Adaptive behavior planning system for a perception robot

Abstract

The application provides an adaptive behavior planning system for a perception robot, and relates to the technical field of perception robots, and is based on a perception robot extracting a multi-sensor, performing environment perception, obtaining an environment perception dataset, preprocessing the environment perception dataset, dimensionally reducing the environment perception dataset according to a preprocessing result, generating an environment perception dimensionally reduced data array, performing behavior analysis on the perception robot according to the environment perception dimensionally reduced data array, combining multiple behavior information, constructing a behavior tree, performing strategy training through the behavior tree combined with the environment perception dimensionally reduced data array, and executing a behavior strategy to plan adaptive behaviors for the perception robot. The technical problem of low behavior planning efficiency and poor adaptability of the perception robot in a complex environment in the prior art is solved. The technical effects of improving the behavior planning efficiency of the perception robot, enhancing the adaptability to the complex environment and realizing intelligent behavior decision are achieved.

Description

Technical Field

[0001] This invention relates to the field of sensory robot technology, and more specifically to an adaptive behavior planning system for sensory robots. Background Technology

[0002] In the field of sensory robotics, behavior planning is one of the core technologies for achieving robot autonomy. The goal of behavior planning is to enable robots to generate reasonable behavioral strategies in real time based on perceived environmental information to complete predetermined tasks. However, most existing behavior planning methods are based on fixed rules or simple path planning algorithms, which suffer from poor adaptability to environmental changes and low computational efficiency. Furthermore, these methods typically require significant computational resources and time for path planning and decision-making, limiting the real-time performance and efficiency of sensory robots in practical applications.

[0003] Existing technologies suffer from low efficiency and poor adaptability in behavior planning for sensing robots in complex environments. Summary of the Invention

[0004] This application provides an adaptive behavior planning system for sensing robots, which addresses the technical problems of low efficiency and poor adaptability in behavior planning of sensing robots in complex environments in the prior art.

[0005] In view of the above problems, this application provides an adaptive behavior planning system for a sensing robot. The system includes: an environment perception module, used to extract multiple sensors based on the sensing robot, perform environmental perception through the multiple sensors, and obtain an environment perception dataset; a data dimensionality reduction module, used to preprocess the environment perception dataset, reduce the dimensionality of the environment perception dataset according to the preprocessing results, and generate an environment perception dimensionality reduction data array; a behavior analysis module, used to perform behavior analysis on the sensing robot based on the environment perception dimensionality reduction data array, and generate multiple behavior information; a behavior tree construction module, used to combine the multiple behavior information to construct a behavior tree; and a behavior planning module, used to perform policy training by combining the behavior tree with the environment perception dimensionality reduction data array, generate a behavior policy, and execute the behavior policy to perform adaptive behavior planning for the sensing robot.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] An environmental perception module is used to extract data from multiple sensors on the sensing robot, and obtain an environmental perception dataset through these sensors. A data dimensionality reduction module is used to preprocess the environmental perception dataset and reduce its dimensionality based on the preprocessing results, generating an environmental perception dimensionality-reduced data array. A behavior analysis module is used to analyze the behavior of the sensing robot based on the environmental perception dimensionality-reduced data array, generating multiple behavioral information. A behavior tree construction module is used to combine the multiple behavioral information to construct a behavior tree. A behavior planning module is used to train a strategy using the behavior tree and the environmental perception dimensionality-reduced data array, generate a behavior strategy, and execute the behavior strategy to adaptively plan the behavior of the sensing robot. This achieves the technical effects of improving the efficiency of the sensing robot's behavior planning, enhancing its adaptability to complex environments, and realizing intelligent behavior decision-making. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A schematic diagram of the adaptive behavior planning system for a sensing robot provided in this application;

[0010] Figure 2 This is a flowchart illustrating the process of generating multiple behavioral information in the adaptive behavior planning system for sensing robots provided in this application.

[0011] Figure labeling: 11 Environmental perception module, 12 Data dimensionality reduction module, 13 Behavior analysis module, 14 Behavior tree construction module, 15 Behavior planning module. Detailed Implementation

[0012] This application provides an adaptive behavior planning system for sensing robots, addressing the technical problems of low efficiency and poor adaptability in behavior planning for sensing robots in complex environments in existing technologies. It achieves the technical effects of improving the efficiency of behavior planning for sensing robots, enhancing adaptability to complex environments, and realizing intelligent behavioral decision-making.

[0013] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0014] like Figure 1 As shown, this application provides an adaptive behavior planning system for a sensing robot, the system comprising:

[0015] The environmental perception module 11 is used to extract multiple sensors based on the perception robot, perform environmental perception through the multiple sensors, and obtain an environmental perception dataset.

[0016] Specifically, the environmental perception module 11 is one of the core components of the sensing robot. It implements data acquisition and processing based on multiple sensors to provide comprehensive perception of the surrounding environment. The environmental perception module 11 extracts data from multiple sensors integrated into the sensing robot. These multiple sensors refer to various sensor devices integrated into the sensing robot, such as cameras, ultrasonic sensors, lidar, and infrared sensors, used to perceive different types of information in the environment, such as visual, sound, and distance information. The integration of multiple sensors on the sensing robot not only covers different information levels such as vision, sound, and distance, but also improves the sensing robot's perception accuracy and anti-interference capability by fusing data from various sensors. In its implementation, the environmental perception module 11 perceives the environment through each sensor, extracting its corresponding data signals, such as images, distance, or infrared radiation, to form an environmental perception dataset. This dataset is a collection of data acquired through multiple sensors, containing various environmental features. The environmental perception module 11 achieves efficient environmental perception, providing a fundamental guarantee for the robot's behavior planning.

[0017] The data dimensionality reduction module 12 is used to preprocess the environmental perception dataset, reduce the dimensionality of the environmental perception dataset according to the preprocessing results, and generate an environmental perception dimensionality reduction data array.

[0018] Specifically, the environmental sensing dataset obtained by the environmental sensing module 11 contains multidimensional data from multiple sensors, which contains a large amount of redundant information or noise. The data dimensionality reduction module 12 receives the environmental sensing dataset obtained by the environmental sensing module 11 and preprocesses it. The preprocessing includes signal filtering and noise reduction to remove random interference and useless information that may be encountered during sensor acquisition, improve data quality, and ensure the authenticity and validity of the sensing data. After the data preprocessing is completed, the data dimensionality reduction module 12 selects an appropriate dimensionality reduction algorithm, such as principal component analysis or linear discriminant analysis, based on the characteristics of the data to perform dimensionality reduction. The dimensionality-reduced data is mapped to a new low-dimensional space, which represents the core information of the original environmental sensing data, generating an environmental sensing dimensionality-reduced data array. This array represents the dimensionality-reduced data result of the environmental sensing dataset, preserving the core information of the data. By processing the environmental sensing dataset through the data dimensionality reduction module 12, the dimensionality of the data is reduced while retaining the most informative features, thereby making subsequent analysis and processing more efficient.

[0019] Furthermore, the environmental perception dataset is preprocessed. The data dimensionality reduction module 12 is also used to: perform multidimensional analysis based on the environmental perception dataset to determine multidimensional environmental state signals; perform wavelet decomposition on the multidimensional environmental state signals to obtain wavelet coefficients of the multidimensional environmental state signals; perform threshold quantization on the wavelet coefficients of the multidimensional environmental state signals to determine a wavelet selection threshold for the multidimensional environmental state signals; truncate the wavelet coefficients of the multidimensional environmental state signals according to the wavelet selection threshold, set noise signals smaller than the wavelet selection threshold to zero, and obtain effective signal information greater than the wavelet selection threshold; and filter and reconstruct the effective signal information to obtain the preprocessing result.

[0020] Specifically, in the data preprocessing stage, the data dimensionality reduction module 12 performs multidimensional analysis on the environmental perception dataset using multidimensional data analysis techniques such as cluster analysis and association rule mining. Through the multidimensional analysis results, a multidimensional environmental state signal is obtained. This multidimensional environmental state signal includes different dimensional features of the environment perceived by the sensing robot, such as the distance to the robot's location, the state of surrounding objects, and temperature changes. Then, wavelet decomposition technology is used to process the multidimensional environmental state signal. Wavelet decomposition is an effective signal processing method that can decompose complex signals into sub-signals of different frequencies, revealing different scale features within the signal. Wavelet decomposition transforms the multidimensional environmental state signal into wavelet coefficients, forming multidimensional environmental state signal wavelet coefficients. Each wavelet coefficient represents the variation characteristics of the environmental state signal at a specific scale. Next, threshold quantization is performed on the obtained multidimensional environmental state signal wavelet coefficients. Based on the distribution and importance of the wavelet coefficients, a wavelet selection threshold is determined. This multidimensional environmental state signal is used to distinguish between valid signals and noise. Based on the set wavelet selection threshold for the multidimensional environmental state signal, the wavelet coefficients of the multidimensional environmental state signal are truncated. Wavelet coefficients smaller than the wavelet selection threshold are set to zero. These zeroed-out wavelet coefficients are non-essential signals caused by external interference or sensor accuracy limitations; by setting them to zero, noise is removed. The remaining wavelet coefficients larger than the multidimensional environmental state signal wavelet selection threshold are considered valid signal information. Finally, the extracted valid signal information is filtered and reconstructed. The reconstruction process smooths the signal using filtering methods, removes remaining noise, and restores the true characteristics of the signal, obtaining the preprocessed result. Filtering methods include Gaussian filtering, mean filtering, and Kalman filtering. For example, Gaussian filtering can be performed using the following formula: Where μ is the mean of x, σ is the standard deviation of x, representing the dispersion of the data and determining the shape of the distribution; the smaller σ is, the more concentrated the distribution and the thinner the shape; e is the type of noise in the process. Through processing, the data dimensionality reduction module 12 transforms the original multidimensional sensing data into a more concise and reliable set of signals, providing accurate data support for the robot's subsequent behavior planning and decision-making.

[0021] Furthermore, the environmental perception dataset is dimensionality reduced based on the preprocessing results to generate an environmental perception dimensionality-reduced data array. The data dimensionality reduction module 12 is also used to: perform distribution calculation on the environmental perception dataset based on the wavelet selection threshold of the multidimensional environmental state signal to obtain a data distribution probability array; construct a similarity probability distribution function to perform dimensionality reduction processing on the point distribution of the data distribution probability array to generate a data dimensionality reduction result; and construct the environmental perception dimensionality-reduced data array based on the data dimensionality reduction result.

[0022] Specifically, the environmental perception dataset is analyzed and calculated using wavelet selection thresholds for the multidimensional environmental state signals generated in the preprocessing stage. By calculating the distribution characteristics of the environmental perception dataset, a data distribution probability array is obtained. This array describes the distribution characteristics of the data in multidimensional space in mathematical probability form, intuitively reflecting the concentration and distribution pattern of the data. The signals filtered based on the wavelet selection thresholds for the multidimensional environmental state signals more accurately reflect the actual effective distribution of the environmental perception data, eliminating interference from noise data. After obtaining the data distribution probability array, for each pair of data points in the environmental perception dataset, the similarity value between them is calculated based on Euclidean distance, Manhattan distance, or cosine similarity calculation methods. All similarity values ​​are arranged to generate a similarity matrix. The elements of the similarity matrix are normalized so that the sum of the elements in each row is 1, obtaining the normalized similarity matrix, i.e., the similarity probability distribution function. The similarity probability distribution function reflects the similarity distribution of each data point relative to other data points. Based on the constructed similarity probability distribution function, principal component analysis, t-SNE, or other nonlinear dimensionality reduction methods are used to reduce the dimensionality of the data distribution probability array, mapping data points in the high-dimensional space to the low-dimensional space, generating a data dimensionality reduction result. This result is a simplified low-dimensional dataset that still accurately reflects the characteristics of the original data. Finally, based on the dimensionality reduction result, the dimensionality-reduced data is further structured to generate an environmental perception dimensionality-reduced data array. This array is a standardized representation of the dimensionality-reduced data, containing core environmental features and exhibiting low dimensionality and high information content. The data dimensionality reduction module 12 generates a concise yet information-dense environmental perception dimensionality-reduced data array by performing dimensionality reduction on the environmental perception dataset, ensuring that key features are preserved while compressing data dimensionality, providing accurate and efficient data input for subsequent behavior planning of the perception robot.

[0023] The behavior analysis module 13 is used to perform behavior analysis on the sensing robot based on the environmental perception dimensionality reduction data array and generate multiple behavior information.

[0024] Specifically, the behavior analysis module 13 receives the dimensionality-reduced environmental perception data array from the data dimensionality reduction module 12, and performs behavior analysis on the perception robot in conjunction with the robot's task objectives, generating multiple behavioral information items. These behavioral information items refer to the robot's specific behavioral instructions in the current environment, including behavior type, execution conditions, priority, etc., used to guide the robot's behavior. The behavior analysis module 13 can efficiently analyze environmental perception data and task objectives, generating accurate and reliable behavioral information, thereby improving the accuracy and reliability of the perception robot's adaptive behavior planning.

[0025] In one embodiment, such as Figure 2 As shown, the behavior analysis module 13 performs behavior analysis on the sensing robot based on the environmental perception dimensionality reduction data array to generate multiple behavior information. The behavior analysis module 13 is further used to: retrieve the target task; perform turning analysis based on the environmental perception dimensionality reduction data array according to the target task to determine the target direction information; perform path analysis based on the environmental perception dimensionality reduction data array according to the target direction information to obtain a target path map; perform obstacle detection based on the target direction information, the target path map, and the environmental perception dimensionality reduction data array to obtain behavior obstacle information; classify the target direction information, the target path map, and the behavior obstacle information to determine multiple behavior patterns; and add the multiple behavior patterns to the multiple behavior information.

[0026] Specifically, the behavior analysis module 13 first retrieves the current target task of the sensing robot. The target task refers to a specific behavioral task that the sensing robot needs to complete, such as reaching a designated location, path navigation, or avoiding dynamic obstacles. Based on the target task and combined with the environmental perception dimensionality reduction data array, a turning analysis is performed. This turning analysis calculates the optimal movement direction based on the relative position between the robot and the target, ensuring the robot can efficiently move towards the target location. The turning analysis determines the angular deviation between the current robot position and the target position using geometric methods, and simultaneously avoids possible collision paths using environmental data, thereby generating target direction information. This target direction information, including the target azimuth angle and distance, indicates the direction the robot should move next and is an important input for path planning.

[0027] Based on the target direction information and combined with the environmental perception dimensionality reduction data array, the robot performs path analysis from its current starting point to the target point, calculates the robot's travel path in the current environment, ensuring the path is safe, reachable, and efficient, and generates a target path map. The target path map is a graph structure generated through path analysis, describing the path information from the current starting point to the target point. Path analysis is achieved by combining environmental perception data and using either the A* algorithm or Dijkstra's algorithm to traverse the environmental data. Based on path generation, the behavior analysis module 13 performs obstacle detection by combining the target direction information and the target path map. Obstacle detection utilizes feature information about obstacles in the environmental perception dimensionality reduction data array, such as position, size, and dynamics, to analyze objects or areas on the path that may hinder the robot's movement, and marks their positions and characteristics, generating behavioral obstacle information. The obstacle information describes the obstacle information on the path, including the type of obstacle, its location, and its degree of influence on the path. Furthermore, based on the target direction information, the target path map, and the behavioral obstacle information, the possible actions that the sensing robot may perform are classified and mapped to specific behavioral patterns. These behavioral patterns are the types of actions the sensing robot performs in the current environment, such as obstacle avoidance, detour, or target approach. Finally, multiple behavioral patterns are integrated into behavioral information, which is a detailed description of the behavioral patterns, including the action type, specific execution conditions such as trigger signals or path nodes, and the priority of each behavior, used to guide the sensing robot's actions. The behavioral information is output in a structured form, providing a clear execution basis for the behavior planning module. For example, the behavior analysis module 13 generates a set of behavioral information: first avoid obstacles, then drive in a straight line, and finally reach the target point. By combining the target task with environmental perception data, the behavior analysis module 13 can dynamically adapt to environmental changes, generate optimal behavioral information, and ensure the accuracy of behavior classification by comprehensively considering target direction, path planning, and obstacle detection results. It can also generate complete behavioral analysis information, enhancing the flexibility of the sensing robot.

[0028] Furthermore, based on the target direction information, the target path map, and the behavioral obstacle information, multiple behavioral patterns are determined. The behavior analysis module 13 is also used to: extract features from the sensing robot based on the target direction information, the target path map, and the behavioral obstacle information to obtain multiple behavioral features; classify the multiple behavioral features using a support vector machine to determine multiple behavioral types; combine the multiple behavioral types with the environmental perception dimensionality reduction data array to obtain behavior learning results; and perform pattern recognition based on the behavior learning results to determine the multiple behavioral patterns.

[0029] Specifically, based on the target direction information, the target path map, and the behavioral obstacle information, features are extracted from the current environment and task state of the sensing robot to obtain multiple behavioral features. These behavioral features include direction features, path features, and obstacle characteristic information. Direction features include angular deviation of the target direction and distance to the target position; path features include path complexity, number of turning points, and path length; and obstacle characteristics include obstacle position, size, and dynamics (dynamics refers to whether the obstacle is moving). The extracted behavioral features are then input into a machine learning model for classification. This machine learning model is a Support Vector Machine (SVM), which acts as the classifier. The SVM constructs a hyperplane in a high-dimensional feature space to separate different behavior types. For example, straight-line driving, turning and obstacle avoidance, and pausing / waiting are classified into different labels. The SVM classification process includes: mapping behavioral features to a high-dimensional space, finding a hyperplane that maximizes the class margin to ensure classification accuracy, and outputting the behavior type to which each input feature vector belongs. Next, the multiple behavior types are combined with the environmental perception dimensionality-reduced data matrix for behavior learning. Using a data-driven approach, the robot's behavioral decisions under current environmental conditions are obtained. After behavior learning is complete, the results are used to guide behavior pattern recognition. Optionally, based on the classification results, the robot's behavior priorities in different scenarios are adjusted. Through continuous learning and historical data feedback, the classification criteria and applicable scope of behavior types are optimized. Finally, based on the behavior learning results, the correlations between behavior types are identified. Based on these correlations, the logical relationships between behavior patterns are constructed; for example, which behavior types can be combined to form a complete pattern. Multiple behavior patterns are then output for subsequent behavior planning. A behavior pattern refers to a high-level set of behaviors composed of multiple behavior types, representing the robot's action logic under specific conditions. By utilizing feature extraction, support vector machine classification, behavior learning, and pattern recognition, environmental and task data are transformed into multiple specific behavior patterns, thereby dynamically guiding the robot to adopt adaptive behavior strategies in different scenarios, achieving efficient and accurate task execution.

[0030] The behavior tree construction module 14 is used to combine the multiple behavioral information to construct a behavior tree.

[0031] Specifically, the behavior tree construction module 14 receives multiple behavioral information entries, including multiple behavior types, triggering conditions, and execution priorities. It then performs logical analysis on these behavioral information entries to determine the relationships between behaviors. Based on task requirements and environmental characteristics, it combines related behavioral information; for example, obstacle avoidance behavior might consist of two specific behavioral entries: turning obstacle avoidance and detour obstacle avoidance. A behavior tree is then constructed. This hierarchical structure flexibly describes various behavioral paths of the sensing robot under different environmental and task conditions. By generating the behavior tree, support is provided for the sensing robot's behavior planning, enabling the robot to complete task objectives efficiently and in real-time in complex environments.

[0032] Furthermore, the multiple behavioral information is combined to construct a behavior tree. The behavior tree construction module 14 is also used to: perform behavioral association analysis based on the multiple behavioral information to obtain multiple association coefficients; construct multiple association edges according to the multiple association coefficients and the multiple behavioral information; map the multiple behavioral information to multiple behavior tree nodes; and connect the multiple behavior tree nodes according to the multiple association edges to construct the behavior tree.

[0033] Specifically, firstly, behavioral correlation analysis is performed on multiple behavioral information items. By evaluating the dependencies and logical relationships between behaviors, multiple correlation coefficients are generated. Correlation coefficients are numerical values ​​used to quantify the strength of the correlation between two behaviors, ranging from 0 (no correlation) to 1 (complete correlation). These coefficients are calculated based on temporal correlation, spatial correlation, and conditional correlation. Temporal correlation refers to whether one behavior requires the completion of another behavior to execute; spatial correlation refers to whether the behaviors involve adjacent or identical spatial regions; and conditional correlation refers to whether the behaviors share the same triggering conditions, such as target direction or obstacle information. Then, based on the behavioral correlation analysis results, multiple correlation edges are constructed using these correlation coefficients to connect related behavioral information items. Correlation edges are the basic structure of the behavior tree, representing the logical or conditional relationships between behavioral nodes. When the correlation coefficient is higher than a set threshold, a correlation edge is added between two behavioral information items. The weight of the edge is determined by the correlation coefficient, representing the strength of the correlation between the behaviors. For example, if a condition node detecting an obstacle has a high correlation with a turning obstacle avoidance behavior node, a correlation edge is added between them. Finally, multiple behavioral information items are mapped to behavior tree nodes, that is, the behavioral information is transformed into nodes in the behavior tree and assigned corresponding attributes. Behavior tree nodes include behavior nodes and condition nodes. Behavior nodes represent specific behavioral information, such as straight-line driving or obstacle avoidance, which are the actions ultimately executed by the sensing robot. Condition nodes represent specific conditions that trigger the behavior, such as detecting an obstacle, used to determine whether to execute the corresponding behavior. Furthermore, the behavior tree nodes are logically connected according to multiple constructed association edges to form a complete behavior tree structure. The behavior tree is a hierarchical structure, with the root node representing the main task objective, branch nodes representing different behavioral paths, and leaf nodes representing specific behavioral actions. By constructing the behavior tree, a complete logical framework for adaptive behavior planning of the robot is provided, guiding the robot to efficiently and accurately complete task objectives in complex and dynamic environments.

[0034] Furthermore, the multiple behavioral information is mapped to multiple behavioral tree nodes, and the multiple behavioral tree nodes are connected according to the multiple associated edges to construct the behavioral tree. The behavioral tree construction module 14 is also used to: analyze the multiple behavioral information according to the multiple behavioral types according to the environmental perception dimensionality reduction data array, and set multiple behavioral triggering conditions; allocate the multiple behavioral information according to the multiple behavioral triggering conditions, and set execution priorities; map the multiple behavioral information to multiple behavioral nodes according to the execution priorities, and map the multiple behavioral triggering conditions to multiple condition nodes according to the execution priorities; match the multiple behavioral nodes with the multiple condition nodes to determine the multiple behavioral tree nodes.

[0035] Specifically, multiple behavioral information is analyzed based on the environmental perception dimensionality reduction data array. Multiple behavioral trigger conditions are set according to different behavior types. A behavioral trigger condition refers to the environmental or state conditions that a behavior must meet; it is the basis for activating a behavior node and defines the conditions under which a certain behavior of the perception robot can be activated. For example, a target direction deviation less than a certain value can be used as a trigger condition for straight-line behavior, and the detection of an obstacle can be used as a trigger condition for turning and obstacle avoidance behavior. By analyzing the logical attributes of the behavioral information and the characteristics of the environmental data, the accuracy and comprehensiveness of the trigger conditions are ensured. After the trigger conditions are set, multiple behavioral information are allocated according to their corresponding trigger conditions. Based on the importance of the behavioral information and the task objective, an execution priority is set for each behavioral information. Behaviors with higher priority have a higher execution order in the behavior tree. Optionally, the priority setting can be based on the urgency and importance of the behavioral information. The execution priority indicates the order in which behavioral information is executed in the behavior tree, determined by the urgency and importance of the task. According to the allocation results, multiple behavioral information are mapped to multiple behavior nodes according to their execution priorities, and the corresponding behavioral trigger conditions are mapped to condition nodes. The mapping between behavior nodes and condition nodes is based on the trigger conditions and execution logic described in the behavioral information. For example, the condition node corresponding to an obstacle avoidance behavior node might be "obstacle distance less than 0.5 meters". Finally, based on priority and logical relationships, behavior nodes are matched with condition nodes to determine multiple behavior tree nodes. These nodes are then connected by associative edges according to the execution order to construct a complete behavior tree. The connection rules include: condition nodes are connected to their corresponding behavior nodes via associative edges, indicating the behavior executed after the trigger condition is met; higher-priority behavior nodes are placed earlier in the tree to ensure priority execution; different branches represent different behavior paths to handle different environmental or task situations. Through the analysis of behavior information, priority setting, node mapping, and logical connections, a hierarchical behavior tree structure is generated, improving the flexibility of adaptive behavior planning for the perception robot and ensuring that the perception robot can efficiently and flexibly complete adaptive tasks in diverse scenarios.

[0036] The behavior planning module 15 is used to train a strategy by combining the behavior tree with the environmental perception dimensionality reduction data array, generate a behavior strategy, and execute the behavior strategy to perform adaptive behavior planning for the perception robot.

[0037] Specifically, the behavior planning module 15 is the core module for the perceptual robot to achieve adaptive behavior planning. It traverses the behavior tree to evaluate the applicability and execution probability of each behavior node in the current environment. During the traversal process, it combines the environmental perception dimensionality reduction data array for policy training, continuously adjusting and optimizing the nodes and paths in the behavior tree to generate behavior strategies adapted to the current environment. Once a behavior strategy is generated, it is executed to guide the perceptual robot in adaptive behavior planning. By combining the behavior tree with the environmental perception dimensionality reduction data array, the behavior planning module 15 achieves efficient and intelligent behavior planning for the perceptual robot, enabling it to make reasonable decisions and take corresponding actions in complex and ever-changing environments, efficiently completing task objectives.

[0038] Furthermore, by combining the behavior tree with the environmental perception dimensionality reduction data array for policy training, a behavior policy is generated. The behavior planning module 15 is also used to: map the target task to the behavior tree for node traversal to determine multiple behavior nodes; introduce a reward function to determine whether the multiple behavior nodes conform to the expected behavior node interval according to the environmental perception dimensionality reduction data array; if all the multiple behavior nodes conform to the expected behavior node interval, then calculate according to the reward function to generate positive reward data; if any of the multiple behavior nodes does not conform to the expected behavior node interval, then calculate according to the reward function to generate negative reward data; and update the behavior tree based on the positive reward data or the negative reward data to formulate the behavior policy.

[0039] Specifically, the current target task of the perception robot is mapped to a behavior tree, transforming it into specific nodes within the behavior tree and clarifying the execution path. The behavior tree is traversed, starting from the root node and proceeding downwards based on the current environmental state and the requirements of the target task, identifying multiple behavior nodes. During this traversal, the perception robot performs corresponding operations or makes conditional judgments based on the type of behavior node. To optimize the behavior strategy, a reward function is introduced during the traversal of the behavior tree nodes to quantitatively evaluate the state and execution result of each node. Based on the input of the environmental perception dimensionality reduction data array, it is determined whether each node is within the expected behavior node range, which is the execution range of the behavior node under ideal conditions. The reward function is used to evaluate the execution result of the behavior node. Rewards guide strategy optimization, including positive and negative rewards. A positive reward is generated when the execution result of a behavior node meets the task objective or its state satisfies expected conditions; for example, a high positive reward is given when obstacle avoidance behavior successfully bypasses an obstacle. When the execution result of a certain behavior node deviates from the task objective, or its state does not meet the expected conditions, a negative reward is generated. For example, when the straight-line driving behavior causes a deviation from the target direction, a negative reward is given.

[0040] Based on the states of multiple behavior nodes and the reward function, reward data is calculated and classified as positive or negative rewards. Positive reward data is generated when all traversed behavior nodes conform to the expected behavior node interval. For example, if the robot successfully completes all obstacle avoidance behaviors and reaches the target point, positive reward data is generated. If any behavior node fails to meet the expected behavior node interval, negative reward data is generated according to the reward function. For example, if the robot fails to complete path planning within the specified time, negative reward data is generated. Positive reward data indicates that the behavior node's execution result meets expectations and is used to strengthen the weight of related nodes in the behavior tree. Negative reward data indicates that the behavior node's execution result does not meet expectations and is used to adjust the strategy or weights. The behavior tree is updated based on the reward data to optimize the behavior strategy. Optionally, for behavior nodes that generate positive rewards, their execution priority is increased or their triggering conditions are adjusted to make them easier to select. For behavior nodes that generate negative rewards, their execution priority is decreased or their associated conditions are adjusted to avoid repeating errors. Behavioral path planning is optimized by adjusting the weights of associated edges; for example, paths with high positive rewards are set as the primary execution paths. Finally, specific behavior strategies are generated based on the optimized behavior tree. These strategies are then output and applied to the robot's real-time behavior planning, ensuring that the perceptual robot can flexibly and efficiently complete its target tasks in the current environment.

[0041] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0042] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. An adaptive behavior planning system for a sensing robot, characterized in that, The system includes: The environmental perception module is used to extract multiple sensors based on the perception robot, perform environmental perception through the multiple sensors, and obtain an environmental perception dataset. The data dimensionality reduction module is used to preprocess the environmental perception dataset, reduce the dimensionality of the environmental perception dataset based on the preprocessing results, and generate an environmental perception dimensionality reduction data array. The behavior analysis module is used to perform behavior analysis on the sensing robot based on the environmental perception dimensionality reduction data array, generating multiple behavior information; the behavior analysis module is also used for: The target task is retrieved, and based on the environmental perception dimensionality reduction data array, a turning analysis is performed according to the target task to determine the target direction information; Based on the environmental perception dimensionality reduction data array, path analysis is performed according to the target direction information to obtain the target path map; Based on the target direction information, obstacle detection is performed according to the target path map and the environmental perception dimensionality reduction data array to obtain behavioral obstacle information; Based on the target direction information, the target path map, and the behavioral obstacle information, multiple behavioral patterns are classified and determined. Add the multiple behavioral patterns to the multiple behavioral information; The behavior analysis module is also used for: Based on the target direction information, the target path map, and the behavioral obstacle information, the perception robot is subjected to feature extraction to obtain multiple behavioral features; Support vector machines are used to classify based on the aforementioned multiple behavioral features to determine multiple behavioral types; The multiple behavior types are combined with the environmental perception dimensionality reduction data array for learning to obtain behavior learning results; Based on the behavior learning results, pattern recognition is performed to determine the multiple behavior patterns; The behavior tree construction module is used to combine the multiple behavioral information to construct a behavior tree; the behavior tree construction module is also used for: Based on the aforementioned multiple behavioral information, behavioral correlation analysis is performed to obtain multiple correlation coefficients; Multiple association edges are constructed based on the multiple association coefficients and the multiple behavioral information; The multiple behavioral information is mapped to multiple behavioral tree nodes, and the multiple behavioral tree nodes are connected according to the multiple associated edges to construct the behavioral tree; The behavior tree construction module is also used for: The environmental perception dimensionality reduction data array is used to analyze the multiple behavioral information according to the multiple behavioral types, and multiple behavioral triggering conditions are set. The multiple behavioral information items are allocated according to the multiple behavioral triggering conditions, and execution priorities are set; The multiple behavior information is mapped to multiple behavior nodes according to the execution priority, and the multiple behavior triggering conditions are mapped to multiple condition nodes according to the execution priority; The multiple behavior nodes are matched with the multiple condition nodes to determine the multiple behavior tree nodes; The behavior planning module is used to train a strategy by combining the behavior tree with the environmental perception dimensionality reduction data array, generate a behavior strategy, and execute the behavior strategy to perform adaptive behavior planning for the perception robot.

2. The adaptive behavior planning system for a sensing robot as described in claim 1, characterized in that, The data dimensionality reduction module is also used for: Multidimensional analysis is performed based on the environmental perception dataset to determine multidimensional environmental state signals. The multidimensional environmental state signal is decomposed into wavelet coefficients to obtain the multidimensional environmental state signal wavelet coefficients. Threshold quantization is performed on the wavelet coefficients of the multidimensional environmental state signal to determine the wavelet selection threshold of the multidimensional environmental state signal. The wavelet coefficients of the multidimensional environmental state signal are truncated according to the wavelet selection threshold of the multidimensional environmental state signal, and the noise signals smaller than the wavelet selection threshold of the multidimensional environmental state signal are set to zero to obtain effective signal information greater than the wavelet selection threshold of the multidimensional environmental state signal. The effective signal information is filtered and reconstructed to obtain the preprocessing result.

3. The adaptive behavior planning system for a sensing robot as described in claim 2, characterized in that, The data dimensionality reduction module is also used for: Based on the wavelet selection threshold of the multidimensional environmental state signal, the distribution of the environmental perception dataset is calculated to obtain a data distribution probability array. A similarity probability distribution function is constructed, and the dimensionality of the point distribution of the data distribution probability array is reduced to generate the data dimensionality reduction result. The environmental perception dimensionality reduction data array is constructed based on the dimensionality reduction results of the data.

4. The adaptive behavior planning system for a sensing robot as described in claim 1, characterized in that, The behavior planning module is also used for: The target task is mapped to the behavior tree, and node traversal is performed to determine multiple behavior nodes; A reward function is introduced, and based on the environmental perception dimensionality reduction data array, it is determined whether the multiple behavioral nodes conform to the expected behavioral node range; If all of the multiple behavior nodes conform to the expected behavior node range, then positive reward data is generated according to the reward function. If any of the multiple behavioral nodes does not conform to the expected behavioral node range, then negative reward data is generated according to the reward function. The behavior tree is updated based on the positive reward data or the negative reward data to formulate the behavior strategy.

Images