Robot stability and balance control method based on embedded depth fusion algorithm

Abstract

The invention relates to the technical field of robots, in particular to a robot stability and balance control method based on an embedded depth fusion algorithm, and the method comprises the following steps: S1, sensor data collection: configuring a plurality of sensors to collect the motion state and environment information of a robot in real time; s2, dynamic scene understanding: performing real-time analysis on the acquired sensor data, and identifying potential obstacles and changes in the environment; s3, cooperative control based on behavior prediction: predicting future behaviors of a plurality of robots, and dynamically adjusting a cooperative strategy; and S4, real-time feedback adjustment: comparing the current state of the robot with a preset stable state through a multi-level feedback mechanism, and intelligently adjusting a control strategy based on real-time environment change and task requirements. According to the invention, the adaptability and task execution efficiency of the system are improved, and more intelligent and flexible cooperative work can be realized in a complex scene.

Description

Technical Field

[0001] The present invention relates to the field of robotics technology, and in particular to a robot stability and balance control method based on an embedded deep fusion algorithm. Background Art

[0002] With the rapid development of robotics technology, robots are increasingly used in industry, medical care, services and other fields. In order to ensure that robots have efficient operation capabilities and stability in complex and changing environments, more and more sensors and intelligent algorithms are being applied to robotic systems to achieve environmental perception, analysis and behavior control. Especially in multi-robot collaborative tasks, how to ensure efficient collaboration between robots and avoid resource conflicts has become one of the key issues in the development of robotics technology today.

[0003] However, the existing technology still has many shortcomings in the robot stability and balance control. On the one hand, although the robot can collect environmental data through a variety of sensors, there are still certain limitations in the fusion of sensor data and the recognition accuracy of obstacles when dealing with complex dynamic scenes, especially in real-time analysis and adaptation to environmental changes. On the other hand, the existing collaborative control methods lack intelligence in multi-robot task scheduling and path planning, and are unable to handle the dynamic changes of obstacles and resource conflicts between robots in a timely manner, resulting in poor execution efficiency and collaboration of multi-robot systems in complex scenarios, and are prone to problems such as task delays and waste of resources.

[0004] The purpose of the present invention is to propose a robot stability and balance control method based on an embedded deep fusion algorithm to address the above-mentioned problems. Through a multi-level feedback mechanism and intelligent collaboration strategy adjustment, efficient collaboration and optimal resource allocation of multi-robot systems in complex environments are achieved, ensuring that the robots can maintain balance and stability in a dynamic environment, further improving the safety of operations and the efficiency of task execution. Summary of the invention

[0005] The present invention provides a robot stability and balance control method based on an embedded deep fusion algorithm.

[0006] A robot stability and balance control method based on an embedded deep fusion algorithm comprises the following steps: S1, sensor data acquisition: configure multiple sensors, including accelerometers, gyroscopes, visual sensors, and acoustic sensors to collect the robot's motion status and environmental information in real time; S2, Dynamic Scene Understanding: Dynamic scene understanding algorithms are used to analyze the collected sensor data in real time to identify potential obstacles and changes in the environment, including: S21, data preprocessing: preprocessing the collected sensor data, including filtering and denoising; S22, Obstacle Identification: Using dynamic scene understanding algorithms, we analyze pre-processed sensor data to identify potential obstacles in the environment in real time and determine their location, shape, and dynamic characteristics; S23, scene change prediction: Based on the obstacle recognition results and historical sensor data, the scene change prediction model is used to analyze the environmental change trend and evaluate the impact of potential scene changes on the robot; S3, collaborative control based on behavior prediction: Combines the identification results of potential obstacles and changes in the environment to predict the future behavior of multiple robots and dynamically adjust the collaborative strategy; S4, real-time feedback adjustment: Through a multi-level feedback mechanism, the current state of the robot is compared with the preset stable state, and the control strategy is intelligently adjusted based on real-time environmental changes and task requirements.

[0007] Optionally, the sensor data collection in S1 includes: S11, accelerometer configuration: configure accelerometers at different positions of the robot to measure the acceleration of the robot in various directions to obtain motion state data; S12, gyroscope configuration: configure the gyroscope at the core of the robot to detect the angular velocity of the robot and maintain the stability and direction control of the robot by real-time monitoring of posture changes; S13, visual sensor configuration: configure the visual sensor in front of the robot to capture environmental images and videos, and identify surrounding obstacles, terrain features and environmental information; S14, acoustic sensor configuration: configure acoustic sensors around the robot to capture sound information of the environment.

[0008] Optionally, the data preprocessing in S21 includes: S211, data filtering: using low-pass filtering algorithm to smooth the sensor data; S212, data denoising: denoising the sensor data using a Kalman filter algorithm.

[0009] Optionally, the obstacle identification in S22 includes: S221, data input: taking the preprocessed sensor data as input and constructing a multi-dimensional feature vector; S222, obstacle detection: using convolutional neural network (CNN) algorithm for obstacle recognition; S223, position, shape and dynamic feature extraction: Use the Canny edge detection algorithm to process the output feature map to determine the boundary, position and shape of the obstacle.

[0010] Optionally, the scene change prediction model in S23 adopts a graph neural network (GNN) model, and the graph neural network (GNN) model includes: S231, Graph Construction: Build a graph based on sensor data and obstacle recognition results in the environment. , where the node Represents obstacles in the environment, edges represents the relationship between these obstacles; S232, feature initialization: on the constructed graph, initialize the feature vector for each node; S233, information aggregation: Aggregate information through neighboring nodes, consider the influence of different obstacles, introduce weight factors, and perform weighted aggregation on each node according to the characteristics of its neighboring nodes; S234, feature update: after information aggregation, update the features of each node; S235, scene change prediction: based on the updated node features, the scene change is predicted through the fully connected layer and activation function.

[0011] Optionally, the behavior prediction-based collaborative control in S3 includes: S31, environmental status information extraction: combining sensor data and obstacle recognition results to extract the current environmental status information, including the location, dynamic characteristics and environmental change trend of obstacles; S32, Robot Behavior Prediction: Based on the extracted environmental state information and the dynamic motion model of each robot, the behavior prediction algorithm is applied to predict the future behavior trajectories of multiple robots, including movement direction, speed, and obstacle avoidance behavior; S33, dynamic adjustment of collaboration strategy: According to the predicted robot behavior trajectory and the dynamic changes of obstacles, the collaboration strategy is dynamically adjusted, including priority allocation, task switching or path replanning.

[0012] Optionally, the environmental status information extraction in S31 includes: S311, obstacle position extraction: According to the obstacle recognition result, the obstacle position is calculated based on the monocular camera. ; S312, extraction of dynamic features and environmental change trends: By analyzing historical sensor data, the Kalman filter algorithm is used to predict the speed and acceleration of obstacles.

[0013] Optionally, the robot behavior prediction in S32 includes: S321, environmental status information input: the extracted environmental status information As input, it includes the location of obstacles, dynamic characteristics, and environmental change trends; S322, dynamic motion model: calculating the future motion state of each robot according to the uniform motion model; S323, behavior prediction algorithm: uses recurrent neural network (RNN) as a behavior prediction algorithm to predict the future behavior trajectories of multiple robots; S324, obstacle avoidance behavior prediction: combining the dynamic changes of obstacles and the robot's own trajectory prediction results to adjust the robot's future movement path.

[0014] Optionally, the dynamic adjustment of the collaboration strategy in S33 includes: S331, priority allocation: Calculate the task priority of each robot according to the predicted behavior trajectory of each robot and the current task requirements. The priority is determined based on the distance between the robot and the target, the urgency of the task, and the current resource utilization; S332, Task Switching: When new obstacles appear in the environment or task requirements change, the task allocation is automatically adjusted based on the robot's predicted behavior trajectory; S333, path replanning: Combining the dynamic changes of obstacles and the robot behavior prediction results, the A* algorithm is used to recalculate the robot's optimal path to the target while avoiding dynamic obstacles.

[0015] Optionally, the real-time feedback adjustment in S4 includes: S41, state monitoring and feedback: real-time monitoring of the robot's current motion state , including location ,speed , acceleration and task progress, and compare it to the preset steady state Compare and calculate the state error ; S42, error feedback adjustment: according to the error , error adjustment is performed through the proportional-integral-derivative (PID) control algorithm; S43, intelligent adjustment of control strategy: Based on real-time environmental changes and task requirements, by predicting future states And the current state error , dynamically adjust the robot motion control signal and optimize the robot's movement direction, speed or task execution strategy.

[0016] Beneficial effects of the present invention: The present invention, by configuring multiple sensors and combining dynamic scene understanding algorithms, can collect and analyze the robot's motion state and surrounding environment information in real time, which not only improves the robot's perception ability and can effectively identify obstacles and their changing trends in complex environments, but also improves the accuracy and processing speed of obstacle identification through multimodal data fusion and precise feature extraction, so that the robot has higher operational safety and adaptability in dynamic environments.

[0017] The present invention predicts the future behavior trajectories and obstacle avoidance paths of multiple robots in advance, dynamically adjusts the collaboration strategy, and uses multi-level dynamic collaboration mechanisms such as priority allocation, task switching, and path replanning. Robots can achieve efficient collaboration in complex task environments, reduce resource conflicts, improve the system's adaptability and task execution efficiency, and ensure more intelligent and flexible collaborative work in complex scenarios.

[0018] The present invention, by introducing a multi-level feedback mechanism, combined with a PID control algorithm and future state prediction, can intelligently adjust the robot's motion state and stable state in real time, ensuring that the robot can quickly respond to real-time environmental changes and task requirements. By dynamically adjusting the control strategy, the robot's stability, autonomous control capability and accuracy of task execution in complex environments are improved, ensuring efficient task completion in a changing environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 A schematic flow chart of a control method according to an embodiment of the present invention; Figure 2 A schematic diagram of dynamic scene understanding according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0022] It should be noted that the references to "one embodiment", "embodiment", "exemplary embodiments", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0023] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0024] like Figure 1-Figure 2 As shown, a robot stability and balance control method based on an embedded deep fusion algorithm comprises the following steps: S1, sensor data acquisition: configure multiple sensors, including accelerometers, gyroscopes, visual sensors, and acoustic sensors to collect the robot's motion status and environmental information in real time; S2, Dynamic Scene Understanding: Use dynamic scene understanding algorithms to analyze collected sensor data in real time, identify potential obstacles and changes in the environment, and improve the robot's adaptability in complex and dynamic environments. Specifically, it includes: S21, data preprocessing: preprocessing the collected sensor data, including filtering and denoising; S22, Obstacle Identification: Using dynamic scene understanding algorithms, we analyze pre-processed sensor data to identify potential obstacles in the environment in real time and determine their location, shape, and dynamic characteristics; S23, scene change prediction: Based on the obstacle recognition results and historical sensor data, the scene change prediction model is used to analyze the environmental change trend and evaluate the impact of potential scene changes on the robot; S3, collaborative control based on behavior prediction: Combines the identification results of potential obstacles and changes in the environment, predicts the future behavior of multiple robots, and dynamically adjusts the collaborative strategy to optimize collective action and improve stability; S4, real-time feedback adjustment: Through a multi-level feedback mechanism, the current state of the robot is compared with the preset stable state, and the control strategy is intelligently adjusted based on real-time environmental changes and task requirements to ensure the stability and balance of the robot in complex environments; Through the above content, the robot's adaptability and stability in complex and dynamic environments have been improved. It can not only identify environmental changes and potential obstacles in real time, but also optimize the collective actions of multiple robots through collaborative control, thereby effectively improving the safety and efficiency of operations. It has strong practicality and innovation.

[0025] Sensor data collection in S1 includes: S11, accelerometer configuration: configure accelerometers at different positions of the robot to measure the acceleration of the robot in various directions to obtain motion state data; S12, gyroscope configuration: configure the gyroscope at the core of the robot to detect the angular velocity of the robot and maintain the stability and direction control of the robot by real-time monitoring of posture changes; S13, visual sensor configuration: configure the visual sensor in front of the robot to capture environmental images and videos, and identify surrounding obstacles, terrain features and environmental information; S14, acoustic sensor configuration: configure acoustic sensors around the robot to capture sound information of the environment; Through the above content, not only the robot's dynamic perception ability is enhanced, enabling it to identify surrounding obstacles and environmental changes in real time, but also the motion stability and direction control accuracy are improved, thereby significantly improving the robot's adaptability and safety in complex and dynamic environments, ensuring more efficient operation and decision-making.

[0026] The data preprocessing in S21 includes: S211, data filtering: Use low-pass filtering algorithm to smooth the sensor data, expressed as: ; in, is the output data after filtering, For the currently collected data, is the filter factor ; S212, data denoising: The sensor data is denoised using the Kalman filter algorithm, expressed as: ; in, is the estimated value of the current state, is the current observation value, is the Kalman gain, is the observation matrix; Through the above, the accuracy and reliability of sensor data are effectively improved. This dual processing not only enhances the effectiveness of subsequent dynamic scene understanding and decision-making processes, but also significantly improves the stability and response speed of the robot in complex environments, thereby improving overall operational efficiency and safety.

[0027] Obstacle recognition in S22 includes: S221, data input: take the preprocessed sensor data as input and construct a multi-dimensional feature vector, expressed as: ; in, is a multidimensional feature vector, Represents the characteristic information provided by different sensors; S222, obstacle detection: The convolutional neural network (CNN) algorithm is used for obstacle recognition, which is expressed as: ; in, is the convolution kernel, is the bias, is the activation function, is the output feature map; S223, position, shape and dynamic feature extraction: Use the Canny edge detection algorithm to process the output feature map to determine the boundary, position and shape of the obstacle, expressed as: ; in, and are the gradients of the image in the x and y directions respectively; Through the above content, the recognition accuracy and speed of potential obstacles in the environment are improved. The convolution operation can effectively extract complex features, and the nonlinearity introduced by the activation function enhances the expressive ability of the model, enabling the robot to accurately identify the position, shape and dynamic characteristics of obstacles. It not only improves the robot's environmental perception ability, but also provides reliable information support for subsequent path planning and decision-making, thereby enhancing the robot's operational safety and efficiency in complex and dynamic environments.

[0028] The scene change prediction model in S23 adopts the graph neural network (GNN) model, which includes: S231, Graph Construction: Build a graph based on sensor data and obstacle recognition results in the environment. , where the node Represents obstacles in the environment, edges represents the relationship between these obstacles; S232, feature initialization: On the constructed graph, initialize the feature vector for each node, expressed as: ; in, , , They are nodes 1, 2, The initial eigenvector of S233, information aggregation: Aggregate information through neighboring nodes, take into account the influence of different obstacles, introduce weight factors to ensure the effectiveness of information transmission, and perform weighted aggregation based on the characteristics of its neighboring nodes, so that the node can absorb information about the surrounding environment and form a more accurate feature representation, which is expressed as: ; in, For Node In the The updated feature vector after layer information aggregation, For the node The set of all connected adjacent nodes, For Node and nodes The weight factor between For Layer The weight matrix of For Node and nodes The connection relationship between is the ReLU activation function, For the Layer Node The characteristic vector of For the Layer Node The eigenvector of S234, feature update: After information aggregation, the features of each node are updated. The update process combines the node's own features and the information aggregated from neighboring nodes, which can gradually enhance the node features in multiple rounds of information propagation, thereby better reflecting the dynamic changes of the environment, expressed as: ; in, is the attention mechanism function for information aggregation, which is used to calculate the node and nodes The weight of the relationship between ; in, is the normalization function, is the dimension of the feature vector; S235, scene change prediction: According to the updated node features, the scene change is predicted through the fully connected layer and activation function, which is expressed as: ; in, is the final prediction output, The last layer of the graph neural network ( The feature matrix of all nodes in the layer, is the weight matrix for linear transformation, is the bias vector; Through the above content, it is possible to more accurately capture the obstacle relationships and dynamic changes in complex environments, and perform well in real-time prediction of scene changes. Especially when processing diverse and interrelated environmental information, it is able to integrate the features of multiple nodes to improve the accuracy of environmental understanding. By introducing attention mechanisms and feature updates, the model is more flexible and adaptable in dealing with dynamic changes, thereby enhancing the robot's decision-making ability, stability and operational efficiency in complex environments, and significantly improving the robustness and response speed of the overall system.

[0029] The behavior prediction-based collaborative control in S3 includes: S31, environmental status information extraction: combining sensor data and obstacle recognition results to extract the current environmental status information, including the location, dynamic characteristics and environmental change trend of obstacles; S32, Robot Behavior Prediction: Based on the extracted environmental state information and the dynamic motion model of each robot, the behavior prediction algorithm is applied to predict the future behavior trajectories of multiple robots, including movement direction, speed, and obstacle avoidance behavior; S33, dynamic adjustment of collaboration strategy: According to the predicted robot behavior trajectory and the dynamic changes of obstacles, the collaboration strategy is dynamically adjusted, including priority allocation, task switching or path replanning; Through the above content, the robot can perceive and predict potential obstacles and dynamic changes in complex environments in advance, and ensure efficient collaboration between robots and accuracy of task completion by dynamically adjusting collaboration strategies such as priority allocation, task switching or path replanning. This improves the system's adaptability, stability and efficiency, reduces the possibility of resource conflicts and path conflicts, and is suitable for complex scenarios where multiple robots work together.

[0030] The environmental status information extraction in S31 includes: S311, obstacle position extraction: According to the obstacle recognition result, the obstacle position is calculated based on the monocular camera. , expressed as: ; in, is the distance between the obstacle and the sensor, is the pixel coordinate of the obstacle in the image, is the focal length of the camera; S312, dynamic characteristics and environmental change trend extraction: By analyzing historical sensor data, the Kalman filter algorithm is used to predict the speed and acceleration of the obstacle, which is expressed as: ; in, is the predicted state at the next moment, is the state transfer matrix, is the control matrix, is the control input, is the noise term; Through the above content, the position, dynamic characteristics and environmental change trends of obstacles are effectively extracted. By combining the perspective projection method of the monocular camera and the Kalman filter algorithm, the three-dimensional position and motion trajectory of the obstacle can be accurately predicted. This not only improves the robot's perception of the environment, enabling it to cope with complex and dynamic scenes, but also enhances the robot's real-time response capability, ensuring that it has greater operational flexibility and stability in a changing environment.

[0031] Robot behavior prediction in S32 includes: S321, environmental status information input: the extracted environmental status information As input, it includes the location of obstacles, dynamic characteristics, and environmental change trends, expressed as: ; in, are the three-dimensional coordinates of the obstacle, It's speed. is the acceleration; S322, dynamic motion model: Calculate the future motion state of each robot according to its uniform motion model, expressed as: ; in, For the future position of the robot, is the current position of the robot, is the current speed, is the time step; S323, behavior prediction algorithm: Recurrent neural network (RNN) is used as the behavior prediction algorithm to predict the future behavior trajectories of multiple robots, expressed as: ; ; in, For the moment The hidden state vector represents the current behavior state of the robot. For the moment The hidden state vector of is the input state vector (including robot position, speed and environment state information), is the activation function, is the weight matrix of the hidden state, is the weight matrix of the input state, is the bias term of the hidden layer, The output prediction result is the future behavior trajectory of the robot. is the output weight matrix, is the bias term of the output layer; S324, obstacle avoidance behavior prediction: Combine the dynamic changes of obstacles and the robot's own trajectory prediction results to adjust the robot's future motion path to avoid collision with obstacles. The objective function of obstacle avoidance is expressed by minimizing the distance to the obstacle, which is expressed as: ; in, Represents the robot and obstacles The distance between is the number of obstacles; Through the above content, it is possible to effectively capture time series information and consider the relationship between past behaviors and current states, making predictions more accurate. By optimizing the distance between the robot and obstacles, the path can be dynamically adjusted to ensure the effectiveness of obstacle avoidance behavior. This not only improves the robot's autonomous decision-making ability in complex environments, but also greatly improves the safety and collaborative efficiency of task execution, making it more suitable for complex scenarios of multi-robot collaborative operations.

[0032] The dynamic adjustment of the collaboration strategy in S33 includes: S331, priority allocation: Calculate the task priority of each robot based on the predicted behavior trajectory of each robot and the current task requirements. The priority is determined based on the distance between the robot and the target, the urgency of the task, and the current resource utilization, expressed as: ; in, For robots The priority score of For robots The distance to the target (the closer the target, the higher the priority), is the task urgency, which indicates the urgency of task completion (the higher the task urgency, the higher the priority), For robots Current resource utilization, such as power or processing power (the fewer resources, the higher the priority, ensuring that the task is completed before the power or resources are exhausted), It is a weighted parameter that controls the impact of each factor on the priority and adjusts the weight according to the application scenario; Weighting parameters The setting basis and value range include: Weighting parameters The setting basis is the influence of the distance to the target on the priority. In some tasks, the robot closer to the target should have a higher priority. Therefore, when the influence of distance is greater, Should be set to a larger value, the value range is ; Weighting parameters The basis for setting the priority is the impact of task urgency on priority. If some tasks have high urgency, such as tasks that need to be completed in a short time or have time limits, the urgency weight Should be larger, the value range is ; Weighting parameters The setting basis is the impact of the robot's current resource utilization (such as power, computing power, etc.) on the priority. If the robot is short of resources, such as the battery is close to exhaustion or the processing power is limited, the resource utilization weight Should be large to ensure that the task can be completed before resources are exhausted. The value range is ; S332, task switching: When new obstacles appear in the environment or the task requirements change, the task allocation is automatically adjusted based on the robot's predicted behavior trajectory, expressed as: ; in, For robots The obstacle avoidance risk metric is calculated by calculating the distance between the robot and the obstacle. The sum of the reciprocals of the distances between them. If the distance is smaller, the risk is greater. For robots With obstacles The distance between is the number of obstacles in the robot’s current path or environment; , perform task switching; in, is the risk threshold of task switching, when the risk metric Exceeding risk threshold When the system determines that the robot Unable to complete the task safely and transfer the task to other robots; Risk Threshold A risk measure based on the robot's past collisions or near collisions with obstacles during task execution. To set, including: Collect historical data: record the distance between the robot and obstacles during the execution of multiple tasks and the corresponding risk measures ; Statistical analysis: These historical data are divided into two categories: one is the data of successful obstacle avoidance by the robot, and the other is the data of collision between the robot and obstacles. These two types of data are statistically analyzed to find the risk measurement when the collision occurs. The typical value range of , and the risk metric when the obstacle is successfully avoided; Calculate the risk threshold: Based on historical data, the threshold is determined by the mean plus standard deviation of historical collision data to ensure that the system can avoid collisions in most cases, expressed as: ; in, Representing risk measures for historical collision data The mean of represents the standard deviation of historical collision data, measuring the risk metric distribution under collision scenarios, is the safety factor, If the environment is complex and requires higher security, Can take larger value; S333, path replanning: Combine the dynamic changes of obstacles and the robot behavior prediction results, and recalculate the optimal path of the robot to the target through the A* algorithm, while avoiding dynamic obstacles, expressed as: ; in, For the current node The comprehensive evaluation function of The total cost to reach the destination, From the starting point to the current node The actual cost of For the current node Heuristic estimate of the target node; Through the above content, it is ensured that robots can collaborate efficiently in complex and changing environments to avoid task conflicts and waste of resources. When the environment or task requirements change, the system can adjust task allocation and path planning in real time to effectively improve the efficiency and safety of task execution. The task switching threshold set based on historical data further ensures the stability and adaptability of the system, enabling multi-robot systems to achieve smarter and more flexible collaborative work in complex environments.

[0033] Real-time feedback adjustments in S4 include: S41, state monitoring and feedback: real-time monitoring of the robot's current motion state , including location ,speed , acceleration and task progress, and compare it to the preset steady state Compare and calculate the state error , expressed as: ; in, Indicates the error between the current state of the robot and the preset stable state, is the current state, It is the preset stable state; S42, error feedback adjustment: according to the error , error adjustment is performed through the proportional-integral-derivative (PID) control algorithm, expressed as: ; in, is the adjusted control signal, are proportional, integral, and differential gain coefficients respectively; S43, intelligent adjustment of control strategy: Based on real-time environmental changes and task requirements, by predicting future states And the current state error , dynamically adjust the robot motion control signal and optimize the robot's moving direction, speed or task execution strategy, which is expressed as: ; Through the above content, the robot is ensured to maintain stable operation in a complex dynamic environment. At the same time, combined with future state prediction, the control strategy is intelligently adjusted, so that the robot can quickly respond to environmental changes and task requirements and maintain efficient autonomous control capabilities. The robot's adaptability and stability in complex scenarios are improved, effectively avoiding the accumulation of state deviations and ensuring the precise execution of tasks.

[0034] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0035] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A robot stability and balance control method based on an embedded deep fusion algorithm, characterized in that: The following steps are involved: S1, sensor data acquisition: configure multiple sensors, including accelerometers, gyroscopes, visual sensors, and acoustic sensors to collect the robot's motion status and environmental information in real time; S2, Dynamic Scene Understanding: Dynamic scene understanding algorithms are used to analyze the collected sensor data in real time to identify potential obstacles and changes in the environment, including: S21, data preprocessing: preprocessing the collected sensor data, including filtering and denoising; S22, Obstacle Identification: Using dynamic scene understanding algorithms, we analyze pre-processed sensor data to identify potential obstacles in the environment in real time and determine their location, shape, and dynamic characteristics; S23, scene change prediction: Based on the obstacle recognition results and historical sensor data, the scene change prediction model is used to analyze the environmental change trend and evaluate the impact of potential scene changes on the robot; S3, collaborative control based on behavior prediction: Combines the identification results of potential obstacles and changes in the environment to predict the future behavior of multiple robots and dynamically adjust the collaborative strategy; S4, real-time feedback adjustment: Through a multi-level feedback mechanism, the current state of the robot is compared with the preset stable state, and the control strategy is intelligently adjusted based on real-time environmental changes and task requirements.

2. According to claim 1, a robot stability and balance control method based on an embedded deep fusion algorithm is characterized in that: The sensor data collection in S1 includes: S11, accelerometer configuration: configure accelerometers at different positions of the robot to measure the acceleration of the robot in various directions to obtain motion state data; S12, gyroscope configuration: configure the gyroscope at the core of the robot to detect the angular velocity of the robot and maintain the stability and direction control of the robot by real-time monitoring of posture changes; S13, visual sensor configuration: configure the visual sensor in front of the robot to capture environmental images and videos, and identify surrounding obstacles, terrain features and environmental information; S14, acoustic sensor configuration: configure acoustic sensors around the robot to capture sound information of the environment.

3. The robot stability and balance control method based on the embedded deep fusion algorithm according to claim 1 is characterized in that: The data preprocessing in S21 includes: S211, data filtering: using low-pass filtering algorithm to smooth the sensor data; S212, data denoising: denoising the sensor data using a Kalman filter algorithm.

4. The robot stability and balance control method based on the embedded deep fusion algorithm according to claim 3 is characterized in that: The obstacle identification in S22 includes: S221, data input: taking the preprocessed sensor data as input and constructing a multi-dimensional feature vector; S222, obstacle detection: using convolutional neural network algorithm for obstacle recognition; S223, position, shape and dynamic feature extraction: Use the Canny edge detection algorithm to process the output feature map to determine the boundary, position and shape of the obstacle.

5. A robot stability and balance control method based on an embedded deep fusion algorithm according to claim 4, characterized in that: The scene change prediction model in S23 adopts a graph neural network model, and the graph neural network model includes: S231, Graph Construction: Build a graph based on sensor data and obstacle recognition results in the environment. , where the node Represents obstacles in the environment, edges represents the relationship between these obstacles; S232, feature initialization: on the constructed graph, initialize the feature vector for each node; S233, information aggregation: Aggregate information through neighboring nodes, consider the influence of different obstacles, introduce weight factors, and perform weighted aggregation on each node according to the characteristics of its neighboring nodes; S234, feature update: after information aggregation, update the features of each node; S235, scene change prediction: based on the updated node features, the scene change is predicted through the fully connected layer and activation function.

6. The robot stability and balance control method based on the embedded deep fusion algorithm according to claim 1 is characterized in that: The behavior prediction-based collaborative control in S3 includes: S31, environmental status information extraction: combining sensor data and obstacle recognition results to extract the current environmental status information, including the location, dynamic characteristics and environmental change trend of obstacles; S32, Robot Behavior Prediction: Based on the extracted environmental state information and the dynamic motion model of each robot, the behavior prediction algorithm is applied to predict the future behavior trajectories of multiple robots, including movement direction, speed, and obstacle avoidance behavior; S33, dynamic adjustment of collaboration strategy: According to the predicted robot behavior trajectory and the dynamic changes of obstacles, the collaboration strategy is dynamically adjusted, including priority allocation, task switching or path replanning.

7. A robot stability and balance control method based on an embedded deep fusion algorithm according to claim 6, characterized in that: The environmental status information extraction in S31 includes: S311, obstacle position extraction: According to the obstacle recognition result, the obstacle position is calculated based on the monocular camera. ; S312, extraction of dynamic features and environmental change trends: By analyzing historical sensor data, the Kalman filter algorithm is used to predict the speed and acceleration of obstacles.

8. The robot stability and balance control method based on the embedded deep fusion algorithm according to claim 7 is characterized in that: The robot behavior prediction in S32 includes: S321, environmental status information input: the extracted environmental status information As input, it includes the location of obstacles, dynamic characteristics, and environmental change trends; S322, dynamic motion model: calculating the future motion state of each robot according to the uniform motion model; S323, behavior prediction algorithm: using recursive neural network as a behavior prediction algorithm to predict the future behavior trajectories of multiple robots; S324, obstacle avoidance behavior prediction: combining the dynamic changes of obstacles and the robot's own trajectory prediction results to adjust the robot's future movement path.

9. The robot stability and balance control method based on the embedded deep fusion algorithm according to claim 8 is characterized in that: The dynamic adjustment of the collaboration strategy in S33 includes: S331, priority allocation: Calculate the task priority of each robot according to the predicted behavior trajectory of each robot and the current task requirements. The priority is determined based on the distance between the robot and the target, the urgency of the task, and the current resource utilization; S332, Task Switching: When new obstacles appear in the environment or task requirements change, the task allocation is automatically adjusted based on the robot's predicted behavior trajectory; S333, path replanning: Combining the dynamic changes of obstacles and the robot behavior prediction results, the A* algorithm is used to recalculate the robot's optimal path to the target while avoiding dynamic obstacles.

10. A robot stability and balance control method based on an embedded deep fusion algorithm according to claim 9, characterized in that: The real-time feedback adjustments in S4 include: S41, state monitoring and feedback: real-time monitoring of the robot's current motion state , including location ,speed , acceleration and task progress, and compare it to the preset steady state Compare and calculate the state error ; S42, error feedback adjustment: according to the error , error adjustment is performed through proportional-integral-derivative control algorithm; S43, intelligent adjustment of control strategy: Based on real-time environmental changes and task requirements, by predicting future states And the current state error , dynamically adjust the robot motion control signal and optimize the robot's moving direction, speed or task execution strategy.

Images