A human-shaped robot interaction control system and method based on a communication module

Abstract

This invention relates to the field of humanoid robot intelligent control technology, and more particularly to a humanoid robot interactive control system and method based on a communication module. It obtains a real-time network score representing the current network quality through a network evaluation module, then predicts a predicted network score representing the future network quality through a network prediction module, and classifies each interactive control command by combining the two scores through an instruction interaction module. Finally, an instruction processing module performs differentiated processing on the different categories of interactive control commands. This invention combines a command classification mechanism based on real-time and predicted dual-dimensional network scores, ensuring that the same command obtains a classification result matching its actual needs in different network environments. Furthermore, it incorporates a differentiated processing strategy based on the classification results using a local lightweight model, thus solving the problem that existing cloud-based humanoid robot control methods cannot guarantee real-time performance under fluctuating network quality conditions.

Description

Technical Field

[0001] This invention relates to the field of humanoid robot intelligent control technology, and in particular to a humanoid robot interactive control system and method based on a communication module. Background Technology

[0002] Unlike fixed-base industrial robotic arms or wheeled service robots, humanoid robots rely on complex motion control through multi-joint collaboration and need to perceive environmental changes in real time and make dynamic adjustments. This "human-like" interactive characteristic makes its interactive control have stringent requirements for real-time performance. Even millisecond-level latency differences can directly affect the reliability of task completion and the naturalness of human-robot interaction.

[0003] However, given the limitations of current humanoid robot hardware, it is difficult to deploy complex semantic understanding, multimodal fusion, or high-precision motion planning models locally. One feasible solution is to upload user commands and environmental perception data to a cloud server, where the cloud model makes decisions and then sends back control commands, thereby achieving more intelligent interactive functions with limited local resources.

[0004] However, in practical applications, when the communication network quality is unstable, the transmission of cloud control commands will experience significant delays or even packet loss. This can lead to lag in robot movements, stuttering in interaction, and in severe cases, even safety hazards. Therefore, a humanoid robot interaction control method that can effectively address the impact of network quality fluctuations on real-time performance is needed. Summary of the Invention

[0005] Therefore, the present invention provides a humanoid robot interactive control system and method based on a communication module to solve the problem that the existing cloud control method for humanoid robots cannot guarantee real-time performance under network quality fluctuation conditions.

[0006] This invention provides a humanoid robot interactive control system based on a communication module, comprising:

[0007] The network evaluation module is used to obtain network status parameters based on the communication module, and to obtain a real-time network score that represents the current network quality based on the network status parameters.

[0008] The network prediction module is used to obtain a predicted network score that represents the future quality of the network based on network state parameters.

[0009] The command interaction module is used to acquire interactive control commands from the humanoid robot.

[0010] The instruction classification module is used to classify each interactive control instruction based on real-time network scores and predicted network scores;

[0011] The instruction processing module is used to differentiate the processing of different types of interactive control instructions based on preset cloud models and local lightweight models.

[0012] In a preferred embodiment: network status parameters are obtained based on the communication module, and a real-time network score representing the current network quality is obtained based on the network status parameters, including:

[0013] Network status parameters are obtained based on the communication module. These parameters include end-to-end latency, packet loss rate, available bandwidth, signal strength, and network congestion index.

[0014] Based on a pre-defined state assessment model, a real-time network score is obtained by calculating the network state parameters.

[0015] In a preferred embodiment: based on network state parameters, a predicted network score representing the future network quality is obtained, including:

[0016] Obtain historical network state parameters and construct an input tensor based on various historical network state parameters;

[0017] The input tensor is fed into a preset recurrent neural network prediction model to obtain the predicted value of each network state parameter output by the recurrent neural network prediction model.

[0018] Based on the preset state evaluation model, the predicted network score is calculated according to the predicted values ​​of the network state parameters.

[0019] In a preferred embodiment: each interactive control instruction is categorized based on real-time network scores and predicted network scores, including:

[0020] Obtain the preset urgency and complexity of each interactive control command;

[0021] Based on real-time network score, predicted network score, urgency, and complexity, a comprehensive risk value is obtained for each interactive control instruction. The comprehensive risk value is used to represent the degree to which the execution effect of an interactive control instruction is affected by network quality.

[0022] Dynamic risk thresholds are obtained based on real-time network scoring;

[0023] Each interactive control instruction is classified based on a comparison of its comprehensive risk value and dynamic risk threshold.

[0024] In a preferred embodiment: based on real-time network score, predicted network score, urgency, and complexity, a comprehensive risk value for each interactive control instruction is obtained, including:

[0025] Based on urgency and complexity, the inherent risks of obtaining instructions are identified.

[0026] The current network compatibility risk is determined based on real-time network scoring;

[0027] Based on the predicted network score, the future network adaptation risk is obtained;

[0028] Based on the inherent risks of the instructions, the current network adaptation risks, and the future network adaptation risks, a comprehensive risk value is obtained.

[0029] In a preferred embodiment: based on the predicted network score, the future network adaptation risk is obtained, including:

[0030] Based on the queue length of the interactive control instructions, the estimated time window from classification to execution of the interactive control instructions is obtained;

[0031] The time decay factor is obtained based on the expected time window;

[0032] The future network adaptation risk is obtained based on the time decay factor and the predicted network score.

[0033] In a preferred embodiment: interactive control commands include level 1 commands, level 2 commands, and level 3 commands; based on a preset cloud model and a local lightweight model, different types of interactive control commands are differentiated, including:

[0034] If the interactive control command is a level 1 command, a dedicated communication channel is allocated to the interactive control command, and the interactive control command is processed based on the dedicated communication channel and the cloud model;

[0035] If the interactive control command is a level 2 command, then the interactive control command is processed based on the local lightweight model;

[0036] If the interactive control command is a level 3 command, it will be processed based on the conventional communication channel and the cloud model.

[0037] The present invention also provides a humanoid robot interactive control method based on a communication module, comprising:

[0038] Network status parameters are obtained based on the communication module, and a real-time network score representing the current network quality is obtained based on the network status parameters.

[0039] Based on the network state parameters, a predicted network score is obtained to represent the future network quality.

[0040] Obtain interactive control commands from the humanoid robot;

[0041] Each interactive control command is classified based on real-time network scores and predicted network scores;

[0042] Based on preset cloud models and local lightweight models, different types of interactive control commands are processed differently.

[0043] The present invention also provides an electronic device, comprising:

[0044] Memory and processor;

[0045] The memory is used to store the program, and the processor is used to execute the steps in any of the above-described humanoid robot interaction control methods based on the communication module when the program is executed.

[0046] The present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps in any of the above-described humanoid robot interactive control methods based on a communication module.

[0047] The beneficial effects of adopting the above scheme are:

[0048] This invention provides a humanoid robot interactive control system based on a communication module. It uses a network evaluation module to obtain a real-time network score representing the current network quality based on network state parameters. Then, a network prediction module predicts a future network quality score. An instruction interaction module acquires interactive control instructions, and an instruction classification module categorizes each instruction based on both the real-time and predicted network scores. Finally, an instruction processing module differentiates the different categories of interactive control instructions based on a preset cloud model and a local lightweight model. This invention combines a real-time and predicted dual-dimensional network score instruction classification mechanism, enabling dynamic evaluation of the current network state and future network trends. It not only reflects immediate network quality but also proactively predicts network fluctuation risks, allowing instruction classification to be flexibly adjusted according to the real-time evolution of network quality. This ensures that the same instruction receives a classification result that matches the actual needs in different network environments. Furthermore, the differentiated processing strategy based on classification results, combined with the local lightweight model in this invention, not only ensures low latency and high accuracy of key instructions, but also effectively reduces the dependence of non-critical tasks on communication resources. By maintaining real-time performance through the local lightweight model, the overall interaction efficiency of the system is significantly improved under cloud-based control conditions, solving the problem that existing cloud-based control methods for humanoid robots cannot guarantee real-time performance under fluctuating network quality conditions. Attached Figure Description

[0049] Figure 1 The system architecture diagram of the humanoid robot interactive control system based on the communication module provided by the present invention;

[0050] Figure 2 for Figure 1 A diagram illustrating the specific steps executed by the instruction classification module in the middle section;

[0051] Figure 3The flowchart of the humanoid robot interaction control method based on the communication module provided by the present invention is shown. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Combination Figure 1 A specific embodiment of the present invention discloses a humanoid robot interaction control system based on a communication module, comprising:

[0054] The network evaluation module 110 is used to obtain network status parameters based on the communication module and obtain a real-time network score to represent the current network quality based on the network status parameters.

[0055] The network prediction module 120 is used to obtain a predicted network score representing the future network quality based on network state parameters.

[0056] The instruction interaction module 130 is used to acquire the interactive control instructions of the humanoid robot;

[0057] The instruction classification module 140 is used to classify each interactive control instruction based on the real-time network score and the predicted network score;

[0058] The instruction processing module 150 is used to differentiate the processing of different types of interactive control instructions based on the preset cloud model and local lightweight model.

[0059] It should be noted that the interactive control commands in this embodiment are not signals that directly drive the mechanisms in the humanoid robot to operate. Rather, they refer to any data or set of data that needs to be processed and can trigger the robot to perform relevant interactive actions. These include user commands (voice / gesture / text; taking voice as an example, interactive control commands can be the original voice data received by the robot or the specific voice content obtained through voice recognition. The specific type of data can be flexibly designed according to the actual robot's control software architecture), environmental perception data (ToF camera images, LiDAR point clouds), and robot status data (joint angles, IMU attitude, end effector force feedback), etc.

[0060] This invention combines a real-time and predictive dual-dimensional network scoring instruction classification mechanism, enabling dynamic assessment of the current network state and future network trends. It not only reflects immediate network quality but also proactively predicts network fluctuation risks, allowing instruction classification to flexibly adjust according to the real-time evolution of network quality. This ensures that the same instruction receives classification results that match its actual needs in different network environments. Furthermore, the invention employs a differentiated processing strategy based on classification results, coupled with a local lightweight model. This guarantees low latency and high accuracy for critical instructions while effectively reducing the dependence of non-critical tasks on communication resources. The local lightweight model maintains real-time performance, significantly improving the overall system interaction efficiency under cloud-based control conditions. This solves the problem in existing humanoid robot cloud control methods that cannot guarantee real-time performance under fluctuating network quality conditions.

[0061] Specifically, the network quality assessment method in this invention can be performed in any manner according to actual needs. This invention provides a preferred method; specifically, in another embodiment, the steps performed by the network assessment module 110—obtaining network status parameters based on the communication module and obtaining a real-time network score representing the current network quality based on the network status parameters—specifically include:

[0062] Network status parameters are obtained based on the communication module. These parameters include end-to-end latency, packet loss rate, available bandwidth, signal strength, and network congestion index.

[0063] Based on a pre-defined state assessment model, a real-time network score is obtained by calculating the network state parameters.

[0064] This invention uses a communication module to collect multiple key network parameters in real time, including end-to-end latency, packet loss rate, available bandwidth, signal strength, and network congestion index, forming a comprehensive perception of network quality. It covers all key dimensions of network performance and obtains a real-time network score that accurately reflects the actual performance of the current network.

[0065] Similarly, the above-mentioned state assessment model can also be flexibly designed by those skilled in the art based on experiments or experience and in combination with actual conditions. A specific example is as follows:

[0066]

[0067] in, For real-time online rating, End-to-end latency (in milliseconds, representing the time it takes for a command to travel between the cloud and the robot). Packet loss rate (in percentage, representing the proportion of transmitted data packets lost). Available bandwidth (in Mbps, representing the maximum data rate that the current link can support). This represents the theoretical maximum available bandwidth. Signal strength (unit: dBm, representing the strength of the wireless signal received by the communication module; priority signal can be a fixed value). and These represent the minimum and maximum possible signal strength values, respectively. This is a network congestion index (a value between 0 and 1, where 0 indicates no congestion and 1 indicates severe congestion, which can be estimated by the queue length reported by the router / base station). , , , , The corresponding weighting coefficients.

[0068] Similarly, the predicted network score representing future network quality can be implemented using any existing method as needed. For example, by fitting real-time network scores from historical analysis, a curve representing changes in network quality can be obtained, and thus the future network score can be derived.

[0069] The present invention provides a more preferred embodiment in which the steps performed by the network prediction module 120, namely obtaining a predicted network score representing the future network quality based on network state parameters, specifically include:

[0070] Obtain historical network state parameters and construct an input tensor based on various historical network state parameters;

[0071] The input tensor is fed into a preset recurrent neural network prediction model to obtain the predicted value of each network state parameter output by the recurrent neural network prediction model.

[0072] Based on the preset state evaluation model, the predicted network score is calculated according to the predicted values ​​of the network state parameters.

[0073] This embodiment employs a recurrent neural network prediction model (such as a lightweight GRU model or LSTM model) to accurately predict each network parameter. Compared to traditional overall network quality prediction methods, this parameter-based prediction mechanism can more meticulously grasp the future trends of each key network indicator. Furthermore, this embodiment uses a preset state evaluation model that is exactly the same as the real-time network score to calculate the predicted network score, ensuring that the predicted score and the real-time score are completely consistent in terms of evaluation dimensions, weight allocation, and calculation logic. This enables the system to make unified comparisons and decisions on the current and future network states based on the same standards, greatly improving the accuracy and reliability of instruction classification.

[0074] Furthermore, in combination Figure 2As shown, in a specific embodiment, the steps performed by the instruction classification module 140 above include: classifying each interactive control instruction based on the real-time network score and the predicted network score, specifically including:

[0075] S201. Obtain the preset urgency and complexity of each interactive control command;

[0076] S202. Based on the real-time network score, predicted network score, urgency, and complexity, obtain the comprehensive risk value of each interactive control instruction. The comprehensive risk value is used to represent the degree to which the execution effect of an interactive control instruction is affected by network quality.

[0077] S203. Obtain the dynamic risk threshold based on the real-time network score;

[0078] S204. Based on the comparison of the comprehensive risk value and dynamic risk threshold of each interactive control instruction, classify each interactive control instruction.

[0079] In the above, urgency and complexity are preset values ​​set by the user. In practice, they can be determined based on factors such as the data source, the time the data was generated, or the data flow in the code. These values ​​can be preset by instruction type or dynamically calibrated by user manual annotation based on historical execution data. For example:

[0080] Urgency can be represented by a value from 0 to 1, where 0 means not urgent and 1 means urgent. For example, the urgency of "obstacle avoidance" is 1 and the urgency of "voice greeting data" is 0.

[0081] Complexity can also be represented by a value from 0 to 1, where 0 represents low complexity and 1 represents high complexity. For example, the complexity of "millimeter-level positioning and grasping" is 1, the complexity of "normal path movement" is 0.5, and the complexity of "light display" is 0.1.

[0082] The overall risk value indicates the degree to which the execution effect of an interactive control command is affected by network quality. This can be understood from two perspectives: First, it represents the sensitivity of an interactive command to network quality. For example, simple commands that do not require the transmission of large amounts of data are less sensitive to network quality, while complex commands that require the transmission of large amounts of data are more sensitive. Network fluctuations will affect the latency and accuracy of the returned data. Second, the overall risk value also indicates the impact of the interactive control command on the robot's operation. For example, commands such as maintaining balance will severely affect the robot's operation, while functions such as voice chat, even if they fail, will not have a significant impact on the robot.

[0083] This invention constructs a comprehensive risk value that accurately reflects the degree to which the execution effect of instructions is affected by network quality by comprehensively considering the preset urgency and complexity of interactive control instructions, as well as a two-dimensional network quality assessment of real-time network scores and predicted network scores. This value not only reflects the key characteristics of the instructions themselves, but also dynamically reflects the potential risks to instruction execution from the current and future network states, achieving a quantitative assessment of the network dependence of instructions. Furthermore, this invention dynamically adjusts the risk threshold based on the real-time network score, allowing the classification criteria to adaptively adjust with real-time changes in the network environment. When network quality is good, the classification threshold is appropriately relaxed to make reasonable use of network resources; when network quality is poor, the classification criteria are raised to prioritize the reliability of critical instructions.

[0084] Based on the intelligent comparison of comprehensive risk value and dynamic risk threshold, the system can accurately classify each instruction, enabling it to intelligently identify instructions that truly require high protection in complex and ever-changing network environments, thereby achieving optimal resource allocation and maximizing interaction effects.

[0085] Similarly, the specific methods for determining the aforementioned comprehensive risk value and dynamic risk threshold can also be obtained using any existing method based on the actual situation. For example, both can be calculated using relevant linear functions. However, this invention provides a preferred implementation:

[0086] In a specific embodiment, step S202 above, obtaining a comprehensive risk value for each interactive control command based on real-time network score, predicted network score, urgency, and complexity, specifically includes:

[0087] Based on urgency and complexity, the inherent risks of obtaining instructions are identified.

[0088] The current network compatibility risk is determined based on real-time network scoring;

[0089] Based on the predicted network score, the future network adaptation risk is obtained;

[0090] Based on the inherent risks of the instructions, the current network adaptation risks, and the future network adaptation risks, a comprehensive risk value is obtained.

[0091] This embodiment calculates the inherent risk of instructions based on their pre-defined urgency and complexity, accurately quantifying the criticality of the task itself and ensuring the basic weight of high-risk instructions. Real-time network scoring assesses the immediate adaptability of the current network environment to the instructions, reflecting whether the current network state meets the basic requirements for instruction execution. Predictive network scoring calculates future network adaptability risks, proactively predicting the potential impact of network quality changes on instruction execution, enabling the system to prepare for impending network fluctuations. Finally, by integrating these three types of risks, a comprehensive risk value is formed that fully reflects the degree to which network quality affects the effectiveness of instruction execution.

[0092] Similarly, it is understood that the inherent risks of instructions, current network adaptation risks, and future network adaptation risks in the above process can also be flexibly designed according to actual conditions. Furthermore, this invention provides a preferred method for calculating future network adaptation risks, namely, in a new embodiment according to the claim, the above step: obtaining future network adaptation risks based on predicted network scores, specifically includes:

[0093] Based on the queue length of the interactive control instructions, the estimated time window from classification to execution of the interactive control instructions is obtained;

[0094] The time decay factor is obtained based on the expected time window;

[0095] The future network adaptation risk is obtained based on the time decay factor and the predicted network score.

[0096] This embodiment analyzes the length of the interactive control instruction queue, fully considering the actual waiting time of instructions in the system task queue. Based on this estimated time window, a time decay factor is dynamically generated, allowing the influence weight of the predicted network score to be intelligently adjusted as the instruction execution time approaches. For instructions about to be executed immediately, the predicted network risk maintains a high weight to ensure timely response to network fluctuations, while for instructions requiring a longer execution time, the decision weight of the predicted risk is reduced accordingly to avoid overreacting to uncertain future network changes. This dynamic time-aware risk assessment mechanism ensures that instructions nearing execution can avoid network risks in a timely manner while avoiding over-reliance on long-term predictions.

[0097] The calculation process for the above comprehensive risk value can be expressed by the following formula:

[0098]

[0099]

[0100] in, This represents the overall risk value. Indicates the degree of urgency. Indicate complexity, This indicates real-time online ratings. Indicates the predicted network score. , , and These are the corresponding weights. This represents the time decay factor. This represents the natural exponential function. Indicates the expected time window, This is a preset time constant used to control the decay rate, which can be adjusted experimentally.

[0101] In the above formula, both the real-time network score and the predicted network score are normalized to 0-1, where 0 represents the worst network condition and 1 represents the best network condition. The first two terms can be regarded as the inherent risk of the instruction, the third term can be regarded as the current network adaptation risk, and the fourth term can be regarded as the future network adaptation risk.

[0102] Furthermore, in a preferred embodiment, three levels can be designed based on real-time network scoring, with three dynamic risk thresholds designed for each level, and interactive control instructions can be sequentially divided into Level 1 instructions, Level 2 instructions, and Level 3 instructions (from high to low importance) based on the dynamic risk thresholds.

[0103] Specifically, a real-time network score greater than 0.8 can be considered a high-quality network, a real-time network score between 0.5 and 0.8 can be considered a normal network, and a real-time network score less than 0.5 can be considered a poor network.

[0104] In a high-quality network, the two dynamic risk thresholds are 0.3 and 0.1, respectively. Interactive control instructions with a comprehensive risk value greater than 0.3 are classified as Level 1 instructions, interactive control instructions with a comprehensive risk value between 0.1 and 0.3 are classified as Level 2 instructions, and interactive control instructions with a comprehensive risk value below 0.1 are classified as Level 3 instructions.

[0105] Similarly, the two dynamic risk thresholds are 0.5 and 0.3 for a typical network, and 0.7 and 0.4 for a poor network.

[0106] Based on this, in a specific embodiment, the steps performed by the instruction processing module 150 are as follows: Differentiated processing of different categories of interactive control instructions based on a preset cloud model and a local lightweight model, specifically including:

[0107] If the interactive control command is a level 1 command, a dedicated communication channel is allocated to the interactive control command, and the interactive control command is processed based on the dedicated communication channel and the cloud model;

[0108] If the interactive control command is a level 2 command, then the interactive control command is processed based on the local lightweight model;

[0109] If the interactive control command is a level 3 command, it will be processed based on the conventional communication channel and the cloud model.

[0110] In this embodiment, since the primary commands are usually of high importance, they are generally commands for emergency obstacle avoidance, fine operations (such as surgical assistance), and highly complex interactions (such as dynamic object tracking). Therefore, this embodiment uses a dedicated link (or dedicated channel) + cloud model processing to provide high-security control for them. For example, the communication module allocates 5G network slices or satellite communication backup links for primary commands, isolates interference from other services, and ensures bandwidth ≥20Mbps and packet loss rate ≤0.1%. Commands and context data (robot joint angles, environmental point clouds, user intentions) are uploaded to the cloud through a dedicated link. The cloud calls a complex model to perform calculations and returns high-precision control commands (such as joint velocity sequences and end-effector poses), which are then sent to the robot actuator with low latency (≤50ms) through a dedicated link.

[0111] Secondary commands are of medium importance and generally include routine movement, simple grasping, and voice interaction responses. They can be generated using a lightweight model running locally on the robot (such as MobileNetV3+LSTM with 50M parameters, which is periodically optimized through transfer learning based on user characteristics collected from edge nodes in the cloud). These commands are generated based on local sensor data (IMU attitude, LiDAR local map, user historical command preferences, etc.) and sent directly to the actuators without the need for cloud transmission.

[0112] Level 3 commands are of lower importance and are generally used for non-emergency information queries, entertainment controls, and background data synchronization. Level 3 commands can be transmitted via conventional links, such as through non-dedicated frequency bands (e.g., Wi-Fi 6 standard channels) to the cloud for processing, or they can be transmitted to edge nodes or local storage areas for batch caching (e.g., packaging every 10 commands) to reduce transmission frequency and network congestion.

[0113] This embodiment, under suitable network conditions, ensures that high-risk, urgent, and complex commands receive priority processing from the high-precision cloud model, medium-risk commands can be responded to quickly through a local lightweight model, and low-risk commands can reduce resource consumption through conventional links or edge caching. This hierarchical and differentiated processing architecture not only ensures the real-time performance and reliability of critical tasks in human-computer interaction, but also significantly improves the overall communication efficiency of the system, reduces energy consumption and computational load, and achieves an optimized balance between performance and cost through a reasonable resource allocation mechanism, perfectly adapting to the complex interaction needs of humanoid robots in diverse application scenarios.

[0114] Combination Figure 3 As shown, the present invention also provides a humanoid robot interactive control method based on a communication module, comprising:

[0115] S301. Obtain network status parameters based on the communication module, and obtain a real-time network score to represent the current network quality based on the network status parameters;

[0116] S302. Based on the network state parameters, obtain the predicted network score to represent the future network quality;

[0117] S303, Obtain the interactive control commands of the humanoid robot;

[0118] S304. Classify each interactive control instruction based on the real-time network score and the predicted network score;

[0119] S305, based on preset cloud models and local lightweight models, performs differentiated processing on different categories of interactive control commands.

[0120] It should be noted that the systems provided in the above embodiments are computer program products, while the methods in this embodiment can implement the technical solutions described in the above system embodiments. The specific implementation principles can be found in the corresponding content in the above embodiments, which will not be repeated here.

[0121] The present invention also provides an electronic device, comprising:

[0122] Memory and processor;

[0123] The memory is used to store the program, and the processor is used to execute the steps in any of the above-described humanoid robot interaction control methods based on the communication module when the program is executed.

[0124] The present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps in any of the above-described humanoid robot interactive control methods based on a communication module.

[0125] This invention provides a humanoid robot interactive control system based on a communication module. It uses a network evaluation module to obtain a real-time network score representing the current network quality based on network state parameters. Then, a network prediction module predicts a future network quality score. An instruction interaction module acquires interactive control instructions, and an instruction classification module categorizes each instruction based on both the real-time and predicted network scores. Finally, an instruction processing module differentiates the different categories of interactive control instructions based on a preset cloud model and a local lightweight model. This invention combines a real-time and predicted dual-dimensional network score instruction classification mechanism, enabling dynamic evaluation of the current network state and future network trends. It not only reflects immediate network quality but also proactively predicts network fluctuation risks, allowing instruction classification to be flexibly adjusted according to the real-time evolution of network quality. This ensures that the same instruction receives a classification result that matches the actual needs in different network environments. Furthermore, the differentiated processing strategy based on classification results, combined with the local lightweight model in this invention, not only ensures low latency and high accuracy of key instructions, but also effectively reduces the dependence of non-critical tasks on communication resources. By maintaining real-time performance through the local lightweight model, the overall interaction efficiency of the system is significantly improved under cloud-based control conditions, solving the problem that existing cloud-based control methods for humanoid robots cannot guarantee real-time performance under fluctuating network quality conditions.

[0126] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0127] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A humanoid robot interactive control system based on a communication module, characterized in that, include: The network evaluation module is used to obtain network status parameters based on the communication module, and to obtain a real-time network score that represents the current network quality based on the network status parameters. The network prediction module is used to obtain a predicted network score that represents the future quality of the network based on network state parameters. The command interaction module is used to acquire interactive control commands from the humanoid robot. The instruction classification module is used to classify each interactive control instruction based on real-time network scores and predicted network scores; The instruction processing module is used to differentiate the processing of different types of interactive control instructions based on the preset cloud model and local lightweight model. This includes obtaining network status parameters based on the communication module, and obtaining a real-time network score representing the current network quality based on these parameters, including: Network status parameters are obtained based on the communication module. These parameters include end-to-end latency, packet loss rate, available bandwidth, signal strength, and network congestion index. Based on a pre-defined state assessment model, a real-time network score is obtained by calculating the network state parameters. The preset state assessment model is: ； in, For real-time online rating, For end-to-end delay, For packet loss rate, For available bandwidth, This represents the theoretical maximum available bandwidth. For signal strength, and These represent the minimum and maximum possible signal strength values, respectively. This is a network congestion index. , , , , The corresponding weighting coefficients.

2. The humanoid robot interactive control system based on a communication module according to claim 1, characterized in that, Based on the network state parameters, a predicted network score is obtained to represent the future network quality, including: Obtain historical network state parameters and construct an input tensor based on various historical network state parameters; The input tensor is fed into a preset recurrent neural network prediction model to obtain the predicted value of each network state parameter output by the recurrent neural network prediction model. Based on the preset state evaluation model, the predicted network score is calculated according to the predicted values ​​of the network state parameters.

3. The humanoid robot interactive control system based on a communication module according to claim 1, characterized in that, Each interactive control instruction is categorized based on real-time network scores and predicted network scores, including: Obtain the preset urgency and complexity of each interactive control command; Based on real-time network score, predicted network score, urgency, and complexity, a comprehensive risk value is obtained for each interactive control instruction. The comprehensive risk value is used to represent the degree to which the execution effect of an interactive control instruction is affected by network quality. Dynamic risk thresholds are obtained based on real-time network scoring; Each interactive control instruction is classified based on a comparison of its comprehensive risk value and dynamic risk threshold.

4. The humanoid robot interactive control system based on a communication module according to claim 3, characterized in that, Based on real-time network scores, predicted network scores, urgency, and complexity, a comprehensive risk value is obtained for each interactive control instruction, including: Based on urgency and complexity, the inherent risks of obtaining instructions are identified. The current network compatibility risk is determined based on real-time network scoring; Based on the predicted network score, the future network adaptation risk is obtained; Based on the inherent risks of the instructions, the current network adaptation risks, and the future network adaptation risks, a comprehensive risk value is obtained.

5. The humanoid robot interactive control system based on a communication module according to claim 4, characterized in that, Based on the predicted network score, the future network adaptation risk is obtained, including: Based on the queue length of the interactive control instructions, the estimated time window from classification to execution of the interactive control instructions is obtained; The time decay factor is obtained based on the expected time window; The future network adaptation risk is obtained based on the time decay factor and the predicted network score.

6. The humanoid robot interactive control system based on a communication module according to claim 1, characterized in that, Interactive control commands include level 1 commands, level 2 commands, and level 3 commands; Based on preset cloud models and local lightweight models, different types of interactive control commands are processed differently, including: If the interactive control command is a level 1 command, a dedicated communication channel is allocated to the interactive control command, and the interactive control command is processed based on the dedicated communication channel and the cloud model; If the interactive control command is a level 2 command, then the interactive control command is processed based on the local lightweight model; If the interactive control command is a level 3 command, it will be processed based on the conventional communication channel and the cloud model.

7. A humanoid robot interactive control method based on a communication module, characterized in that, include: Network status parameters are obtained based on the communication module, and a real-time network score representing the current network quality is obtained based on the network status parameters. Based on the network state parameters, a predicted network score is obtained to represent the future network quality. Obtain interactive control commands from the humanoid robot; Each interactive control command is classified based on real-time network scores and predicted network scores; Based on the preset cloud model and local lightweight model, different types of interactive control commands are processed differently. This includes obtaining network status parameters based on the communication module, and obtaining a real-time network score representing the current network quality based on these parameters, including: Network status parameters are obtained based on the communication module. These parameters include end-to-end latency, packet loss rate, available bandwidth, signal strength, and network congestion index. Based on a pre-defined state assessment model, a real-time network score is obtained by calculating the network state parameters. The preset state assessment model is: ； in, For real-time online rating, For end-to-end delay, For packet loss rate, For available bandwidth, This represents the theoretical maximum available bandwidth. For signal strength, and These represent the minimum and maximum possible signal strength values, respectively. This is a network congestion index. , , , , The corresponding weighting coefficients.

8. An electronic device, characterized in that, include: Memory and processor; The memory is used to store the program, and the processor is used to execute the steps of the humanoid robot interactive control method based on the communication module as described in claim 7 when the program is executed.

9. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the humanoid robot interactive control method based on a communication module as described in claim 7.

Images