Robot control method and device, computer device and storage medium

Abstract

The application belongs to the technical field of artificial intelligence, and relates to a robot control method and device, computer equipment and a storage medium, comprising: collecting full-link behavior data of a robot in a historical time period; filtering target feature data related to a robot task from the full-link behavior data; training a preset decision tree algorithm based on the target feature data to obtain a trained target decision tree, and generating a corresponding interpretable decision package based on the target decision tree; constructing a corresponding risk feature library based on the interpretable decision package; obtaining behavior data of the robot in real time; performing risk assessment processing on the behavior data based on the risk feature library to generate a corresponding risk assessment result; and performing corresponding control processing on the robot based on the risk assessment result. The application can be applied to robot control scenes in the fields of financial technology and digital medicine, can realize early intervention and active control of safety risks, and effectively improves the safety and controllability of the robot.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology and can be applied to fields such as fintech and digital healthcare, particularly to robot control methods, devices, computer equipment, and storage media. Background Technology

[0002] Despite the rapid development of robotics technology, robots still face significant limitations in human-robot interaction and autonomous decision-making, generally operating in a passive "remedial after failure" mode. This mode lacks proactive early warning capabilities based on behavioral data, making it impossible to anticipate and effectively control potential risks, resulting in low safety and controllability of robots.

[0003] In the financial and insurance sector, insurance claims review robots face similar challenges. Traditional claims review methods rely on human experience and simple rules. Robots only begin processing claims after receiving a clear claim application with complete documentation. They lack the ability to proactively identify and warn of potential problems such as fraud or incomplete documentation. For example, in auto insurance claims, if a customer submits false information or missing key details, traditional review robots struggle to detect this early on, often revealing the problem only later in the review process. This not only increases review costs and time but can also potentially lead to financial losses for insurance companies.

[0004] In the field of digital healthcare, taking elderly care robots as an example, they can only trigger alarms after obvious safety events such as falls and collisions occur. They cannot detect and take preventive measures in advance for some potential risk factors that may cause such events, such as abnormal body posture of the elderly or deviation from the normal movement trajectory, making it difficult to ensure the safety of the elderly.

[0005] Therefore, there is an urgent need for a robotics technology with proactive early warning capabilities based on behavioral data, in order to enhance the initiative and effectiveness of robots in human-computer interaction and autonomous decision-making, and to ensure the safety and reliability of applications in various fields. Summary of the Invention

[0006] The purpose of this application is to provide a robot control method, device, computer equipment, and storage medium to solve the technical problem that existing robots lack proactive early warning capabilities based on behavioral data, making it impossible to provide early warning and effective control of potential risks, resulting in low safety and controllability of robots.

[0007] Firstly, a robot control method is provided, including: Collect end-to-end behavior data of the robot over a historical time period; Target feature data related to the robot task are filtered out from the full-link behavior data; The preset decision tree algorithm is trained based on the target feature data to obtain a trained target decision tree, and a corresponding interpretable decision package is generated based on the target decision tree. A corresponding risk feature library is constructed based on the interpretable decision package; Real-time acquisition of the robot's behavior data; Based on the risk feature library, the behavioral data is processed for risk assessment to generate corresponding risk assessment results; Based on the risk assessment results, the robot is subjected to corresponding control processing.

[0008] Secondly, a robot control device is provided, comprising: The data acquisition module is used to collect the robot's full-link behavior data over a historical time period; The filtering module is used to filter out target feature data related to the robot task from the full-link behavior data; The processing module is used to train a preset decision tree algorithm based on the target feature data to obtain a trained target decision tree, and to generate a corresponding interpretable decision package based on the target decision tree. The construction module is used to build a corresponding risk feature library based on the interpretable decision package; The acquisition module is used to acquire the robot's behavior data in real time; The assessment module is used to perform risk assessment processing on the behavioral data based on the risk feature library and generate corresponding risk assessment results; The control module is used to perform corresponding control processing on the robot based on the risk assessment results.

[0009] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the robot control method described above.

[0010] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the robot control method described above.

[0011] In the above-mentioned robot control method, device, computer equipment, and storage medium, the following steps are taken: First, the robot's full-link behavior data over a historical time period is collected; then, target feature data related to the robot's task is filtered out from the full-link behavior data; next, a preset decision tree algorithm is trained based on the target feature data to obtain a trained target decision tree, and a corresponding interpretable decision package is generated based on the target decision tree; then, a corresponding risk feature library is constructed based on the interpretable decision package; subsequently, the robot's behavior data is acquired in real time; further, risk assessment processing is performed on the behavior data based on the risk feature library to generate a corresponding risk assessment result; finally, corresponding control processing is performed on the robot based on the risk assessment result. Based on the above automated processing flow, this application collects the robot's full-link behavior data over a historical time period, then trains a preset decision tree algorithm based on target feature data related to the robot's task selected from the full-link behavior data to obtain a trained target decision tree, generates an interpretable decision package based on the target decision tree, constructs a risk feature library based on the interpretable decision package, and then performs risk assessment processing on the behavior data based on the use of the risk feature library. It also automatically performs corresponding control processing on the robot according to the generated risk assessment results, thereby realizing early intervention and proactive management of safety risks and effectively improving the robot's safety and controllability. Attached Figure Description

[0012] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is an exemplary system architecture diagram to which this application can be applied; Figure 2 This is a flowchart of one embodiment of the robot control method according to this application; Figure 3 This is a schematic diagram of one embodiment of the robot control device according to this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0014] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0015] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0016] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0017] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0018] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0019] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0020] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0021] It should be noted that the robot control method provided in this application embodiment is generally executed by a server / terminal device, and correspondingly, the robot control device is generally set in the server / terminal device.

[0022] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0023] Continue to refer to Figure 2 A flowchart illustrating an embodiment of the robot control method according to this application is shown. The order of steps in the flowchart can be changed, and some steps can be omitted, depending on different requirements. The robot control method provided in this application embodiment can be applied to any scenario requiring robot control processing, and therefore can be applied to products in these scenarios, such as robot control products in the financial insurance field or the digital healthcare field. The robot control method includes the following steps: Step S201: Collect the robot's full-link behavior data within the historical time period.

[0024] In this embodiment, the robot control method operates on electronic devices (e.g., Figure 1The server / terminal device shown can acquire the robot's full-link behavior data over a historical time period via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future-developed wireless connection methods. The implementing entity of this application can be a robot control system, which can be simply referred to as the system. This application can be applied to robot control scenarios in the fields of fintech and digital healthcare. For example, in the fintech field, robots can automatically handle tasks such as customer service and financial report review by analyzing image and text information. In the digital healthcare scenario, robots can assist doctors in medical image analysis and provide more accurate analytical support by combining patient medical record information. The numerical selection of the aforementioned historical event period is not specifically limited and can be determined according to actual business needs, for example, it can be set to the past two years.

[0025] Furthermore, the specific implementation process of collecting the robot's full-link behavior data over the historical time period will be described in more detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0026] Step S202: Filter out target feature data related to the robot task from the full-link behavior data.

[0027] In this embodiment, target feature data related to the robot's core tasks is extracted from the end-to-end behavioral data pool. This includes perception layer data and corresponding safety-compliant execution action data, with perception layer data being a crucial component of the data selection process. Taking a robot monitoring vital signs in medical care as an example, perception layer data consists of the patient's physiological sensor data, such as heart rate, blood pressure, and respiratory rate. This data reflects the patient's current physiological state and is a vital basis for the robot's decision-making. Only by acquiring accurate and comprehensive perception layer data can the model understand the actual situation and learn appropriate execution actions to take under different physiological states. Similarly, using a medical care robot as an example, execution action data represents the robot's decisions and execution results after receiving the patient's physiological parameters, such as whether to trigger an alarm or adjust the monitoring frequency. This data reflects the robot's correct and safety-compliant responses under specific perception layer data inputs. The model needs to learn the relationship between these inputs (perception layer data) and outputs (execution action data) to master the actions to take in different situations, thereby enabling it to make correct decisions when encountering similar situations in the future.

[0028] Step S203: Train the preset decision tree algorithm based on the target feature data to obtain a trained target decision tree, and generate a corresponding interpretable decision package based on the target decision tree.

[0029] In this embodiment, the training process of the target decision tree includes: 1) Determining input features and output results: The perception layer data is used as the input features of the decision tree, and the safe and compliant execution actions are used as the output results. For example, in a medical care scenario, the input features can be physiological parameters such as the patient's heart rate, blood pressure, and respiratory rate, and the output results can be execution actions such as "normal monitoring", "triggering an emergency alarm", and "adjusting the monitoring frequency". 2) Selecting a decision tree algorithm: A suitable decision tree algorithm, such as ID3, C4.5, or CART, is selected based on the characteristics of the data and the task requirements. These algorithms select the optimal feature partitioning method by calculating indicators such as information gain, information gain ratio, or Gini index to construct a decision tree model. 3) Training the decision tree model: The selected target feature data is used to train the decision tree algorithm. During the training process, the algorithm continuously adjusts the node partitioning and branch structure of the decision tree according to the relationship between the input features and the output results until the preset stopping conditions are reached (such as the maximum depth of the tree, the minimum number of samples per node, etc.), and a trained target decision tree is obtained. For example, when training a decision tree model for a medical companion robot, the algorithm continuously optimizes the decision tree partitioning rules based on the robot's correct decision results under different combinations of heart rate and respiratory rate.

[0030] The specific implementation process of generating the corresponding interpretable decision package based on the target decision tree described above will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0031] Step S204: Construct a corresponding risk feature library based on the interpretable decision package.

[0032] In this embodiment, the specific implementation process of constructing the corresponding risk feature library based on the interpretable decision package will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0033] Step S205: Acquire the robot's behavior data in real time.

[0034] In this embodiment, the aforementioned behavioral data refers to the robot's behavioral data collected in real time, including real-time collected perception layer data, decision layer data, and execution layer data. The specific collection process can be referred to later in the description of the processing of perception layer data, decision layer data, and execution layer data collected by the multi-source data acquisition module within a historical time period; further details will not be elaborated here.

[0035] Step S206: Perform risk assessment processing on the behavioral data based on the risk feature library to generate corresponding risk assessment results.

[0036] In this embodiment, the specific implementation process of performing risk assessment processing on the behavioral data based on the risk feature library to generate the corresponding risk assessment results will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0037] Step S207: Perform corresponding control processing on the robot based on the risk assessment results.

[0038] In this embodiment, the risk assessment results can include high risk, medium risk, and low risk. Corresponding control measures can be taken for different risk levels corresponding to the risk assessment results to perform corresponding control processing on the robot. Specifically, this includes: High-risk control: For high-risk situations, emergency control measures are immediately triggered. For example, in cases of high human-robot collision risk, the robot immediately executes an emergency stop command, ceasing all movement to avoid a collision. Simultaneously, an emergency alarm message is sent to relevant personnel (such as operators and managers) to notify them to handle the situation promptly. Medium-risk control: For medium-risk situations, function degradation measures are initiated. For example, if sensor data drift causes the robot's decisions to be inaccurate but does not immediately cause serious consequences, the robot can suspend high-risk tasks, such as suspending precision assembly operations, and instead perform simpler, lower-risk tasks, while simultaneously calibrating and checking the sensors. Low-risk control: For low-risk situations, early warning messages are issued. For example, if the robot's movements have slight deviations that do not affect the normal completion and safety of the task, an early warning message is sent to the operator, reminding them to pay attention to the robot's operating status so that problems can be identified and adjusted in a timely manner.

[0039] This application first collects end-to-end behavior data of the robot within a historical time period; then filters out target feature data related to the robot's task from the end-to-end behavior data; then trains a preset decision tree algorithm based on the target feature data to obtain a trained target decision tree, and generates a corresponding interpretable decision package based on the target decision tree; then constructs a corresponding risk feature library based on the interpretable decision package; subsequently acquires the robot's behavior data in real time; further performs risk assessment processing on the behavior data based on the risk feature library to generate a corresponding risk assessment result; finally, performs corresponding control processing on the robot based on the risk assessment result. Based on the above automated processing flow, this application collects the robot's full-link behavior data over a historical time period, then trains a preset decision tree algorithm based on target feature data related to the robot's task selected from the full-link behavior data to obtain a trained target decision tree, generates an interpretable decision package based on the target decision tree, constructs a risk feature library based on the interpretable decision package, and then performs risk assessment processing on the behavior data based on the use of the risk feature library. It also automatically performs corresponding control processing on the robot according to the generated risk assessment results, thereby realizing early intervention and proactive management of safety risks and effectively improving the robot's safety and controllability.

[0040] In some alternative implementations, step S201 includes the following steps: Call the preset multi-source data acquisition module.

[0041] In this embodiment, the aforementioned multi-source data acquisition module is a pre-deployed functional module used for collecting and processing multi-dimensional behavioral data related to the robot.

[0042] The multi-source data acquisition module collects perception layer data, decision layer data, and execution layer data of the robot within a historical time period.

[0043] In this embodiment, the data acquisition of the perception layer includes: Visual sensors: High-definition cameras are installed on key parts of the robot (such as the head, end effector of the robotic arm, etc.), and appropriate shooting frequencies and resolutions are set according to the robot's application scenario. For example, in industrial collaborative robots, the camera captures images of the surrounding environment at a frequency of 30 frames per second to identify information such as obstacles and workpiece positions within the work area. Auditory sensors: High-sensitivity microphones are installed to capture voice commands and sound signals from the surrounding environment. For example, in medical companion robots, the microphone can receive voice commands from patients or medical staff, such as "call a doctor" or "adjust the bed height." Tactile sensors: Force feedback sensors are installed on contact points such as the robot's robotic arm and fingers to sense the force applied when contacting objects in real time. In industrial assembly tasks, tactile sensors can detect the pressure between parts during assembly, ensuring appropriate assembly force and avoiding damage to parts. Physiological sensors: When the robot interacts with humans, physiological sensors are installed on interaction points (such as hands, seats, etc.) to collect vital sign data during human-robot interaction. For example, in rehabilitation training robots, physiological sensors worn on the patient's limbs can collect data such as the patient's heart rate and electromyographic signals to assess the effectiveness of rehabilitation training and the patient's physical condition.

[0044] Data acquisition at the decision-making level includes: Algorithm input parameters: Recording various input parameters used by the robot's decision-making algorithm. These parameters may come from the processing results of data from the perception layer or be pre-set fixed parameters. For example, in an autonomous navigation robot, algorithm input parameters may include target position coordinates, current position coordinates, map information, etc. Intermediate decision variables: Recording key intermediate variable values ​​during the operation of the decision-making algorithm. These variables reflect the internal state and calculation process of the algorithm during the decision-making process. For example, in a path planning algorithm, intermediate decision variables may include the evaluation scores of each candidate path, node information on the path, etc. Final decision result: Clearly recording the final decision made by the robot based on the decision-making algorithm, such as direction of movement, action type, etc. For example, in a handling robot, the final decision result may be "go to location A to handle goods".

[0045] Execution layer data acquisition includes: Joint motion angles: Angle sensors are installed at each joint of the robot to measure the joint motion angles in real time. This angle data allows for precise control of the robot's posture and trajectory. For example, in a six-axis industrial robot, angle sensors at each joint can accurately measure the joint's rotation angle, enabling complex industrial operations. Motion execution force: Force or torque sensors are used to measure the force exerted by the robot when performing actions. In welding robots, torque sensors measure the torque applied by the robotic arm to the welding torch during welding, ensuring stable welding quality. Drive system energy consumption: Energy monitoring equipment is installed to monitor the robot's drive system energy consumption in real time, including parameters such as current, voltage, and power. This helps assess the robot's energy efficiency and identify energy waste. For example, in an electric mobile robot, real-time energy consumption is calculated by monitoring the battery's output current and voltage. Task completion status: The completion status of each robot task is recorded, including whether the task was successfully completed, the completion time, and whether any abnormalities occurred. In logistics sorting robots, task completion status records whether goods were successfully sorted to the designated location and the time taken for sorting.

[0046] The perception layer data, the decision layer data, and the execution layer data are integrated and processed to obtain the corresponding first processed data.

[0047] In this embodiment, the collected perception layer data, decision layer data, and execution layer data are integrated and processed to form an integrated data record, namely the first processed data mentioned above.

[0048] The first processed data is optimized based on a preset standardization strategy to obtain the corresponding second processed data.

[0049] In this embodiment, the specific implementation process of optimizing the first processed data based on the preset standardization strategy to obtain the corresponding second processed data will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0050] The second processed data is used as the end-to-end behavior data.

[0051] This application utilizes a pre-defined multi-source data acquisition module to collect sensory, decision-making, and execution layer data from the robot over a historical time period. This data is then integrated to obtain first-processed data. Following this, a pre-defined standardization strategy is applied to optimize the first-processed data, resulting in second-processed data. This second-processed data is then used as the final end-to-end behavioral data. Based on this process, this application, by using the multi-source data acquisition module to collect data from the sensory, decision-making, and execution layers, can comprehensively record the robot's various states and decision-making processes during operation. Furthermore, the standardized strategy optimizes the integrated first-processed data to form the final end-to-end behavioral data, effectively improving the richness and standardization of the obtained end-to-end behavioral data.

[0052] In some optional implementations of this embodiment, the step of optimizing the first processed data based on a preset standardization strategy to obtain the corresponding second processed data includes the following steps: Invoke the preset data association rules.

[0053] In this embodiment, the above data association rule is an association rule that uses timestamp synchronization technology to associate multi-dimensional data.

[0054] Based on the data association rules, the first processed data is associated to obtain the corresponding third processed data.

[0055] In this embodiment, precise timestamps are added to each collected data point, achieving millisecond-level accuracy. Timestamps can be generated using high-precision clock sources (such as atomic clocks or high-precision crystal oscillators). Furthermore, during data transmission and storage, the timestamps are ensured to be transmitted and saved along with the data. When data from different dimensions arrives at the data processing center, they are correlated based on their timestamps. For example, perception layer image data, decision layer algorithm input parameters, and execution layer joint motion angle data collected at the same time are correlated to form a complete data record reflecting the robot's state and decision-making execution at that moment.

[0056] The third processed data is encapsulated based on a preset data format to obtain the corresponding fourth processed data.

[0057] In this embodiment, the selection of the above data format is not specifically limited and can be determined according to actual business needs. For example, JSON-LD (JSON for Linking Data) can be selected. JSON-LD is a JSON-based data format with good readability and scalability, and supports semantic annotation, which facilitates data exchange and sharing.

[0058] Next, the associated multi-dimensional data is encapsulated in JSON-LD format. The data structure is defined, clarifying the meaning and data type of each field. For example, a JSON-LD object containing data from the perception layer, decision layer, and execution layer is defined. The perception layer data can include multiple sub-fields such as visual, auditory, tactile, and physiological sensor data, and each sub-field contains corresponding specific data and descriptive information.

[0059] The fourth processed data is marked based on preset identification information to obtain the corresponding fifth processed data.

[0060] In this embodiment, the aforementioned identification information includes a unique robot identifier and information about the entity operating the robot. Each robot device can be assigned a unique identifier, such as its MAC address or serial number. This unique identifier is embedded in the encapsulated data to accurately identify the device from which the data originates during subsequent data processing and analysis. Furthermore, the entity operating the robot, such as the user ID and administrator privilege level, is recorded. If the robot is operated through a user interface, the user ID is recorded upon login; if an administrator performs system configuration or maintenance operations, the administrator's privilege level is recorded. This information is also embedded in the data to facilitate the traceability and auditing of operational behaviors.

[0061] The fifth processed data is used as the second processed data.

[0062] In this embodiment, all data processed as described above is stored in a centralized database or data storage system, forming a full-link behavioral data pool. This data pool can be a relational database or a non-relational database, selected based on the characteristics of the data and access requirements. The data pool is managed and maintained, including data backup, recovery, and index creation, ensuring data security and accessibility. Simultaneously, a data query interface is established to facilitate subsequent processes' access to and analysis of the data.

[0063] This application invokes preset data association rules; based on these rules, it performs association processing on the first processed data to obtain corresponding third processed data; then, it encapsulates the third processed data according to a preset data format to obtain corresponding fourth processed data; subsequently, it marks the fourth processed data based on preset identification information to obtain corresponding fifth processed data; finally, it uses the fifth processed data as the second processed data. Based on the above processing flow, this application ensures accurate temporal association of data from different dimensions by using data association rules, enabling accurate analysis of the robot's behavior at specific moments. Standardized data format encapsulation and embedding of robot device identification and operating entity information improves data readability, exchangeability, and traceability, facilitating data management and analysis throughout the system. The resulting end-to-end behavioral data can serve as a data warehouse, providing rich data resources for subsequent technical processes.

[0064] In some alternative implementations, step S203 includes the following steps: Generate a visual decision path corresponding to the target decision tree.

[0065] In this embodiment, after completing model training and obtaining the target decision tree, the model structure of the target decision tree is presented in a visual manner to obtain the corresponding visualized decision path. Specifically, a graphical tool can be used to draw the nodes and branches of the target decision tree as graphics. Each node represents a feature judgment condition, each branch represents a judgment result, and the leaf nodes represent the final decision output. For example, when visualizing the decision tree of a medical companion robot, it can be seen that starting from the root node, the tree is gradually divided according to different value ranges of heart rate and respiratory rate, eventually reaching different leaf nodes, corresponding to different execution actions.

[0066] Obtain the decision results of the target decision tree.

[0067] In this embodiment, during the training process of the target decision tree, the decision results are generated naturally as the model structure is determined. This generation is primarily based on the following steps and principles: 1. Data Partitioning and Rule Construction. The decision tree algorithm starts from the root node and partitions the dataset based on input features. It traverses all features, searching for the feature that best distinguishes different output results (i.e., decision categories) as the partitioning criterion. For example, in a medical companion robot scenario, the algorithm analyzes features such as heart rate and respiratory rate to find which feature, under different values, most clearly distinguishes the robot's correct decision outcome (such as taking emergency measures or normal monitoring). Once the partitioning features and partition points are determined, the dataset is divided into two or more subsets, each corresponding to a branch. This process is recursively performed, continuing to search for the optimal partitioning feature in each subset until a preset stopping condition is met, such as the tree reaching its maximum depth or the minimum number of samples in a node reaching a threshold.

[0068] 2. Leaf nodes determine the decision outcome. A leaf node is formed when the stopping condition is met. Each leaf node corresponds to a specific decision outcome. This decision outcome is determined based on the statistical distribution of the output outcomes of the training samples contained in that leaf node. Typically, a majority voting method is used, meaning the output outcome category with the most samples in the leaf node is taken as the decision outcome for that leaf node. For example, in the decision tree of a medical companion robot, if a leaf node contains training samples in which the robot should take emergency measures in most cases, then the decision outcome for that leaf node is to take emergency measures.

[0069] In this way, the entire decision path from the root node to the leaf node constitutes a complete decision rule. For any new input data, as long as the decision tree is traversed downwards according to the node judgment conditions, the decision result corresponding to the final leaf node is the model's decision on that input data.

[0070] Obtain the key feature information corresponding to the target decision tree.

[0071] In this embodiment, the weights of each feature can be calculated using the feature importance evaluation methods (such as information gain, Gini coefficient, etc.) provided by the decision tree algorithm of the target decision tree described above. For example, in Python's scikit-learn library, after training the decision tree model, the weights of each feature can be directly obtained using the `feature_importances_` attribute. Then, based on the weights, key features are selected according to a certain proportion (e.g., selecting the top 20% of features by weight) or by setting a weight threshold (e.g., features with a weight greater than 0.1) to serve as the aforementioned key feature information.

[0072] The visualized decision path, the decision result, and the key feature information are integrated to obtain the corresponding integrated information.

[0073] In this embodiment, the obtained visualized decision path, decision results, and key feature information can be integrated and processed, and the integrated information can be used as a corresponding interpretable decision package.

[0074] The integrated information is used as the interpretable decision package.

[0075] This application generates a visualized decision path corresponding to a target decision tree; then obtains the decision results of the target decision tree; and acquires the key feature information corresponding to the target decision tree. Subsequently, the visualized decision path, decision results, and key feature information are integrated to obtain corresponding integrated information. This integrated information is then used as an interpretable decision package. Based on the above processing flow, this application, by integrating the visualized decision path, decision results, and key feature information of the target decision tree, can efficiently and accurately construct the corresponding interpretable decision package, improving the construction efficiency and intelligence of the interpretable decision package. This is beneficial for understanding the robot's decision logic and can thus quickly assess the rationality and safety of the decisions.

[0076] In some optional implementations, the system also has the capability to handle deep learning decision interpretation for complex tasks. The specific implementation process includes: Using deep learning algorithms for decision-making: For some complex tasks, such as image recognition and natural language processing, decision tree models may not be able to handle them accurately. In this case, deep learning algorithms (such as convolutional neural networks (CNN) for image recognition and recurrent neural networks (RNN) for natural language processing) are used for decision-making.

[0077] Using a decision tree model as an interpreter: After the deep learning algorithm makes a decision, a decision tree model is used as an interpreter. The LIME (Locally Interpretable Model-Agnostic) algorithm is used to extract the decision tree rules corresponding to the deep learning decision result. The basic idea of ​​the LIME algorithm is to generate a set of local samples near the sample point to be interpreted, and then use a simple interpretable model (such as a decision tree) to fit these local samples, thereby obtaining an explanation for the decision at that sample point.

[0078] The output is an interpretable decision package: This package integrates the decision results from the deep learning algorithm, the visualized decision path extracted using the LIME algorithm, and key feature weights, forming an interpretable decision package consisting of "decision result + visualized decision path + key feature weights." For example, in an image recognition task, the deep learning algorithm identifies an object in an image as a "cat." The decision tree rules extracted using the LIME algorithm might show that features of certain specific pixel regions in the image (such as ears, eyes, etc.) led to this recognition result. The package also provides the weights of these key features, explaining their influence on the decision result.

[0079] In complex tasks, although deep learning algorithms are used for decision-making, the decision tree model, acting as an interpreter, and the LIME algorithm transform the "black box" output of deep learning into interpretable decision rules, providing a core basis for behavior explanation and risk analysis. The result of this stage, the interpretable decision package, acts like a decision instruction manual, helping to understand the robot's decision logic and assess the rationality and safety of its decisions.

[0080] In some alternative implementations, step S204 includes the following steps: Information is extracted from the interpretable decision package to obtain the corresponding key feature information.

[0081] In this embodiment, the interpretable decision package consists of a visualized decision path, decision results, and key feature information related to the target decision tree. The required key feature information (i.e., key features) can be extracted by performing information extraction on this interpretable decision package.

[0082] Obtain a pre-defined risk feature construction strategy.

[0083] In this embodiment, the risk feature construction strategy includes the following: 1. Defining risk features. 1) Definition of risk features at the perception level: Analyzing the relationship between key features and perception data: Determining which key features are related to the robot's perception data. For example, in obstacle avoidance by industrial collaborative robots, obstacle distance and speed key features are related to data from perception devices such as vision sensors and lidar. Setting perception anomaly features: Based on the accuracy of the perception devices and the robot's safety requirements, setting perception anomaly features for relevant key features. For example, for the obstacle distance key feature, if the distance measured by the vision sensor deviates from the actual distance by more than 10%, it is defined as a perception anomaly feature; for the speed key feature, if the speed fluctuation measured by the lidar exceeds 20% of the normal speed range, it is considered abnormal. 2) Definition of risk features at the decision level: Reviewing decision rules: Understanding the decision rules based on key features in the decision tree model. For example, in obstacle avoidance decision-making, when the obstacle distance is less than a certain value and the speed is greater than a certain value, the decision tree decides that the robot should take emergency braking measures. Determining decision anomaly features: When the decision path deviates from these conventional safety rules, it is defined as a decision anomaly feature. In actual operation, even if the distance and speed of the obstacle meet the conditions for emergency braking, if the decision tree decides that the robot should continue moving forward, this situation is considered an abnormal decision characteristic. 3) Definition of execution-level risk characteristics: Corresponding key characteristics to execution actions: Clarifying how key characteristics affect the robot's execution actions. For example, when an industrial collaborative robot grasps an object, the size and weight of the object, as key characteristics, will affect the grasping force and trajectory of the robotic arm. Setting execution abnormal characteristics: Based on the robot's execution accuracy and safety range, set execution abnormal characteristics for relevant key characteristics. For example, if the actual grasping force deviates from the preset force by more than 15% or the movement angle deviates by more than 5 degrees when the robotic arm grasps an object, it is defined as an execution abnormal characteristic.

[0084] 2. Establish a risk feature database. Design the database structure: A relational or non-relational database can be used to store the risk feature database. When designing the database table structure, fields such as feature name, feature description, level (perception, decision-making, execution), associated key features, threshold range, and anomaly handling measures can be included. Enter risk feature information: Enter the defined risk features into the risk feature database according to the designed database structure.

[0085] For example, for an abnormal obstacle distance perception feature in obstacle avoidance of industrial collaborative robots, the information entered may be as follows: Feature name: Abnormal obstacle distance perception; Feature description: The obstacle distance measured by the vision sensor deviates from the actual distance by more than 10%; Level: Perception; Associated key feature: Obstacle distance; Threshold range: Deviation > 10%; Anomaly handling measures: Recalibrate the sensor and check whether the sensor is damaged.

[0086] The key feature information is processed based on the risk feature construction strategy to construct the corresponding risk features.

[0087] In this embodiment, the risk feature construction process can be performed on the key feature information based on the strategy content of the risk feature construction strategy to obtain the corresponding risk features.

[0088] The risk characteristics are entered into a preset database to obtain the corresponding target database.

[0089] In this embodiment, the selection of the aforementioned database is not specifically limited; either relational or non-relational databases can be used. The selected database can be used to store the aforementioned risk characteristics, and the resulting target database can be used as the corresponding risk characteristic library.

[0090] The target database is used as the risk feature database.

[0091] In this embodiment, as the robot's application scenarios change, new data accumulates, and the decision tree model is optimized, the risk feature library can be updated and maintained periodically. For example, when it is found that the threshold settings for certain risk features are unreasonable, they can be adjusted in a timely manner; when new risk situations arise, they can be defined as new risk features and entered into the library in a timely manner.

[0092] This application extracts information from an interpretable decision package to obtain corresponding key feature information; then it acquires a preset risk feature construction strategy; and processes the key feature information based on the risk feature construction strategy to construct corresponding risk features; subsequently, the risk features are entered into a preset database to obtain a corresponding target database; and finally, the target database is used as a risk feature library. Based on the above processing flow, this application automatically and intelligently completes the construction of the risk feature library by extracting information from an interpretable decision package to obtain key feature information, then processes the key feature information based on the use of a risk feature construction strategy to construct risk features, and then enters the risk features into a preset database to obtain a risk feature library. This improves the efficiency of risk feature library construction.

[0093] In some optional implementations of this embodiment, step S206 includes the following steps: The behavioral data is compared one by one with the features in the risk feature database to obtain the corresponding comparison results.

[0094] In this embodiment, the robot's real-time collected behavior data is compared one by one with features in the risk feature database. Specifically, data matching algorithms (such as string matching, numerical range matching, etc.) can be used to determine whether the current behavior data matches any risk features defined in the risk feature database, and generate corresponding comparison results. For example, the real-time collected sensor data is compared with the sensor data fluctuation threshold defined in the risk feature database. If the fluctuation exceeds the threshold, it is determined to be a perception anomaly. Alternatively, pattern matching, similarity calculation, and other methods can be used to determine whether the current data matches a certain risk event in the risk feature database, and generate corresponding comparison results.

[0095] Risk analysis is performed based on the comparison results and the preset level classification rules to obtain the corresponding risk analysis results.

[0096] In this embodiment, probability assessment and impact assessment are performed based on the comparison results and grading rules to obtain risk analysis results that include the probability of risk occurrence and the degree of impact.

[0097] The aforementioned classification rules include classification rules for the likelihood (probability) of risk occurrence and classification rules for the degree of risk impact.

[0098] Specifically, the rules for classifying the probability of risk occurrence include: 1. Collecting information on robot application scenarios. Organizing cross-departmental teams, including robot R&D personnel, maintenance personnel, safety experts, and business personnel familiar with robot application scenarios, to conduct a comprehensive survey of robot application scenarios. For example, in industrial production scenarios, understanding the robot's working area, workflow, and frequency of interaction with other equipment and personnel; in medical assistance scenarios, understanding the robot's contact methods with patients and medical staff, and working hours. Through on-site inspections, questionnaires, interviews, etc., recording in detail the characteristics of robot application scenarios and potential risk factors. For example, in industrial production workshops, recording the robot's operating speed, operational accuracy, and the complexity of the working environment (such as the presence of obstacles, whether the ground is flat, etc.). 2. Analyzing historical data. Extracting information related to risk occurrence from historical data such as robot maintenance records, fault reports, and accident records. For example, statistically analyzing the number of robot malfunctions, fault types, and time periods of malfunctions over a past period. Classifying and organizing historical data, and conducting statistical analysis according to different fault causes and scenarios. For example, malfunctions can be categorized into mechanical, electrical, and software malfunctions, and the frequency of occurrence for each type can be statistically analyzed. 3. Establish probability level classification standards. Based on collected application scenario information and historical data analysis results, combined with industry experience and expert opinions, establish risk probability level classification standards. For example: High probability: Under similar application scenarios and operating conditions, the frequency of this risk event has exceeded 5 times in the past year, or based on the current situation, the probability of its occurrence in the next month is predicted to exceed 30%. For example, when robots frequently perform high-precision assembly operations and the equipment is aging, the probability of mechanical failure leading to assembly errors is relatively high. Medium probability: The frequency of this risk event has been between 1 and 5 times in the past year, or the probability of its occurrence in the next month is predicted to be between 10% and 30%. For example, the occasional entry of new employees into the robot's working area increases the risk of collisions, but the probability can be reduced through training and management. Low probability: The frequency of this risk event has been less than 1 time in the past year, or the probability of its occurrence in the next month is predicted to be less than 10%. For example, when the robot's working environment is relatively enclosed and has undergone strict safety protection measures, the probability of external interference causing malfunctions is low.

[0099] The rules for classifying the degree of risk impact include: 1. Defining the scope of impact. Identifying the objects and scope that may be affected by robot risk events, including personnel safety, equipment damage, production interruption, and data leakage. For example, in industrial production scenarios, robot malfunctions may cause production line shutdowns, affecting production progress and product quality; in medical assistance scenarios, robot operational errors may cause harm to patients. A detailed analysis of each affected object is conducted to understand its importance and sensitivity. For example, in medical scenarios, patient safety is the primary consideration, and its impact is far greater than the losses from production line shutdowns.

[0100] 2. Assessment Indicators for Impact Level. Develop corresponding assessment indicators for each affected entity. For example, for personnel safety, indicators such as the degree of injury (e.g., minor injury, serious injury, death) and the number of people affected can be set; for equipment damage, indicators such as equipment repair costs and equipment scrap rate can be set; for production interruption, indicators such as downtime and output loss can be set. Based on the assessment indicators, combined with actual conditions and industry experience, determine the quantitative standards for different impact levels. For example: Severe: Leads to death or serious injury, equipment scrapping, production interruption exceeding 72 hours, causing significant economic losses and social impact. For example, an industrial robot malfunction causing serious injury to a worker, resulting in a production line shutdown for more than a week. Significant: Leads to minor injury to personnel, requiring major equipment repairs, production interruption between 24 and 72 hours, causing significant economic losses. For example, a robot malfunction leading to substandard product quality, requiring large-scale rework. Moderate: No or minor injury to personnel, requiring minor equipment repairs, production interruption between 1 and 24 hours, causing some economic losses. For example, a minor robot malfunction causing a temporary halt to some production processes, but which can be quickly resumed. Minor impact: No substantial impact on personnel and equipment, production interruption time is less than 1 hour, and economic loss is minimal. For example, the robot software experiences a brief malfunction, but can be restored to normal operation by restarting.

[0101] The process of determining the likelihood and impact of a risk includes: 1) Likelihood assessment. Based on the comparison results and the aforementioned likelihood grading rules, further assess the likelihood of the risk occurring (i.e., the likelihood level). For example, if the current data matches the risk feature database closely, and similar risk events occur frequently in historical data, then the likelihood of the risk occurring is high. Consider other factors that may affect the likelihood of the risk occurring, such as environmental changes and equipment maintenance. For example, if the robot's work area has recently been modified, adding new obstacles, the likelihood of a collision risk may further increase. 2) Impact assessment. Based on the type of risk event and the objects that may be affected, and in conjunction with the aforementioned impact grading rules, assess the impact after the risk occurs (i.e., the degree of impact). For example, if a robot experiences a collision risk and there are people around, it may cause injury to people, and the impact may be severe. Consider the chain reactions and secondary impacts that the risk event may trigger. For example, robot malfunction may cause the production line to stop, thereby affecting the operation of the entire supply chain and causing a wider impact.

[0102] A corresponding risk matrix is ​​constructed based on the risk analysis results.

[0103] In this embodiment, a two-dimensional risk matrix is ​​constructed based on the determined probability level and impact level. The rows of the risk matrix represent the probability of a risk occurring (high, medium, low), and the columns represent the impact level after the risk occurs (severe, significant, moderate, minor). At each intersection point of the risk matrix, the corresponding risk level is labeled. For example, the intersection of high probability and severe impact is labeled as a high-risk level, and the intersection of low probability and minor impact is labeled as a low-risk level.

[0104] The behavioral data is processed for risk assessment based on the risk matrix to obtain the corresponding assessment results.

[0105] In this embodiment, the risk assessment process described above refers to risk level positioning based on a risk matrix. This is achieved by mapping the assessed probability of risk occurrence and severity of impact to corresponding positions in the risk matrix, thus finding the corresponding intersection points. Furthermore, the risk level of the current risk event is determined based on the annotations at these intersection points. For example, if the probability of a risk occurring is high and the severity of its impact is severe, then the corresponding intersection point in the risk matrix represents a high-risk level.

[0106] The assessment results shall be used as the risk assessment results.

[0107] In this embodiment, relevant experts and personnel can also be organized to verify and discuss the determined risk level. Considering potential assessment biases and uncertainties, necessary adjustments can be made to the risk level.

[0108] This application compares behavioral data with features in a risk feature database one by one to obtain the corresponding comparison results; then, based on the comparison results and preset level classification rules, it performs risk analysis to obtain the corresponding risk analysis results; next, it constructs a corresponding risk matrix based on the risk analysis results; then, it performs risk assessment processing on the behavioral data based on the risk matrix to obtain the corresponding assessment results; finally, it uses the assessment results as the risk assessment results. Based on the above processing flow, this application, by establishing a risk feature database, can identify various potential safety hazards that may occur during robot operation in real time. Furthermore, by using the risk matrix method, it can accurately assess the risk level of behavioral data, ensuring the accuracy of the obtained risk assessment results. Moreover, it can formulate targeted control measures based on the severity and probability of the risk, achieving proactive avoidance of safety hazards.

[0109] In some optional implementations of this embodiment, after step S206, the electronic device may further perform the following steps: The end-to-end behavioral data, the interpretable decision package, and the risk assessment results are combined to obtain the corresponding combined data.

[0110] In this embodiment, when new data needs to be encapsulated into the blockchain, the end-to-end behavioral data, the interpretable decision package, and the risk assessment results are first combined to form the content of the current data block, i.e., combined data.

[0111] The combined data is signed to obtain the corresponding target combined data.

[0112] In this embodiment, the signature processing includes: digesting the end-to-end data, interpretable decision packets, and risk assessment results in the current data block. The digest processing also uses the SHA-256 encryption algorithm to generate a fixed-length hash value as a unique identifier for the current data. Then, a precise timestamp is added to the data block to record the time the data block was generated. The timestamp precision can reach the millisecond level to ensure the timeliness and traceability of the data. Afterward, the node that generated the data block will sign the data block using its private key to prove the source and integrity of the data block. Other nodes can verify the signature using their public key to ensure that the data block has not been tampered with.

[0113] Invoke the preset blockchain network.

[0114] In this embodiment, the blockchain network architecture design includes: 1. Robot Local Node. Function: Responsible for storing the robot's real-time behavioral data. This data includes various operation records during robot operation, real-time data collected by sensors, etc. Deployment: Local node software is deployed on each robot device. This software has the functions of data acquisition, storage, and communication with edge nodes. It acquires the robot's behavioral data in real time through a preset data acquisition interface and stores it in the local node's database. Data Synchronization: Periodically synchronizes the locally stored real-time behavioral data to the edge nodes. The synchronization frequency can be adjusted according to factors such as the importance of the data and network bandwidth; for example, for critical safety data, it can be set to synchronize once per minute.

[0115] 2. Edge Nodes. Functional Role: Primarily stores risk warning records. When a robot's local node detects a potential safety risk, it sends the risk warning information to the edge node for storage. Deployment Location: Edge node servers are deployed in areas close to the robot cluster. There can be one or more edge nodes; the specific number depends on the size and distribution of the robot cluster. Data Processing: The edge node performs preliminary processing and analysis of the received risk warning information, such as classifying and grading risks. Simultaneously, the edge node is also responsible for synchronizing the processed risk warning records to the central node.

[0116] 3. Central Node. Functional Positioning: Stores historical traceability data, including end-to-end data, decision paths, and control results. The central node is the core of the blockchain network, responsible for maintaining the integrity and consistency of the entire blockchain. Deployment Environment: A highly reliable and secure server environment is selected for deploying the central node. A distributed deployment approach can be adopted to improve system fault tolerance and availability. Data Management: Stores and manages data synchronized from edge nodes, while providing data query and verification interfaces for use by the traceability query module.

[0117] The target combination data is stored and processed based on the blockchain network.

[0118] In this embodiment, the storage and processing of the target combination data can be completed by encapsulating the generated target combination data into the aforementioned blockchain network.

[0119] In addition, the system has a traceability and query module that supports multi-dimensional retrieval: administrators can query corresponding behavioral data by task ID, trace the complete processing chain by risk event ID, and query historical security records by device identifier. Simultaneously, a permission management mechanism is set up, with different roles (users, maintenance personnel, regulatory agencies) having different data access permissions.

[0120] This application combines end-to-end behavioral data, interpretable decision packages, and risk assessment results to obtain corresponding combined data; then, it signs the combined data to obtain the corresponding target combined data; subsequently, it invokes a pre-defined blockchain network; and finally, it stores the target combined data on the blockchain network. Based on the above processing flow, this application, by combining and signing end-to-end behavioral data, interpretable decision packages, and risk assessment results, and storing the resulting target combined data on a blockchain network, utilizes its distributed ledger and smart contract technologies to ensure the authenticity and integrity of data during storage and transmission. This provides a solid data foundation for subsequent full-process traceability and auditing of robot behavior, enhances data credibility and security, and improves the safety and reliability of robot operation.

[0121] In some optional implementations of this embodiment, the system also has a feedback-driven iterative optimization function to improve security capabilities. Specific data processing operations include: 1) Feedback analysis: The feedback analysis module collects two types of core data: security incident tracing reports (including risk root causes, decision deviations, and control effects) and daily operation data (such as the path distribution for normal decisions and the accuracy rate of risk warnings). For example, by analyzing the security incident tracing reports of the logistics sorting robot, it can determine whether the missorting of goods is due to order identification errors or path planning deviations, and simultaneously calculate the accuracy rate of risk warnings during daily operation.

[0122] 2) Bottlenecks and Optimization of the Positioning System: By comparing and analyzing the bottlenecks of the positioning system: if the decision tree model has decision bias, the feature weights and branching rules of the decision tree will be optimized by adding new sample data; if the risk feature library is insufficient, feature indicators will be supplemented based on newly emerging risk events; if the blockchain storage efficiency is low, the data block encapsulation strategy and node synchronization mechanism will be optimized. The gradient boosting tree algorithm is used to iteratively train the decision tree model and the risk assessment model, and the optimized parameters and rules are synchronized to the preceding processes to achieve a closed-loop evolution of "collection-modeling-early warning-tracing-optimization". For example, if the decision tree model is found to have decision bias in the voice command recognition task of children's educational robots, the feature weights and branching rules of the decision tree will be optimized by adding new children's voice command sample data to improve the recognition accuracy.

[0123] In some alternative implementations, the user information obtained is subject to user consent and complies with relevant laws and policies.

[0124] Furthermore, any software tools or components not belonging to our company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

[0125] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0126] It should be emphasized that, to further ensure the privacy and security of the above risk assessment results, the risk assessment results can also be stored in a blockchain node.

[0127] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0128] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0129] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0130] Further reference Figure 3 As a response to the above Figure 2The implementation of the method shown in this application provides an embodiment of a robot control device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0131] like Figure 3 As shown, the robot control device 300 described in this embodiment includes: a data acquisition module 301, a screening module 302, a processing module 303, a construction module 304, an acquisition module 305, an evaluation module 306, and a control module 307. Wherein: The data acquisition module 301 is used to collect the robot's full-link behavior data within a historical time period; The filtering module 302 is used to filter out target feature data related to the robot task from the full-link behavior data; The processing module 303 is used to train a preset decision tree algorithm based on the target feature data to obtain a trained target decision tree, and to generate a corresponding interpretable decision package based on the target decision tree. Construction module 304 is used to construct a corresponding risk feature library based on the interpretable decision package; The acquisition module 305 is used to acquire the robot's behavior data in real time; Assessment module 306 is used to perform risk assessment processing on the behavioral data based on the risk feature library and generate corresponding risk assessment results; The control module 307 is used to perform corresponding control processing on the robot based on the risk assessment results.

[0132] In some optional implementations of this embodiment, the acquisition module 301 includes: Call submodules to invoke preset multi-source data acquisition modules; The data acquisition submodule is used to collect the robot's perception layer data, decision layer data, and execution layer data within a historical time period based on the multi-source data acquisition module. The first integration submodule is used to integrate and process the perception layer data, the decision layer data, and the execution layer data to obtain the corresponding first processed data; The optimization submodule is used to perform data optimization processing on the first processed data based on a preset standardization strategy to obtain the corresponding second processed data; The first determining submodule is used to use the second processed data as the end-to-end behavior data.

[0133] In some optional implementations of this embodiment, the optimized submodule includes: The calling unit is used to invoke preset data association rules; The association unit is used to perform association processing on the first processed data based on the data association rules to obtain the corresponding third processed data; The encapsulation unit is used to encapsulate the third processed data based on a preset data format to obtain the corresponding fourth processed data; A marking unit is used to mark the fourth processed data based on preset identification information to obtain the corresponding fifth processed data; A determining unit is used to use the fifth processed data as the second processed data.

[0134] In some optional implementations of this embodiment, the processing module 303 includes: A generation submodule is used to generate a visual decision path corresponding to the target decision tree; The first acquisition submodule is used to acquire the decision results of the target decision tree; The second acquisition submodule is used to acquire key feature information corresponding to the target decision tree; The second integration submodule is used to integrate the visualized decision path, the decision result, and the key feature information to obtain the corresponding integrated information. The second determining submodule is used to use the integrated information as the interpretable decision package.

[0135] In some optional implementations of this embodiment, the construction module 304 includes: The extraction submodule is used to extract information from the interpretable decision package to obtain the corresponding key feature information; The third acquisition submodule is used to acquire the preset risk feature construction strategy; The processing submodule is used to process the key feature information based on the risk feature construction strategy to construct the corresponding risk features; The input submodule is used to input the risk characteristics into a preset database to obtain the corresponding target database; The third determining submodule is used to use the target database as the risk feature library.

[0136] In some optional implementations of this embodiment, the evaluation module 306 includes: The comparison submodule is used to compare the behavioral data with the features in the risk feature database one by one to obtain the corresponding comparison results; The analysis submodule is used to perform risk analysis based on the comparison results and preset level classification rules to obtain the corresponding risk analysis results; A submodule is constructed to build a corresponding risk matrix based on the risk analysis results. The assessment submodule is used to perform risk assessment processing on the behavioral data based on the risk matrix to obtain the corresponding assessment results; The fourth determination submodule is used to use the assessment result as the risk assessment result.

[0137] In some optional implementations of this embodiment, the robot control device further includes: The combination module is used to combine the end-to-end behavioral data, the interpretable decision package, and the risk assessment results to obtain corresponding combined data. The signature module is used to sign the combined data to obtain the corresponding target combined data. The calling module is used to invoke a pre-defined blockchain network; A storage module is used to store and process the target combination data based on the blockchain network.

[0138] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0139] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with components 41-43 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0140] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0141] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may also include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions for robot control methods. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0142] In some embodiments, the processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions stored in the memory 41 or to process data, such as executing computer-readable instructions for the robot control method.

[0143] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0144] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the robot control method described above.

[0145] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0146] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A robot control method, characterized in that, Includes the following steps: Collect end-to-end behavior data of the robot over a historical time period; Target feature data related to the robot task are filtered out from the full-link behavior data; The preset decision tree algorithm is trained based on the target feature data to obtain a trained target decision tree, and a corresponding interpretable decision package is generated based on the target decision tree. A corresponding risk feature library is constructed based on the interpretable decision package; Real-time acquisition of the robot's behavior data; Based on the risk feature library, the behavioral data is processed for risk assessment to generate corresponding risk assessment results; Based on the risk assessment results, the robot is subjected to corresponding control processing.

2. The robot control method according to claim 1, characterized in that, The steps for collecting the robot's full-link behavior data over a historical time period specifically include: Invoke the preset multi-source data acquisition module; The multi-source data acquisition module collects perception layer data, decision layer data, and execution layer data of the robot within a historical time period. The perception layer data, the decision layer data, and the execution layer data are integrated and processed to obtain the corresponding first processed data. The first processed data is optimized based on a preset standardization strategy to obtain the corresponding second processed data; The second processed data is used as the end-to-end behavior data.

3. The robot control method according to claim 2, characterized in that, The step of optimizing the first processed data based on a preset standardization strategy to obtain the corresponding second processed data specifically includes: Invoke the preset data association rules; Based on the data association rules, the first processed data is associated to obtain the corresponding third processed data; The third processed data is encapsulated and processed based on a preset data format to obtain the corresponding fourth processed data. The fourth processed data is marked based on preset identification information to obtain the corresponding fifth processed data; The fifth processed data is used as the second processed data.

4. The robot control method according to claim 1, characterized in that, The step of generating a corresponding interpretable decision package based on the target decision tree specifically includes: Generate a visual decision path corresponding to the target decision tree; Obtain the decision results of the target decision tree; Obtain key feature information corresponding to the target decision tree; The visualized decision path, the decision result, and the key feature information are integrated to obtain corresponding integrated information. The integrated information is used as the interpretable decision package.

5. The robot control method according to claim 1, characterized in that, The step of constructing the corresponding risk feature library based on the interpretable decision package specifically includes: Information is extracted from the interpretable decision package to obtain the corresponding key feature information; Obtain the pre-defined risk feature construction strategy; The key feature information is processed based on the risk feature construction strategy to construct the corresponding risk features; The risk characteristics are entered into a preset database to obtain the corresponding target database; The target database is used as the risk feature database.

6. The robot control method according to claim 1, characterized in that, The step of performing risk assessment processing on the behavioral data based on the risk feature database to generate corresponding risk assessment results specifically includes: The behavioral data is compared one by one with the features in the risk feature database to obtain the corresponding comparison results; Risk analysis is performed based on the comparison results and the preset level classification rules to obtain the corresponding risk analysis results; Construct a corresponding risk matrix based on the risk analysis results; Based on the risk matrix, the behavioral data is subjected to risk assessment processing to obtain the corresponding assessment results; The assessment results shall be used as the risk assessment results.

7. The robot control method according to claim 1, characterized in that, After the step of performing risk assessment processing on the behavioral data based on the risk feature library and generating corresponding risk assessment results, the method further includes: The end-to-end behavioral data, the interpretable decision package, and the risk assessment results are combined to obtain corresponding combined data. The combined data is signed to obtain the corresponding target combined data; Invoke the preset blockchain network; The target combination data is stored and processed based on the blockchain network.

8. A robot control device, characterized in that, include: The data acquisition module is used to collect the robot's full-link behavior data over a historical time period; The filtering module is used to filter out target feature data related to the robot task from the full-link behavior data; The processing module is used to train a preset decision tree algorithm based on the target feature data to obtain a trained target decision tree, and to generate a corresponding interpretable decision package based on the target decision tree. The construction module is used to build a corresponding risk feature library based on the interpretable decision package; The acquisition module is used to acquire the robot's behavior data in real time; The assessment module is used to perform risk assessment processing on the behavioral data based on the risk feature library and generate corresponding risk assessment results; The control module is used to perform corresponding control processing on the robot based on the risk assessment results.

9. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the robot control method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the robot control method as described in any one of claims 1 to 7.

Images