Autonomous Task Planning Method for Underground Space Exploration of Multi-Heterogeneous Robots under Dynamic Networking

Abstract

The present invention discloses an autonomous task planning method for multi-heterogeneous robot underground space exploration under dynamic networking, belonging to the technical field of robot task planning. An intelligent task allocation model for air-ground heterogeneous robots is constructed through a task allocation module, and a matching scheme between robots and tasks is generated based on the mapping relationships at the feature layer, task layer, and robot layer. Through the networking module, the signal strength distribution is predicted, the time-delay boundary of signal transmission is determined, and the networking switching mechanism is triggered according to the time-delay boundary to generate a communication topology scheme that is scalable, strongly adaptable, and highly robust. Through the path planning module, the Zernike network is used to predict the trajectory of moving interference targets, and an obstacle avoidance path is generated in combination with the global path planning algorithm. Through the task reallocation and path replanning module, the task allocation is dynamically adjusted, and the local path is optimized using reinforcement learning. Based on the above steps, task planning and autonomous networking of air-ground heterogeneous robots in an underground unstructured irregular space and a satellite signal rejection environment are realized.

Description

Technical Field

[0001] The invention belongs to the technical field of robot task planning, and particularly relates to an autonomous task planning method for multi-heterogeneous robot underground space detection under dynamic networking. Background Technique

[0002] In recent years, with the continuous advancement of urbanization, the underground unstructured space detection task has shown an upward trend. In the past, most underground space detections were carried out based on manpower, single robots, and homogeneous multi-robots. However, in the underground unstructured irregular space and satellite signal rejection environment, traditional robot detection tasks have characteristics such as time and space constraints, multi-source heterogeneous interference, and high-dynamic multi-targets, resulting in a series of problems such as limited task allocation efficiency, limited action ability, difficult space networking, and low path planning efficiency of traditional robots; the detection efficiency based on manpower and traditional robots is low and the cost is high, making it difficult to complete the detection tasks in the above-mentioned underground unstructured irregular space and satellite signal rejection environment.

[0003] How to use air-ground heterogeneous robots for collaborative detection in the underground unstructured irregular space and satellite signal rejection environment and improve their detection efficiency, as well as effectively realize the high-dynamic task planning and collaborative networking mechanism of air-ground heterogeneous robots under spatio-temporal constraint conditions, has become a technical problem to be solved urgently. Therefore, it is particularly important to develop an autonomous task planning method for multi-heterogeneous robot underground space detection under dynamic networking. Summary of the Invention

[0004] The present invention aims at the above problems, makes up for the deficiencies of the prior art, and provides an autonomous task planning method for multi-heterogeneous robot underground space detection under dynamic networking with multi-objective task allocation for multi-heterogeneous robots, an air-ground collaborative self-networking mechanism, and a path collaborative planning effect. The task planning method of the present invention can effectively improve the organic coordination and cross-domain collaboration capabilities of multi-heterogeneous robots, realize the efficient task planning and reliable networking of air-ground heterogeneous robots in the underground unstructured irregular space and satellite signal rejection environment, and greatly improve the detection efficiency.

[0005] To achieve the above object, the present invention adopts the following technical solutions.

[0006] The autonomous task planning method for multi-heterogeneous robot underground space detection under dynamic networking provided by the present invention includes the following steps:

[0007] S1: Construct an intelligent task allocation model for air-ground heterogeneous robots through a task allocation module, and generate a matching scheme between robots and tasks based on the mapping relationship of the feature layer, task layer, and robot layer;

[0008] S2: Predict the signal strength distribution through the networking module, determine the signal transmission delay boundary, and trigger the networking switching mechanism according to the delay boundary to generate a communication topology solution that is customizable, strongly adaptable, and highly robust.

[0009] S3: Use the Zernike network in the path planning module to predict the trajectory of the moving interference target, and generate an obstacle avoidance path in combination with the global path planning algorithm.

[0010] S4: When the sudden change of the environment causes the task or path to fail, dynamically adjust the task allocation through the task reallocation and path replanning module, and use reinforcement learning to optimize the local path to complete the path replanning.

[0011] S5: Based on the above steps, realize the task planning and autonomous networking of the air-ground heterogeneous robots in the underground unstructured irregular space and the satellite signal denial environment.

[0012] As a preferred solution of the present invention, the intelligent task allocation model is a neural network task allocation model, and the task allocation module includes a feature layer, a task layer, and a robot layer; the feature layer: defines the characteristics of the sensors and action components carried by the robot; the task layer: allocates tasks based on the matching degree between the task requirements and the robot characteristics; the robot layer: consists of 5 types of robots, specifically including multi-rotor drones, tilt-rotor drones, tracked robots, legged robots, and air-ground dual-purpose robots, and the total number of 5 types of robots is 12.

[0013] As another preferred solution of the present invention, the networking module includes a signal strength prediction layer, a signal transmission delay boundary determination layer, and a networking switching mechanism layer; the signal strength prediction layer: uses the Zernike network to predict the wireless signal field strength distribution; the signal transmission delay boundary determination layer: establishes the mapping relationship between the signal field strength and the transmission delay, judges the upper bound of the delay from the strength of the signal field strength, and when the signal transmission delay reaches the upper bound of the delay, it means that the connection signal between the robots is weak at this time. If the transmission delay exceeds the upper bound of the delay, the robots do not transmit signals; the networking switching mechanism layer: when the transmission delay of the communication between the robots exceeds the upper bound of the delay, trigger the networking switching mechanism and switch the networking mode.

[0014] As another preferred solution of the present invention, the path planning module includes a moving interference target estimation layer and a path planning layer; the moving interference target estimation layer: uses the Zernike network to depict the driving trajectory of the moving interference target, and predicts the future driving trajectory of the moving interference target according to the change law; the path planning layer: adopts the A-star algorithm, uses the driving trajectory of the moving interference target as a constraint condition to generate a global path, and completes the path planning.

[0015] As another preferred embodiment of the present invention, the task reallocation and path replanning module includes a task reallocation layer and a path replanning layer; the task reallocation layer: taking the matching degree between the robot features and the task as the objective function, dynamically adjusting the task allocation in combination with the distance constraint; the path replanning layer: optimizing the local path through the reinforcement learning method to complete the path replanning.

[0016] As another preferred embodiment of the present invention, the mapping relationship from the task layer to the feature layer represents the degree of dependence of the task on the feature through weights, and the mapping relationship from the robot layer to the feature layer represents the feature capabilities of the robot through weights.

[0017] As another preferred embodiment of the present invention, the Zernike network has the following expression:

[0018] ;

[0019] where m is a positive integer or 0, R ( ρ ) is ρ of n order polynomial; i is the number of terms; l g is the Zernike network weight coefficient, Z g ( ρ , θ ) is the g th Zernike neuron, ε is the modeling error.

[0020] As another preferred embodiment of the present invention, the Zernike network optimizes its parameters using a learning algorithm, and the learning algorithm uses any one of the Levenberg-Marquardt algorithm, the recursive least squares algorithm, the backpropagation algorithm, and the extended Kalman filter algorithm.

[0021] As another preferred embodiment of the present invention, the reinforcement learning method aims to maximize the cumulative reward and learns the local path optimization strategy through the interaction between the robot and the environment.

[0022] As another preferred embodiment of the present invention, the network switching mechanism supports the green communication mode and reduces the energy consumption by dynamically adjusting the transmission power and communication frequency band.

[0023] In addition, the 12 robots included in the robot layer are of 5 types, namely: 2 multi-rotor unmanned aerial vehicles; 3 tilt-rotor unmanned aerial vehicles; 3 tracked robots; 2 legged robots; 2 air-ground dual-purpose robots.

[0024] Advantages of the present invention:

[0025] A method for autonomous task planning of multi - heterogeneous robots for underground space exploration under dynamic networking provided by the present invention combines the task allocation module, networking module, path planning module, and task re - allocation and path re - planning module to complete the construction of an intelligent task allocation model for air - ground heterogeneous robots under multiple constraints, generate a communication topology generation scheme with cut - ability, strong adaptability, and high robustness, and a green communication mode, realizing the key technology of integrated brain - like planning for multi - heterogeneous robots; moreover, it effectively improves the organic coordination and cross - domain cooperation ability of multi - heterogeneous robots, and endows heterogeneous robots with strong autonomous, adaptable, and survivable intelligent behavior capabilities for underground space exploration. Brief Description of the Drawings

[0026] Figure 1 It is a mapping relationship diagram between the feature layer, task layer, and robot layer of the neural network task allocation model constructed by the present invention through the task allocation module. Detailed Embodiments

[0027] In order to make the technical problems, technical solutions, and beneficial effects solved by the present invention clearer, the following further details the present invention in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0028] The method for autonomous task planning of multi - heterogeneous robots for underground space exploration under dynamic networking provided by the embodiments of the present invention includes the following steps:

[0029] S1: Construct an intelligent task allocation model for air - ground heterogeneous robots under multiple constraints through the task allocation module, and generate a matching scheme for robots and tasks based on the mapping relationship between the feature layer, task layer, and robot layer;

[0030] S2: Predict the signal strength distribution and determine the signal transmission delay boundary through the networking module, and trigger the networking switching mechanism according to the delay boundary to generate a cut - able, strongly adaptable, and highly robust communication topology scheme;

[0031] S3: Use the Zernike network to predict the trajectory of moving interference targets through the path planning module, and generate an obstacle - avoidance path in combination with the global path planning algorithm;

[0032] S4: When the environment suddenly changes resulting in task or path failure, dynamically adjust the task allocation through the task re - allocation and path re - planning module, and optimize the local path using reinforcement learning to complete path re - planning;

[0033] S5: Based on the above steps, realize the task planning and autonomous networking of air - ground heterogeneous robots in an underground unstructured and irregular space and a satellite - signal - denied environment.

[0034] In step S1, the task allocation module includes a feature layer, a task layer, and a robot layer, which are used to build an intelligent task allocation model for heterogeneous air-ground robots. The intelligent task allocation model is a neural network task allocation model. For the neural network task allocation model, corresponding relationships are built, and the mapping relationships from the robot layer to the feature layer and from the task layer to the feature layer are established respectively, such as Figure 1 the mapping relationship diagram shown. Finally, the corresponding relationship between the robot and the task is obtained, so as to complete the task allocation, realize task collaboration and information sharing. The feature layer: defines the characteristics of the sensors and action components carried by the robot, and endows the robot with the characteristics and capabilities. Among them, the sensors carried by the robot include cameras, lidar, infrared imagers, temperature sensors, gas detectors, etc.; the action components include wheels, rotors, feet, etc. The task layer: consists of multiple tasks, and whether the task can be executed is determined by the characteristics and capabilities of the robot; the task is allocated based on the matching degree between the task requirements and the robot characteristics. The robot layer: consists of 5 types of robots, specifically including multi-rotor drones, tilt-rotor drones, tracked robots, legged robots, and air-ground dual-purpose robots. The total number of the 5 types of robots is 12. Among them, the 12 robots of the 5 types included in the robot layer are: 2 multi-rotor drones, denoted as A1 and A2 respectively; 3 tilt-rotor drones, denoted as A3, A4, and A5 respectively; 3 tracked robots, denoted as A6, A7, and A8 respectively; 2 legged robots, denoted as A9 and A10 respectively; 2 air-ground dual-purpose robots, denoted as A11 and A12 respectively.

[0035] In step S2, the networking module includes a signal strength prediction layer, a signal transmission time delay boundary determination layer, and a networking switching mechanism layer. The signal strength prediction layer: uses the Zernike network to predict the wireless signal field strength distribution. Since the signal field strength changes violently in the edge areas of different propagation regions, the Zernike network is used as a prediction model to predict the characteristics of the wireless signal field strength distribution. The signal transmission time delay boundary determination layer: establishes the mapping relationship between the signal field strength and the transmission time delay, and judges the upper bound of the time delay from the strength of the signal field strength. When the signal transmission time delay reaches the upper bound of the time delay, it means that the connection signal between the robots is weak at this time. If the transmission time delay exceeds the upper bound of the time delay, the robots do not transmit signals. The networking switching mechanism layer: when the transmission time delay of the communication between the robots exceeds the upper bound of the time delay, the networking switching mechanism is triggered and the networking mode is switched.

[0036] Specifically, the path planning module includes a moving interference target estimation layer and a path planning layer. The path planning module endows path planning to achieve an efficient and reliable autonomous planning function. The moving interference target estimation layer: uses the Zernike network to depict the driving trajectory of the moving interference target and predicts the future driving trajectory of the moving interference target according to the change law. The path planning layer: adopts the A-star algorithm, uses the driving trajectory of the moving interference target as a constraint condition to generate a global path, and completes path planning.

[0037] Specifically, the task reallocation and path replanning module includes a task reallocation layer and a path replanning layer. When encountering sudden environmental changes, such as partial collapse of the underground space, etc., which cause the original task allocation and path planning to fail, the task reallocation and path replanning module can effectively reallocate tasks and replan paths. The task reallocation layer: takes the matching degree between the robot features and the task as the objective function, and dynamically adjusts the task allocation in combination with the distance constraint. Due to sudden environmental changes or the generation of temporary tasks, local task reallocation is required. Taking the matching degree between the robot features and the task as the allocation objective function, and considering the urgency at the same time, the distance from the robot to the task point is used as the constraint condition. The path replanning layer: optimizes the local path through the reinforcement learning method. Using the reinforcement learning method, the robot learns how to select appropriate actions in different states through interaction with the environment to obtain the maximum cumulative reward for local path planning. That is, the path replanning layer uses the reinforcement learning method to achieve the goal of maximizing the cumulative reward and completes the path replanning strategy through the interaction between the robot and the environment.

[0038] Specifically, the mapping relationship from the task layer to the feature layer represents the dependence degree of the task on the feature through weights; the mapping relationship from the robot layer to the feature layer represents the feature capabilities possessed by the robot through weights. <l

[0039] Specifically, for the Zernike network, its expression is:

[0040] ;

[0041] Among them, m is a positive integer or 0, R ( ρ ) is ρ of [[ID=Z5]] n order polynomial; i is the number of terms; l g is the Zernike network weight coefficient, Z g ( ρ , θ ) is the g th Zernike neuron,ε is the modeling error. The Zernike network has strong nonlinear approximation ability, fast convergence, strong generalization ability, "self-learning ability", robustness and anti-interference performance, and can be well applied as an approximation model and a classification model.

[0042] The Zernike network optimizes its parameters using a learning algorithm, and the learning algorithm can be any one of the Levenberg-Marquardt algorithm, the recursive least squares algorithm, the backpropagation algorithm, and the extended Kalman filter algorithm; the learning algorithm can also use other algorithms that can meet the optimization function, not limited to the above several learning algorithms.

[0043] Specifically, the network switching mechanism supports the green communication mode and reduces energy consumption by dynamically adjusting the transmission power and communication frequency band.

[0044] In summary, through the embodiments provided by the present invention, the following outstanding advantages of the present invention can be summarized:

[0045] (1) The present invention combines task allocation technology, prediction technology, communication technology, intelligent optimization technology, and artificial intelligence technology to design an autonomous task planning method for multi-heterogeneous robots in underground space exploration under dynamic networking. Through the task allocation module, networking module, path planning module, task reallocation and path replanning module, an intelligent task allocation model for air-ground heterogeneous robots is constructed, a communication topology scheme and a green communication mode with cuttable, strong adaptability, and high robustness are generated, and an integrated planning scheme for multi-heterogeneous robots is proposed. The present invention effectively improves the organic coordination and cross-domain cooperation capabilities of multi-heterogeneous robots, and endows heterogeneous robots with intelligent behavior capabilities such as strong autonomy, strong adaptability, and strong survival in underground space exploration; it has the beneficial effects of high allocation efficiency, excellent networking mechanism, and reliable planning scheme.

[0046] (2) The present invention combines task allocation technology and artificial intelligence technology. By establishing the relationship between the feature layer and the task layer, and the feature layer and the robot layer, an intelligent task allocation model for air-ground heterogeneous robots under multiple constraints is constructed to achieve task coordination and information sharing.

[0047] (3) The present invention combines communication technology and prediction technology. Through signal strength prediction, the relationship between signal strength and transmission delay boundary is established, and an adaptive network switching mechanism is generated according to the relationship between transmission delay and delay boundary, so as to generate a communication topology scheme and a green communication mode with cuttable, strong adaptability, and high robustness.

[0048] (4) The present invention integrates artificial intelligence technology and prediction technology, uses the Zernike network to characterize the driving trajectory of the mobile interference target, and uses the A-star algorithm to use the driving trajectory of the mobile interference target as a constraint condition to generate a global path and complete autonomous path planning.

[0049] (5) The present invention integrates task allocation technology, artificial intelligence technology, and intelligent optimization technology. It establishes the matching degree between robot capability and task as the allocation objective function, and uses the distance between the robot and the task point as the constraint condition to achieve task reallocation. In addition, the robot learns how to choose appropriate actions in different states to obtain the maximum cumulative reward by interacting with the environment to perform local path planning and achieve path replanning.

[0050] (6) The present invention integrates artificial intelligence technology and utilizes the Zernike network, which has strong nonlinear approximation, fast convergence, strong generalization ability, "self-learning ability", robustness and anti-interference performance, and can be well applied as an approximation model and classification model.

[0051] In summary, the present invention combines task allocation technology, prediction technology, intelligent optimization technology and artificial intelligence technology to design a multi-heterogeneous robot autonomous task allocation method for underground space exploration, giving heterogeneous robots strong autonomous, strong adaptability, strong survival and other intelligent behavioral capabilities, improving the efficiency and safety of detection task allocation in underground unstructured irregular space and satellite signal denial environment, and has the beneficial effects of automation, digitization and intelligence.

[0052] It can be understood that the above specific description of the present invention is only used to illustrate the present invention and is not limited to the technical solutions described in the embodiments of the present invention. Those skilled in the art should understand that the present invention can still be modified or replaced by equivalents to achieve the same technical effects; as long as the use requirements are met, they are within the scope of protection of the present invention.

Claims

1. An autonomous mission planning method for multi - heterogeneous robot underground space exploration under dynamic networking, characterized in that: Including the following steps: S1: Build an intelligent task allocation model for aerial-ground heterogeneous robots through the task allocation module. Based on the mapping relationships of the feature layer, task layer, and robot layer, generate a matching scheme for robots and tasks. The mapping relationship from the task layer to the feature layer represents the degree of dependence of tasks on features through a weight matrix, and the mapping relationship from the robot layer to the feature layer represents the feature capabilities of robots through a weight matrix. S2: Through the networking module, predict the signal strength distribution, determine the signal transmission delay boundary, and trigger the networking switching mechanism according to the delay boundary to generate a communication topology scheme that is scalable, strongly adaptable, and highly robust. The networking module includes a signal strength prediction layer, a signal transmission delay boundary determination layer, and a networking switching mechanism layer. The signal strength prediction layer: Use the Zernike network to predict the wireless signal field strength distribution. The signal transmission delay boundary determination layer: Establish the mapping relationship between the signal field strength and the transmission delay, judge the upper bound of the delay from the strength of the signal field. When the signal transmission delay reaches the upper bound of the delay, it means that the connection signal between robots is weak at this time. If the transmission delay exceeds the upper bound of the delay, the robots do not transmit signals. The networking switching mechanism layer: When the transmission delay of communication between robots exceeds the upper bound of the delay, trigger the networking switching mechanism and switch the networking mode. S4: When the environment suddenly changes and causes task or path failure, through the task reallocation and path replanning module, dynamically adjust the task allocation, and use reinforcement learning to optimize the local path to complete the path replanning. The path planning module includes a moving interference target estimation layer and a path planning layer. The moving interference target estimation layer: Use the Zernike network to depict the driving trajectory of the moving interference target and predict the future driving trajectory of the moving interference target according to the change law. The path planning layer: Adopt the A-star algorithm, use the driving trajectory of the moving interference target as a constraint condition to generate a global path and complete the path planning. The Zernike network, its expression is: ； wherein, m is a positive integer or 0, R ( ρ ) is ρ 's n th-degree polynomial; i is the number of terms; l g is the Zernike network weight coefficient, Z g ( ρ , θ ) is the g th Zernike neuron, ε is the modeling error; S4: When the environment suddenly changes and causes task or path failure, through the task reallocation and path replanning module, dynamically adjust the task allocation, and use reinforcement learning to optimize the local path to complete the path replanning. S5: Based on the above steps, realize the task planning and autonomous networking of aerial-ground heterogeneous robots in an underground unstructured irregular space and a satellite signal denial environment.

2. The autonomous task planning method for multi - heterogeneous robot underground space exploration under dynamic networking according to claim 1, wherein: The intelligent task allocation model is a neural network task allocation model. The task allocation module includes a feature layer, a task layer, and a robot layer. The feature layer: Define the characteristics of the sensors and action components carried by the robots. The task layer: Allocate tasks based on the matching degree between task requirements and robot features. The robot layer: Consists of multi-rotor drones, tilt-rotor drones, tracked robots, legged robots, and aerial-ground dual-purpose robots.

3. The autonomous task planning method for multi - heterogeneous robot underground space exploration under dynamic networking according to claim 1, characterized in that: The task reallocation and path replanning module includes a task reallocation layer and a path replanning layer. The task reallocation layer: Take the matching degree between robot features and tasks as the objective function, and dynamically adjust the task allocation in combination with distance constraints. The path replanning layer: Optimize the local path through reinforcement learning methods.

4. The autonomous mission planning method for multi - heterogeneous robot underground space exploration under dynamic networking according to claim 1, wherein: The Zernike network optimizes its parameters using a learning algorithm, and the learning algorithm uses any one of the Levenberg-Marquardt algorithm, the recursive least squares algorithm, the backpropagation algorithm, and the extended Kalman filter algorithm.

5. The autonomous task planning method for underground space exploration of multiple heterogeneous robots under dynamic networking according to claim 3, characterized in that: The reinforcement learning method aims to maximize the cumulative reward and learns the local path optimization strategy through the interaction between the robot and the environment.

6. The autonomous task planning method for multi - heterogeneous robot underground space exploration under dynamic networking according to claim 1, wherein: The network switching mechanism supports the green communication mode and reduces energy consumption by dynamically adjusting the transmission power and communication frequency band.

Images