ err,t and σ ref Represents the standard deviation of the error between the current and reference behavior rules, which is used to optimize the learning process.

10. A micro robot motion control system, characterized in that: include: A collection and processing module, used to collect current environmental status information in real time using a non-contact sensor, and synchronize and calibrate the environmental status information to generate a comprehensive perception image; A classification prediction module, used to apply target detection and recognition technology to analyze the comprehensive perception image to determine the objects existing in the current environment and the categories of the objects, obtain classification results, and build an environment state change prediction model based on the classification results and the environment state information. The environment state change prediction model is used to predict the environment state changes that affect the motion trajectory of the micro robot in the future period of time through historical environment state information and the current environment state information, combined with the classification results; An optimization processing module, used to generate candidate paths using a fast random tree according to the comprehensive perception image and the environmental state changes predicted by the environmental state change prediction model, and optimize the candidate paths in combination with an AI search algorithm to obtain a target path; An update generation module is used to construct a three-dimensional dynamic probability map based on the target path using a Bayesian network, wherein the three-dimensional dynamic probability map uses the environmental state at a specific location as a node and the transition probability between the environmental states as an edge, and continuously updates the three-dimensional dynamic probability map using a reinforcement learning algorithm to generate an instant decision map; A simulation adjustment module is used to adjust the motion trajectory of the micro robot when a change in the environmental state is detected, based on the instant decision map and introducing an emotional computing model to simulate emotional factors in the human decision-making process, so as to generate an adaptive behavior rule set; The position adjustment module is used to use a Kalman filter to estimate the position of the micro robot, obtain the position state of the micro robot, and adjust the speed and steering angle of the drive motor of the micro robot based on the position state and the adaptive behavior rule set to ensure that the actual motion trajectory conforms to the ideal motion trajectory, and generate the trajectory tracking of the micro robot. The trajectory tracking refers to the process in which the micro robot moves according to a predetermined motion trajectory during the execution of a task, and adjusts its own speed and steering angle in real time to minimize the deviation between the actual motion trajectory and the ideal trajectory.

Images