Intelligent equipment cluster ultrasonic sensing and digital twinborn path optimization system
By combining an ultrasonic detection module, a cluster collaborative sensing network, and a digital twin platform with an intelligent path optimization module, the problem of perception and path planning for intelligent devices in complex environments is solved, enabling efficient and safe cluster collaborative operation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHENGDU WTHINK TECH CO LTD
- Filing Date
- 2026-01-08
- Publication Date
- 2026-04-21
Smart Images
Figure CN121899831A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of ultrasonic sensing technology, specifically relating to an intelligent device cluster ultrasonic sensing and digital twin path optimization system. Background Technology
[0002] With the clustered application of intelligent devices (such as robots and drones) in industries, logistics, and security, achieving efficient collaborative operation and safe path planning in complex environments has become a key technological challenge. Existing technologies for intelligent devices suffer from the following shortcomings in environmental perception, cluster collaboration, and path optimization:
[0003] Limited environmental perception capabilities: Traditional devices often rely on a single sensor (such as lidar or camera) for obstacle detection, which is costly and susceptible to environmental interference (such as light or smoke). Although some devices use ultrasonic sensors, they are mostly used for simple obstacle avoidance and lack the ability to accurately detect the distance, direction, and speed of obstacles. Moreover, the perception range is limited (such as only covering a narrow area directly in front of the device), making it difficult to meet the needs of large-scale, multi-dimensional perception in complex environments.
[0004] Low efficiency of cluster collaborative perception: When multiple devices work together, traditional solutions rely on fixed communication infrastructure (such as base stations), and it is difficult to establish communication links autonomously in scenarios without infrastructure (such as in the wild or disaster areas). At the same time, the perception data of each device is isolated and lacks an efficient multi-source data fusion mechanism (such as simply splicing data), resulting in low accuracy and poor real-time performance of global obstacle distribution perception, making it difficult to support cluster collaborative decision-making.
[0005] Digital twin models lack dynamism: Existing digital twin technologies are mostly based on static geographic information (GIS) to build models, lacking the ability to update dynamic obstacles (such as moving people and vehicles) in real time. This leads to a disconnect between the virtual environment and the actual working environment, making it impossible to effectively support path simulation and evaluation, and making it difficult to detect potential problems in path planning in advance.
[0006] Path optimization and formation adjustment are not intelligent: Traditional path planning often uses fixed algorithms (such as A*, Dijkstra) and relies on preset environment models, which cannot dynamically adapt to real-time obstacle changes; formation adjustment is mostly based on simple rules (such as fixed spacing and arrangement), lacking comprehensive optimization of multiple objectives such as energy consumption, path length, and obstacle avoidance time, resulting in low operation efficiency and poor safety.
[0007] Therefore, in order to solve the problems of low environmental perception accuracy, poor cluster collaboration efficiency, insufficient model dynamism, and unintelligent path optimization in the existing technology, this invention proposes an intelligent device cluster ultrasonic sensing and digital twin path optimization system. Summary of the Invention
[0008] To overcome the shortcomings and deficiencies of the existing technology, the present invention adopts the following technical solution:
[0009] The intelligent device cluster ultrasonic sensing and digital twin path optimization system includes an ultrasonic detection module, a cluster collaborative sensing network, a digital twin platform, and an intelligent path optimization module.
[0010] Preferably, the ultrasonic detection module is configured in each smart device to obtain the distance, direction and relative speed data of obstacles around the device by emitting high-frequency pulse signals and receiving echoes, so as to provide basic data for obstacle avoidance and path planning of the device.
[0011] Preferably, the cluster collaborative sensing network is used to realize the sharing and fusion processing of ultrasonic data among various intelligent devices, and to generate a global 3D obstacle map to reflect the distribution of obstacles in the cluster operation environment;
[0012] Preferably, the digital twin platform is used to construct a virtual operating environment that includes static environment and dynamic obstacle information, supporting the simulation and evaluation of equipment operation paths;
[0013] Preferably, the intelligent path optimization module is used to dynamically optimize the operating path and cluster formation of intelligent devices based on the simulation evaluation information of the digital twin platform and the real-time data of the ultrasonic detection module, so as to ensure the efficiency and safety of cluster operations.
[0014] Preferably, the ultrasonic detection module uses a piezoelectric ceramic transducer as the ultrasonic transmitter; the distance to the obstacle is calculated by combining the echo time difference with the speed of sound, and the relative speed of the obstacle is calculated by combining the frequency shift of the detected echo with the transmission frequency.
[0015] Preferably, the cluster collaborative sensing network adopts a self-organizing network communication method, enabling each intelligent device to autonomously establish a communication link without fixed communication infrastructure, thereby realizing the real-time transmission and sharing of obstacle distance, direction, and speed data. Each intelligent device uploads its own detection data to the edge computing node, which uses the Kalman filter algorithm to fuse multi-source data and uses triangulation combined with multi-device perspective information to locate the obstacle coordinates, with the positioning error being less than a set value.
[0016] Preferably, the digital twin platform constructs a city digital twin model based on high-precision GIS data, initializes the static environmental information in the model, including buildings, roads, and bridges; integrates the 3D obstacle map generated by the cluster collaborative perception network into the digital twin model in real time to refresh the location information of moving obstacles in real time; the platform simulates the running path of intelligent devices in a virtual environment, evaluates the obstacle avoidance effect of the path, and discovers path planning problems in advance through simulation and makes optimization adjustments.
[0017] Preferably, the intelligent path optimization module constructs a network topology model of the device cluster operation path, abstracts the device operation environment into a graph structure, generates multiple feasible paths from the starting position to the target position and corresponding formation strategies as training samples, and calculates the indicators of each path and formation strategy, including obstacle avoidance success rate, energy consumption, path length, formation adjustment time and obstacle avoidance time, and uses these indicators as quality evaluation indicators of the corresponding feasible solutions.
[0018] Preferably, the intelligent path optimization module constructs a deep reinforcement learning agent, which includes a state space, an action space, and a reward function. The state space is a vector containing task information and a feasible solution state matrix. The task information includes the device's starting position, target position, and operating radius. The feasible solution state matrix is a k×5 matrix used to describe the obstacle avoidance success rate, energy consumption, path length, formation adjustment time, and obstacle avoidance time of k feasible solutions. The action space uses an ε-greedy policy to decide on path selection and formation parameter adjustment, where ε is a random exploration probability control variable with an initial value of 1 that gradually decays with training iterations. The reward function calculates a comprehensive reward based on the weights of obstacle avoidance success rate, energy consumption, path length, formation adjustment time, and obstacle avoidance time, with the sum of all weights being 1. The obstacle avoidance success rate is a positive weight, and the others are negative weights.
[0019] Preferably, the deep reinforcement learning agent selects the optimal solution by iteratively updating the action value function to maximize the expected reward value. Preferably, the action value function comprehensively considers the immediate reward value of selecting the action in the current state and the maximum reward value of selecting the optimal action in the new state, and introduces a discount factor to reflect the importance of future rewards. Preferably, the discount factor takes values in the range of [0,1]. The agent selects samples from the training samples by combining the SumTree priority sampling mechanism with uniform sampling and reweighted sampling mechanisms. The sampling efficiency is optimized by calculating the sample priority and temporal difference error. Preferably, a smoothing constant is introduced when calculating the priority to prevent the priority from being 0. The loss function dynamically adjusts the loss weights based on the sample temporal difference error.
[0020] Preferably, during the operation of the equipment cluster, the intelligent path optimization module monitors the quality evaluation indicators of the scheme in real time. When it detects that the obstacle avoidance success rate is lower than the preset value or the energy consumption is higher than the threshold, the module dynamically updates the optimal path and formation strategy based on the real-time obstacle data from the ultrasonic detection module and the environmental status data from the digital twin platform.
[0021] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:
[0022] 1. This invention enables a single device to accurately perceive obstacles, including distance, direction, and speed, through an ultrasonic detection module. Combined with multi-source data fusion of a cluster collaborative perception network, the obstacle positioning error is controlled within 0.5m, significantly reducing the risk of collision. At the same time, the real-time simulation function of the digital twin platform can predict potential path hazards in advance, and the intelligent path optimization module can dynamically adjust strategies to ensure that the equipment can efficiently avoid obstacles in complex dynamic environments, significantly improving operational safety.
[0023] 2. The intelligent path optimization module of this invention is based on deep reinforcement learning and comprehensively evaluates multiple indicators such as obstacle avoidance success rate, energy consumption, and path length. By dynamically optimizing the path and formation strategy, it reduces invalid paths and formation adjustment time. The self-organizing network communication and edge computing technology of the cluster collaborative perception network realizes real-time data sharing and fusion, avoids efficiency loss caused by the perception limitations of a single device, and ultimately reduces overall energy consumption, shortens the operation cycle, and improves resource utilization efficiency.
[0024] 3. The real-time mapping between the digital twin platform and the physical environment of this invention provides a virtual-real combined decision-making basis for path optimization; the deep reinforcement learning agent, through the mechanism of combining SumTree priority sampling and uniform sampling, efficiently utilizes training samples to optimize the decision model, and can update the path and formation strategy in real time according to environmental changes, breaking through the limitations of traditional fixed path planning, and enabling the system to have stronger autonomous decision-making ability and dynamic adaptability. Attached Figure Description
[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0026] Figure 1 The diagram shows the structure of the intelligent device cluster ultrasonic sensing and digital twin path optimization system of the present invention;
[0027] Figure 2 The flowchart of the intelligent device cluster ultrasonic sensing and digital twin path optimization system of the present invention is shown. Detailed Implementation
[0028] 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.
[0029] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more exemplary embodiments. Numerous specific details are provided in the following description to give a full understanding of exemplary embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, steps, etc., can be employed. In other instances, well-known structures, methods, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.
[0030] Example 1:
[0031] See Figure 1 and Figure 2 As shown, the intelligent device cluster ultrasonic sensing and digital twin path optimization system of this embodiment includes an ultrasonic detection module.
[0032] Each smart device is equipped with an ultrasonic detection module. This module emits high-frequency pulse signals via an ultrasonic transmitter, with a frequency range between 40kHz and 200kHz, covering a 180° fan-shaped area in front of the device. Simultaneously, sensors within the module receive the reflected echoes and accurately calculate the distance and direction of obstacles based on the echo information, providing fundamental data for obstacle avoidance and path planning.
[0033] The specific implementation method is as follows:
[0034] A piezoelectric ceramic transducer is used as the ultrasonic transmitter, with a transmission period set to 10ms. The sensor calculates the distance to obstacles based on the time difference, using the following formula: ;
[0035] Where c is the speed of sound, approximately 343 m / s, and t is the echo time. By accurately measuring the echo time, the distance between the obstacle and the equipment is calculated using the formula.
[0036] The frequency shift of the echo is detected using Doppler analysis technology, and then the relative velocity of the obstacle is calculated. The calculation formula is as follows: ;in, For frequency offset, This refers to the transmission frequency.
[0037] The beneficial effects of this embodiment are as follows: By using a high-frequency ultrasonic detection module and a piezoelectric ceramic transducer, combined with Doppler analysis technology, high-precision real-time calculation of obstacle distance, direction, and velocity is achieved (with controllable positioning error). This module provides dynamic and reliable basic data for obstacle avoidance and path planning of intelligent devices, improving the autonomous navigation capability and safety of the devices in complex environments, while reducing the risk of collisions caused by misjudgment and improving operational efficiency.
[0038] Example 2:
[0039] See Figure 1 and Figure 2 As shown, the intelligent device cluster ultrasonic sensing and digital twin path optimization system of this embodiment includes a cluster collaborative sensing network.
[0040] The cluster-based collaborative sensing network uses communication protocols to share ultrasonic data among intelligent devices and fuses this data to generate a global 3D obstacle map. This map comprehensively and accurately reflects the distribution of obstacles in the cluster's operating environment, providing a global perspective for collaborative operations and path planning.
[0041] The specific implementation method is as follows:
[0042] By adopting a self-organizing network communication method, each smart device can autonomously establish a communication link in the absence of fixed communication infrastructure, thereby enabling real-time data transmission and sharing.
[0043] Each smart device uploads data such as the distance, direction, and speed of obstacles it detects to the edge computing node. The edge computing node runs a Kalman filter algorithm to fuse the multi-source data.
[0044] By utilizing triangulation and combining perspective information from multiple devices, the coordinates of obstacles were located with an error of less than 0.5m. Through collaborative observation and data fusion across multiple devices, the accuracy of obstacle localization was significantly improved, providing strong support for subsequent path optimization and obstacle avoidance decisions.
[0045] The beneficial effects of this embodiment are as follows: Through self-organizing network communication and Kalman filtering algorithm, the cluster cooperative perception network realizes real-time sharing and fusion processing of data from multiple devices, generating a global 3D obstacle map (positioning error <0.5m). This network provides comprehensive environmental information for cluster operations, optimizes cooperative path planning and obstacle avoidance strategies, improves the global cooperation efficiency of the device cluster and the task completion rate in complex scenarios, and reduces redundant detection and resource waste.
[0046] Example 3:
[0047] See Figure 1 and Figure 2As shown, the intelligent device cluster ultrasonic sensing and digital twin path optimization system of this embodiment includes a digital twin platform.
[0048] The digital twin platform constructs a digital twin model of the city based on GIS data and integrates a 3D obstacle map generated by a cluster collaborative sensing network, updating dynamic obstacle information in real time. This platform provides users with an intuitive and accurate virtual operating environment, supporting the simulation and evaluation of equipment operating paths.
[0049] The specific implementation method is as follows:
[0050] The static environment of the city's digital twin model, such as buildings, roads, and bridges, is initialized using high-precision GIS data. A 3D obstacle map generated by a clustered collaborative sensing network is input into the digital twin platform in real time, allowing the platform to update the location information of moving obstacles based on the input data. Within the virtual environment of the digital twin platform, the operating path of intelligent devices is simulated, and the obstacle avoidance effect of the path is evaluated. Through simulation evaluation, problems in path planning can be identified in advance and optimized, improving the safety and efficiency of actual operations.
[0051] The beneficial effects of this embodiment are as follows: The digital twin platform integrates high-precision GIS static environmental data with dynamic obstacle maps to construct a real-time updated virtual working environment. By simulating path obstacle avoidance effects, potential problems can be identified and optimized in advance, reducing trial-and-error costs in actual operations. This platform provides users with intuitive decision support, improves the scientific nature and safety of path planning, shortens the debugging cycle, and enhances the adaptability of equipment clusters in dynamic environments.
[0052] Example 4:
[0053] See Figure 1 and Figure 2 As shown, the intelligent device cluster ultrasonic sensing and digital twin path optimization system of this embodiment includes an intelligent path optimization module.
[0054] The intelligent path optimization module dynamically optimizes the operating path and cluster formation of intelligent devices based on information from the digital twin platform and data acquired by the ultrasonic detection module, ensuring that the devices can operate efficiently and safely in complex environments.
[0055] The specific implementation method is as follows:
[0056] S1. Construct a network topology model of the intelligent device cluster operation path, abstract the device operation environment into a graph structure, and form multiple feasible paths from the starting position to the target position and corresponding formation strategies as training samples. Calculate the obstacle avoidance success rate, energy consumption, path length, formation adjustment time and obstacle avoidance time of each path and formation strategy as quality evaluation indicators of the corresponding feasible solutions.
[0057] S2. Construct a deep reinforcement learning agent based on the state space, action space, and reward function. The specific steps are as follows:
[0058] S21. Establish the state space of the intelligent agent. The state space includes two pieces of information: task information and feasible solution state matrix, represented by a vector: S(t) = [T, SM]; where S(t) is a vector representing the state at the current time t; T is the task information, including the starting position of the equipment. Target location The operational scope (operation radius R); SM is the feasible solution state matrix, used to describe the quality indicators of k feasible solutions. The feasible solution state matrix includes obstacle avoidance success rate, energy consumption, path length, formation adjustment time, and obstacle avoidance time. The feasible solution state matrix is represented as:
[0059] The matrix has a dimension of k×5 (k is the number of feasible solutions, and 5 corresponds to 5 quality evaluation indicators); each row represents a feasible solution (a path + the corresponding formation strategy); each column represents a quality evaluation indicator, s is the obstacle avoidance success rate (range [0,1]), which represents the probability that the solution successfully avoids all obstacles; Energy consumption represents the total energy consumed by the cluster of devices executing this scheme; The path length represents the total path distance from the starting position to the target position. Formation adjustment time indicates the time it takes for a cluster of devices to transition from its current formation to the target formation. Obstacle avoidance time represents the extra time that the device cluster spends avoiding obstacles in the path.
[0060] S22. Establish the action space of the agent, which represents the set of feasible paths and formation strategies. The agent makes corresponding decisions through the ε-greed strategy. The decision-making method is as follows: ;in, The optimal strategy is determined under the current environmental state s; 'a' represents the action (including path selection and formation parameter adjustment, such as formation spacing and arrangement); A is the action space (the set of all possible actions); ε is a random number in the range [0,1], and ε is a variable with an initial value of 1 that controls the probability of random exploration (gradually decaying with training iterations); θ is a common parameter of the environment and equipment cluster; β and α are unique parameters of the value function and advantage function, respectively (α is the learning rate); Q(s,a;θ,α,β) is the action value function, used to evaluate the expected benefit of choosing action a under state s.
[0061] S23. Establish the reward function for the agent, expressed as follows: Where R is the reward function; , , , and These represent the weights of the obstacle avoidance success rate, energy consumption, path length, formation adjustment time, and obstacle avoidance time indicators, respectively, and satisfy the following conditions: (Among them, the obstacle avoidance success rate weight) One is a positive weight, and the rest are negative weights, meaning that the lower the energy consumption and the shorter the path, the higher the corresponding reward.
[0062] S24. Define the method by which the agent selects the optimal solution. Specifically, the method involves iteratively updating the action value function using the environment state, the solution selection strategy, and the reward, with the goal of maximizing the expected reward value. The optimal path and formation strategy are selected for each environment state, and the expression for the action value function is: ,in, Indicates the expected reward value; The immediate reward value for choosing action a in state s; The new state transitioned to after performing action a; For the new state Optional actions below; Indicates a new state The maximum reward value for selecting the optimal action; γ is a discount factor (range: [0,1]), which reflects the importance of future rewards (the closer γ is to 1, the higher the weight of future rewards).
[0063] S3. The agent selects samples from the training samples by combining the SumTree priority sampling mechanism with uniform sampling and reweighted sampling mechanisms. The specific method is as follows:
[0064] S31. The agent uses the SumTree priority sampling mechanism to select samples: calculate the priority of each sample, and select a set number of samples in descending order of priority. The formula for calculating the priority is: Where i is the sample number, and each sample corresponds to a feasible solution; The priority of sample i; Let be the temporal difference error of sample i; This is a smoothing constant used to prevent a priority of 0.
[0065] The formula for calculating the timing difference error is as follows: ;in, Let be the temporal difference error of sample i; The immediate reward for sample i; The new state output by the target neural network The action value below; The current network output status ,action The action value below.
[0066] S32. The agent selects samples from the training samples using a uniform sampling and reweighted sampling mechanism: all feasible solution samples are stored in a circular queue, and a set number of samples are uniformly and randomly drawn during each training session. The probability of each sample being selected is: Where N is the total capacity of the sample pool; i is the sample number; Let be the probability that sample i is selected.
[0067] For each selected sample, its temporal difference error is calculated, and the loss function is reweighted. During backpropagation, the loss weights are dynamically adjusted based on the sample temporal difference error. The loss function is defined as: ;in, For loss weights; is the loss weight for sample i; M is the batch size; Let be the temporal difference error of sample i.
[0068] S33. Merge the samples obtained from S31 and S32 to form a sample set.
[0069] S4. During the operation of the equipment cluster, if the quality evaluation index of the monitoring scheme exceeds the threshold (such as the obstacle avoidance success rate being lower than the preset value, energy consumption being higher than the threshold, etc.), the optimal path and formation strategy will be dynamically updated based on the real-time obstacle data of the ultrasonic detection module and the environmental status data of the digital twin platform.
[0070] The beneficial effects of this embodiment are as follows: The intelligent path optimization module based on deep reinforcement learning achieves adaptive adjustment of path and formation strategies by dynamically modeling environmental states (such as obstacle avoidance success rate, energy consumption, etc.), prioritizing sampling mechanisms, and optimizing reward functions. This module can quickly generate optimal solutions based on real-time obstacle data and environmental changes, reducing energy consumption (such as reducing redundant paths), shortening operation time (such as formation adjustment time), and improving the robustness and collaborative efficiency of the equipment cluster in complex scenarios, ensuring efficient task completion.
[0071] The weights in this invention are used to measure the degree of influence of different factors or variables on a particular outcome or decision. Weights are defined as numerical values assigned to each factor when comparing and evaluating multiple factors to reflect their importance or priority. These weights can be determined based on specific circumstances and needs, and are typically developed and confirmed jointly by professionals or relevant stakeholders. By setting appropriate weights, programs or systems can make more accurate decisions or predictions.
[0072] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
[0073] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims
1. An intelligent device cluster ultrasonic sensing and digital twin path optimization system, characterized in that, The system includes an ultrasonic detection module, a cluster collaborative sensing network, a digital twin platform, and an intelligent path optimization module. The ultrasonic detection module is configured in each smart device and is used to obtain the distance, direction and relative speed data of obstacles around the device by emitting high-frequency pulse signals and receiving echoes, so as to provide basic data for obstacle avoidance and path planning of the device. The cluster collaborative sensing network is used to realize the sharing and fusion processing of ultrasonic data among various intelligent devices, and generate a global 3D obstacle map to reflect the distribution of obstacles in the cluster operation environment. The digital twin platform is used to construct a virtual operating environment that includes static environment and dynamic obstacle information, and supports the simulation and evaluation of equipment operation paths; The intelligent path optimization module is used to dynamically optimize the operating path and cluster formation of intelligent devices based on simulation evaluation information from the digital twin platform and real-time data from the ultrasonic detection module, ensuring the efficiency and safety of cluster operations.
2. The intelligent device cluster ultrasonic sensing and digital twin path optimization system according to claim 1, characterized in that, The ultrasonic detection module uses a piezoelectric ceramic transducer as the ultrasonic transmitter; the distance to the obstacle is calculated by combining the echo time difference with the speed of sound, and the relative speed of the obstacle is calculated by combining the frequency offset of the detected echo with the transmission frequency.
3. The intelligent device cluster ultrasonic sensing and digital twin path optimization system according to claim 1, characterized in that, The cluster collaborative sensing network adopts a self-organizing network communication method, enabling each intelligent device to autonomously establish a communication link without fixed communication infrastructure, and realize the real-time transmission and sharing of obstacle distance, direction, and speed data. Each intelligent device uploads its own detection data to the edge computing node, which uses the Kalman filter algorithm to fuse multi-source data and uses triangulation combined with multi-device perspective information to locate the obstacle coordinates, with the positioning error being less than the set value.
4. The intelligent device cluster ultrasonic sensing and digital twin path optimization system according to claim 1, characterized in that, The digital twin platform constructs a city digital twin model based on high-precision GIS data, initializes the static environmental information in the model, including buildings, roads, and bridges; integrates the 3D obstacle map generated by the cluster collaborative perception network into the digital twin model in real time to refresh the location information of moving obstacles in real time; the platform simulates the running path of intelligent devices in a virtual environment, evaluates the obstacle avoidance effect of the path, and discovers path planning problems in advance through simulation and makes optimization adjustments.
5. The intelligent device cluster ultrasonic sensing and digital twin path optimization system according to claim 1, characterized in that, The intelligent path optimization module constructs a network topology model of the device cluster operation path, abstracts the device operation environment into a graph structure, generates multiple feasible paths from the starting position to the target position and corresponding formation strategies as training samples, and calculates the indicators of each path and formation strategy, including obstacle avoidance success rate, energy consumption, path length, formation adjustment time and obstacle avoidance time, and uses these indicators as quality evaluation indicators for the corresponding feasible solutions.
6. The intelligent device cluster ultrasonic sensing and digital twin path optimization system according to claim 5, characterized in that, The intelligent path optimization module constructs a deep reinforcement learning agent, which includes a state space, an action space, and a reward function. The state space is a vector containing task information and a feasible solution state matrix. The task information includes the device's starting position, target position, and operating radius. The feasible solution state matrix is a k×5 matrix used to describe the obstacle avoidance success rate, energy consumption, path length, formation adjustment time, and obstacle avoidance time of k feasible solutions. The action space uses an ε-greedy policy to decide path selection and formation parameter adjustment, where ε is a random exploration probability control variable with an initial value of 1 that gradually decays with training iterations. The reward function calculates a comprehensive reward based on the weights of obstacle avoidance success rate, energy consumption, path length, formation adjustment time, and obstacle avoidance time, with the sum of all weights being 1. The obstacle avoidance success rate is a positive weight, and the others are negative weights.
7. The intelligent device cluster ultrasonic sensing and digital twin path optimization system according to claim 6, characterized in that, The deep reinforcement learning agent selects the optimal solution by iteratively updating the action value function to maximize the expected reward value. The action value function comprehensively considers the immediate reward value of selecting the action in the current state and the maximum reward value of selecting the optimal action in the new state, and introduces a discount factor to reflect the importance of future rewards. The discount factor has a value range of [0,1]. The agent selects samples from the training samples by combining the SumTree priority sampling mechanism with uniform sampling and reweighted sampling mechanisms. The sampling efficiency is optimized by calculating the sample priority and the temporal difference error. A smoothing constant is introduced when calculating the priority to prevent the priority from being 0. The loss function dynamically adjusts the loss weights based on the temporal difference error of the samples.
8. The intelligent device cluster ultrasonic sensing and digital twin path optimization system according to claim 5, characterized in that, During the operation of the equipment cluster, the intelligent path optimization module monitors the quality evaluation indicators of the scheme in real time. When the quality evaluation indicators exceed the set threshold, the module dynamically updates the optimal path and formation strategy based on the real-time obstacle data from the ultrasonic detection module and the environmental status data from the digital twin platform.