Disaster area robot operation path and energy consumption optimization method and system

Abstract

The application discloses a kind of disaster area robot operation path and energy consumption optimization method and system, the method includes: constructing the unit path energy consumption prediction model of multiple factor coupling, fusion disaster area topography, real-time wind resistance and battery attenuation factor, generate accurate energy consumption map;Build hierarchical reinforcement learning path scheduling framework, the upper strategy network is with energy consumption and time optimal as double target to generate global path, lower execution network is based on MPC real-time processing sensor data, when obstacle avoidance, give consideration to path stability and low energy consumption;Design event-triggered task plug-in mechanism optimization scheduling;Compress upper strategy network using knowledge distillation, deploy to robot edge end.The application improves endurance and task completion rate through accurate energy consumption model, hierarchical architecture solves global-local control disconnection, plug-in mechanism meets high real-time demand, edge deployment adapts to weak network / offline environment, suitable for post-disaster inspection, material transportation and other high real-time scenarios.

Description

Technical Field

[0001] This invention relates to the field of disaster relief robot path planning and energy management technology, specifically a method and system for optimizing robot operation paths and energy consumption in disaster areas. Background Technology

[0002] In disaster-stricken environments, characterized by complex and varied terrain, severely damaged roads, and harsh weather conditions, traditional robot path planning methods mainly face the following three types of technical problems:

[0003] 1. Inaccurate energy consumption prediction: Existing methods often use the shortest path or shortest time as the objective, and the energy consumption model often only considers distance factors, failing to fully consider factors such as terrain elevation difference, slope, surface material, weather conditions and battery performance degradation, resulting in a significant difference between the planned path and the actual energy consumption, which affects the robot's endurance and task completion rate.

[0004] 2. Slow dispatch response: Disaster area tasks often change frequently, such as adding, canceling, or adjusting task locations. Traditional methods usually require recalculating the entire path, leading to response delays and affecting the real-time performance of tasks such as repair and material delivery.

[0005] 3. Disconnect between global and local control: Existing path planning algorithms lack an effective coordination mechanism between global decision-making and local obstacle avoidance control. Local obstacle avoidance is prone to deviating from the global planned path, resulting in increased energy consumption or task delays. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method and system for optimizing robot operation paths and energy consumption in disaster areas, aiming to solve the problems mentioned in the background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for optimizing the operation path and energy consumption of robots in disaster areas, comprising the following steps:

[0008] Step S1: Construct a unit path energy consumption prediction model for the robot;

[0009] Step S2: Construct a hierarchical reinforcement learning path scheduling framework based on the unit path energy consumption prediction model to achieve optimal path scheduling for robot energy consumption and time; the hierarchical reinforcement learning path scheduling framework consists of an upper-level policy network and a lower-level execution network.

[0010] Step S3: Design an event-triggered task plugin mechanism to optimize the hierarchical reinforcement learning path scheduling framework;

[0011] Step S4: Use knowledge distillation to compress the upper-layer policy network, and deploy the hierarchical reinforcement learning path scheduling framework after compression to the robot edge computing platform to achieve autonomous path decision-making and execution under weak network or network outage conditions.

[0012] Furthermore, the unit path energy consumption prediction model is expressed as:

[0013] ;

[0014] In the formula, This represents the total energy consumption of a robot along any path segment in a complex post-disaster environment. Indicates the quality of the robot; Represents gravitational acceleration; Indicates the difference in elevation; Indicates the slope angle; Indicates the horizontal projection distance of the path segment; Indicates the coefficient of surface friction; Indicates the drag coefficient; Indicates air density; This indicates the robot's frontal area; This indicates the relative speed between the robot and the wind. express Battery efficiency at any given time.

[0015] Furthermore, the upper-layer policy network constructs a global topology grid map based on a unit path energy consumption prediction model. , This represents the set of vertices, or nodes, of a global topological raster map. Let be the set of edges; each edge Associate energy consumption and time attributes; energy consumption attribute uses The time attribute is represented as: , Indicates the reference speed under the current terrain; Represents the first in the global topological grid map An edge is a path segment connecting two vertices. Based on a global topological grid map, a reinforcement learning policy network is used to train path decision-making with the composite reward function of "minimum path energy consumption + shortest task time". Finally, the optimal global path selection strategy is output, generating a global path with the best energy consumption and the best time.

[0016] Furthermore, based on the lower-level execution network, local obstacle avoidance and trajectory smoothing are performed on the global path with optimal energy consumption and optimal time:

[0017] After generating the energy-optimal and time-optimal global path, at the beginning of each robot control cycle, the robot's local state is first collected in real time by sensors to construct the robot state vector. , This represents the robot's current two-dimensional coordinates. This indicates the robot's current heading angle. , These represent the robot's current linear velocity and angular velocity, respectively. This indicates the robot's current real-time battery level. Indicates transpose;

[0018] Extract the future from the optimal global path. Global path reference point for steps ;

[0019] Construct a kinematic / dynamic model of the robot to describe its state vector at step k. Add the control command for the robot's k-th step Then, the state vector of the robot at step k+1. The mapping relationship; , This represents the expected linear velocity at step k. This represents the expected angular velocity at step k;

[0020] Construct the MPC objective function and quantify the optimization objective;

[0021] Obstacle information is acquired through real-time perception and transformed into obstacle avoidance constraints for the MPC objective function;

[0022] Solve the MPC objective function to obtain the control commands;

[0023] Execute the current control command and enter the next control cycle.

[0024] Furthermore, the MPC objective function is expressed as:

[0025] ;

[0026] In the formula, This indicates the control commands for the robot; , , Both represent weight matrices.

[0027] Furthermore, the reward function is:

[0028] ;

[0029] In the formula, This represents the reward value of reinforcement learning, user and guidance path optimization; Indicates the total number of path segments; , All represent task weight coefficients; Indicates the first Time consumption of a segment path.

[0030] Furthermore, the specific process of step S3 is as follows: when adding, deleting, or adjusting tasks, first detect the affected areas of the global path with the best energy consumption and best time; only call the upper-layer policy network to replan for the affected areas, while the remaining paths remain unchanged.

[0031] A disaster area robot operation path and energy consumption optimization system includes:

[0032] The energy consumption model building module is used to build a unit path energy consumption prediction model for the robot.

[0033] The scheduling module is used to build a hierarchical reinforcement learning path scheduling framework based on the unit path energy consumption prediction model, so as to achieve the optimal path scheduling for robot energy consumption and time. The hierarchical reinforcement learning path scheduling framework consists of an upper-level policy network and a lower-level execution network.

[0034] The optimization module is used to design an event-triggered task plugin mechanism to optimize the hierarchical reinforcement learning path scheduling framework.

[0035] The deployment module is used to compress the upper-layer policy network using knowledge distillation, and then deploy the resulting hierarchical reinforcement learning path scheduling framework to the robot's edge computing platform to achieve autonomous path decision-making and execution under weak or offline network conditions.

[0036] An electronic device includes a processor, a memory, and a bus, wherein the processor and the memory are connected via the bus, wherein the memory is used to store a set of program code, and the processor is used to call the program code stored in the memory to execute a method for optimizing the operation path and energy consumption of a robot in a disaster area.

[0037] A non-volatile computer storage medium stores computer-executable instructions that execute a method for optimizing the operation path and energy consumption of a robot in a disaster area.

[0038] Compared with existing technologies, the present invention has the following advantages:

[0039] (1) This invention constructs a multi-factor coupled unit path energy consumption prediction model, integrates disaster area terrain (elevation, slope, surface material), real-time wind resistance and battery attenuation factor, dynamically generates accurate energy consumption map, completely solves the deviation problem of traditional methods that rely solely on distance / time to estimate energy consumption, provides a reliable benchmark for path optimization, effectively avoids robot power outages midway, and significantly improves single-operation endurance and task completion rate; then, relying on the hierarchical reinforcement learning architecture of "upper-layer policy network + lower-layer execution network", the upper layer generates the global optimal path with "minimum energy consumption + shortest task time" as dual objectives (different task priorities can be adapted by weights), and the lower layer processes sensor data in real time based on model predictive control (MPC), and by penalizing trajectory deviation and acceleration mutation, it does not deviate from the global strategy when avoiding obstacles locally, and reduces additional energy consumption and mechanical wear, solving the problem of disconnect between global and local control, and achieving the unity of path optimality and execution stability.

[0040] (2) In view of the dynamic change characteristics of disaster area tasks, the present invention designs a task plug-in rescheduling mechanism. The task change is the trigger condition to replan only the affected local path, reuse the existing energy consumption model and strategy network, avoid the redundancy of full path recalculation, realize second-level scheduling response, meet the high real-time requirements of emergency repair and material delivery, and reduce secondary losses caused by delay. Furthermore, the strategy network is compressed by knowledge distillation and deployed at the edge of the robot to ensure that it can still autonomously complete energy consumption prediction, path planning and local obstacle avoidance in weak network or network outage environments. Multiple robots can also achieve regional collaboration through edge parameter interaction, breaking the communication dependency limitation. Attached Figure Description

[0041] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0042] like Figure 1 As shown, the present invention provides a technical solution: a method and system for optimizing the operation path and energy consumption of robots in disaster areas, comprising the following steps:

[0043] Step S1: Construct a prediction model for the robot's energy consumption per unit path.

[0044] The unit path energy consumption prediction model decomposes the total energy consumption of a robot along any path segment in a complex post-disaster environment into mechanical work and battery aging loss, which can be expressed as:

[0045] ;

[0046] In the formula, This represents the total energy consumption of a robot along any path segment in a complex post-disaster environment. Indicates mechanical work; This indicates battery aging and wear.

[0047] Among them, mechanical work It includes four parts: climbing energy consumption, slope resistance, surface friction, and wind resistance.

[0048] Energy consumption during hill climbing (gravitational potential energy):

[0049] ;

[0050] In the formula, Indicates the robot's mass (kg); This represents the acceleration due to gravity (9.8 m / s²). This represents the elevation difference (m).

[0051] Slope resistance (ramp component):

[0052] ;

[0053] In the formula, Indicates the slope angle; This represents the horizontal projection distance (m) of the path segment.

[0054] Surface friction (rolling resistance):

[0055] ;

[0056] In the formula, This represents the surface friction coefficient (dimensionless).

[0057] Wind resistance (air resistance):

[0058] ;

[0059] In the formula, Indicates the drag coefficient; Indicates air density; This indicates the robot's frontal area; This represents the relative speed between the robot and the wind.

[0060] Among them, battery aging loss It can be represented as:

[0061] ;

[0062] In the formula, express Battery efficiency at any given time (reflecting the degree of battery aging).

[0063] Among them, mechanical work Various parts and battery aging and wear Combining these factors, a unit path energy consumption prediction model is finally obtained based on elevation slope, surface material (friction coefficient), wind resistance parameters, and battery degradation:

[0064] .

[0065] Step S2: Construct a hierarchical reinforcement learning path scheduling framework based on the unit path energy consumption prediction model to achieve optimal path scheduling for robot energy consumption and time.

[0066] The hierarchical reinforcement learning path scheduling framework consists of an upper policy layer and a lower execution layer.

[0067] The upper-layer policy network constructs a global topology grid map based on a unit path energy consumption prediction model. , This represents the set of vertices, or nodes, of a global topological raster map. Let be the set of edges; each edge Associate energy consumption and time attributes; energy consumption attribute uses The time attribute can be represented as: , Indicates the reference speed under the current terrain; Represents the first in the global topological grid map An edge is a path segment connecting two vertices. Based on a global topological grid map, a reinforcement learning policy network is used for path decision training, with "minimum path energy consumption + shortest task time" as the composite reward function. The network ultimately outputs the optimal global path selection strategy, generating the global path with optimal energy consumption and optimal time. The reward function is:

[0068] ;

[0069] In the formula, This represents the reward value of reinforcement learning, user and guidance path optimization; Indicates the total number of path segments; , Both represent task weight coefficients, which are used to balance the weights of the two objectives: minimizing path energy consumption and minimizing task time. Indicates the first Time consumption of a segment path.

[0070] Based on the lower-level execution network, local obstacle avoidance and trajectory smoothing are performed on the global path with optimal energy consumption and optimal time to ensure that the optimal global path selection strategy does not deviate and that execution is stable; specifically:

[0071] 1. After generating the global path with optimal energy consumption and optimal time, at the beginning of each robot control cycle, the robot's local state is first collected in real time by sensors to construct the robot state vector. , This represents the robot's current two-dimensional coordinates. This indicates the robot's current heading angle. , These represent the robot's current linear velocity and angular velocity, respectively. This indicates the robot's current real-time battery level. This indicates transpose.

[0072] 2. Extract the future path from the optimal global path. Global path reference point for steps .

[0073] 3. Construct a kinematic / dynamic model of the robot to describe its state vector at step k. Add the control command for the robot's k-th step Then, the state vector of the robot at step k+1. The mapping relationship; , This represents the expected linear velocity at step k. This represents the expected angular velocity at step k.

[0074] 4. Construct the MPC objective function and quantify the optimization objective.

[0075] The objective function's role is to "transform the qualitative requirements of 'precise tracking, stable control, and low energy consumption' into calculable mathematical indicators," which can be expressed as:

[0076] ;

[0077] In the formula, This indicates the control commands for the robot; , , Both represent weight matrices.

[0078] 5. Generate obstacle avoidance constraints and exclude collision zones.

[0079] Obstacle information is acquired through real-time perception and transformed into obstacle avoidance constraints for the MPC objective function, ensuring that the predicted trajectory does not collide.

[0080] 6. Solve for the MPC objective function to obtain the control instructions.

[0081] 7. Execute the current control command and enter the next control cycle.

[0082] Step S3: Design an event-triggered task plugin mechanism to optimize the hierarchical reinforcement learning path scheduling framework.

[0083] Disaster area tasks often change dynamically. When adding, deleting or adjusting tasks, the first step is to detect the affected areas of the global path with the best energy consumption and best time. Only the affected areas are replanned by calling the upper-level policy network, while the remaining paths remain unchanged.

[0084] Step S4: Use knowledge distillation to compress the upper-layer policy network, and deploy the hierarchical reinforcement learning path scheduling framework after compression to the robot edge computing platform to achieve autonomous path decision-making and execution under weak network or network outage conditions.

[0085] A disaster area robot operation path and energy consumption optimization system includes:

[0086] The energy consumption model building module is used to build a unit path energy consumption prediction model for the robot.

[0087] The scheduling module is used to build a hierarchical reinforcement learning path scheduling framework based on the unit path energy consumption prediction model, so as to achieve the optimal path scheduling for robot energy consumption and time. The hierarchical reinforcement learning path scheduling framework consists of an upper-level policy network and a lower-level execution network.

[0088] The optimization module is used to design an event-triggered task plugin mechanism to optimize the hierarchical reinforcement learning path scheduling framework.

[0089] The deployment module is used to compress the upper-layer policy network using knowledge distillation, and then deploy the resulting hierarchical reinforcement learning path scheduling framework to the robot's edge computing platform to achieve autonomous path decision-making and execution under weak or offline network conditions.

[0090] An electronic device includes a processor, a memory, and a bus, wherein the processor and the memory are connected via the bus, wherein the memory is used to store a set of program code, and the processor is used to call the program code stored in the memory to execute a method for optimizing the operation path and energy consumption of a robot in a disaster area.

[0091] A non-volatile computer storage medium stores computer-executable instructions that execute a method for optimizing the operation path and energy consumption of a robot in a disaster area.

[0092] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing the operation path and energy consumption of robots in disaster areas, characterized in that, Includes the following steps: Step S1: Construct a unit path energy consumption prediction model for the robot; Step S2: Construct a hierarchical reinforcement learning path scheduling framework based on the unit path energy consumption prediction model to achieve optimal path scheduling for robot energy consumption and time; the hierarchical reinforcement learning path scheduling framework consists of an upper-level policy network and a lower-level execution network. Step S3: Design an event-triggered task plugin mechanism to optimize the hierarchical reinforcement learning path scheduling framework; Step S4: Use knowledge distillation to compress the upper-layer policy network, and deploy the hierarchical reinforcement learning path scheduling framework after compressing the upper-layer policy network to the robot edge computing platform to realize autonomous path decision-making and execution under weak network or network outage conditions. The unit path energy consumption prediction model is expressed as: ； In the formula, This represents the total energy consumption of a robot along any path segment in a complex post-disaster environment. Indicates the quality of the robot; Represents gravitational acceleration; Indicates the difference in elevation; Indicates the slope angle; Indicates the horizontal projection distance of the path segment; Indicates the coefficient of surface friction; Indicates the drag coefficient; Indicates air density; This indicates the robot's frontal area; This indicates the relative speed between the robot and the wind. express Battery efficiency at any given time; The upper-layer policy network constructs a global topology grid map based on a unit path energy consumption prediction model. , This represents the set of vertices, or nodes, of a global topological raster map. Let be the set of edges; each edge Associate energy consumption and time attributes; energy consumption attribute uses The time attribute is represented as: , Indicates the reference speed under the current terrain; Represents the first in the global topological grid map An edge is a path segment that connects two vertices; Based on a global topological grid map, a reinforcement learning policy network is used to train path decision-making with "minimum path energy consumption + shortest task time" as the composite reward function. Finally, the optimal global path selection strategy is output, generating a global path with the best energy consumption and the best time. Based on the lower-level execution network, local obstacle avoidance and trajectory smoothing are performed on the global path with optimal energy consumption and optimal time: After generating the energy-optimal and time-optimal global path, at the beginning of each robot control cycle, the robot's local state is first collected in real time by sensors to construct the robot state vector. , This represents the robot's current two-dimensional coordinates. This indicates the robot's current heading angle. , These represent the robot's current linear velocity and angular velocity, respectively. This indicates the robot's current real-time battery level. Indicates transpose; Extract the future from the optimal global path. Global path reference point for steps ; Construct a kinematic / dynamic model of the robot to describe its state vector at step k. Add the control command for the robot's k-th step Then, the state vector of the robot at step k+1. The mapping relationship; , This represents the expected linear velocity at step k. This represents the expected angular velocity at step k; Construct the MPC objective function and quantify the optimization objective; Obstacle information is acquired through real-time perception and transformed into obstacle avoidance constraints for the MPC objective function; Solve the MPC objective function to obtain the control commands; Execute the current control command and enter the next control cycle.

2. The method for optimizing the operation path and energy consumption of a robot in a disaster area according to claim 1, characterized in that: The objective function of MPC is expressed as: ； In the formula, This represents the robot's control commands; , , Both represent weight matrices.

3. The method for optimizing the operation path and energy consumption of a robot in a disaster area according to claim 2, characterized in that: The reward function is: ； In the formula, This represents the reward value of reinforcement learning, user and guidance path optimization; Indicates the total number of path segments; , All represent task weight coefficients; Indicates the first Time consumption of a segment path.

4. The method for optimizing the operation path and energy consumption of a robot in a disaster area according to claim 3, characterized in that: The specific process of step S3 is as follows: when adding, deleting or adjusting tasks, first detect the affected areas of the global path with the best energy consumption and best time; only call the upper-level policy network to replan for the affected areas, while the other paths remain unchanged.

5. A disaster area robot operation path and energy consumption optimization system, used to execute the disaster area robot operation path and energy consumption optimization method according to any one of claims 1-4, characterized in that, include: The energy consumption model building module is used to build a unit path energy consumption prediction model for the robot. The scheduling module is used to build a hierarchical reinforcement learning path scheduling framework based on the unit path energy consumption prediction model, so as to achieve the optimal path scheduling for robot energy consumption and time. The hierarchical reinforcement learning path scheduling framework consists of an upper-level policy network and a lower-level execution network. The optimization module is used to design an event-triggered task plugin mechanism to optimize the hierarchical reinforcement learning path scheduling framework. The deployment module is used to compress the upper-layer policy network using knowledge distillation, and then deploy the resulting hierarchical reinforcement learning path scheduling framework to the robot's edge computing platform to achieve autonomous path decision-making and execution under weak or offline network conditions.

6. An electronic device, characterized in that, The system includes a processor, a memory, and a bus. The processor and the memory are connected via the bus. The memory is used to store a set of program code, and the processor is used to call the program code stored in the memory to execute the method for optimizing the operation path and energy consumption of a robot in a disaster area as described in any one of claims 1-4.

7. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer can execute instructions to perform the method for optimizing the operation path and energy consumption of a robot in a disaster area as described in any one of claims 1-4.

Images