Method and System for Optimizing the Path of an Automatic Hospital Delivery Robot Based on Deep Learning

Abstract

The present invention belongs to the field of automated intelligent technologies, and specifically relates to a path optimization method and system for hospital automatic delivery robots based on deep learning. The path optimization method for hospital automatic delivery robots based on deep learning uses terrain, obstacle, and delivery demand data for simulation runs to obtain robot driving data. By comparing the driving data with the demands, the task priorities are determined, and the best delivery paths are matched. The AI model adjusts the proportion of path influence factors according to the priorities, thereby optimizing the paths. This method can effectively handle complex conditions and sudden demands in hospitals and solve traditional problems.

Description

Technical Field

[0001] The present invention belongs to the field of automated intelligent technology, and specifically relates to a path optimization method and system for hospital automatic delivery robots based on deep learning. Background Art

[0002] In the prior art, the internal material distribution in hospitals mainly relies on manual or simple automated equipment. This method is not only inefficient and costly, but also often performs poorly in dealing with emergency distribution requirements. With the progress of technology and the improvement of the level of intelligence, using automatic delivery robots for hospital material distribution has become an effective means to solve these problems. However, traditional automatic delivery robots usually lack flexibility and adaptability in path planning, and cannot fully consider the complex terrain conditions and sudden distribution requirements in hospitals, resulting in limited distribution efficiency and accuracy. Summary of the Invention

[0003] (I) Technical Problems to be Solved

[0004] The present invention mainly aims at the above problems and proposes a path optimization method and system for hospital automatic delivery robots based on deep learning, aiming to solve the problem of how to improve the distribution efficiency and accuracy of hospital automatic delivery robots.

[0005] (II) Technical Solutions

[0006] To achieve the above object, the first aspect of the present invention provides a path optimization method for hospital automatic delivery robots based on deep learning, including the following steps:

[0007] Obtain the terrain and obstacle information of all target areas, as well as the distribution demand data of each target area;

[0008] Layout according to the distribution demand data and the terrain obstacle information of all target areas in the map model, and perform a simulation run on the map model;

[0009] During the simulation run, obtain the simulated driving data of all the delivery robots;

[0010] Compare the simulated driving data with the distribution demand data. If there is simulated driving data of the delivery robot that meets the preset conditions, obtain the priority level of the delivery task;

[0011] Match according to the delivery task priority level in the preset task library to obtain the best delivery path corresponding to the task, all path influencing factors, and the preset proportion of each path influencing factor, and determine the scheduling priority level of the delivery task;

[0012] According to the scheduling priority, the demand data and priority of the delivery tasks that meet the preset conditions are sequentially input into the trained AI model to obtain the latest proportion value;

[0013] Adjust the optimal delivery path corresponding to the task according to all the path impact factors of the delivery task and the latest proportion value of each path impact factor to obtain the optimal path.

[0014] Further, the process of collecting the terrain and obstacle information and delivery demand data includes using a camera to perform real-time scanning of the internal and external environments of the hospital, and automatically updating the obstacle information in the map model through image recognition technology.

[0015] Further, the delivery demand data includes the actual quantity of delivered items and the estimated delivery time, and the simulated driving data includes the simulated delivery time and the simulated delivery efficiency; compare the simulated driving data with the delivery demand data. If the simulated delivery time of the delivery robot is lower than the estimated delivery time, or the simulated delivery efficiency of the delivery robot is higher than the preset efficiency, then obtain the priorities of all delivery tasks.

[0016] Further, the creation of the map model includes:

[0017] Collect map data inside the hospital, including building floor plans or three-dimensional structure data;

[0018] According to the collected map data, establish the data structure of the map;

[0019] Mark the obstacles inside the hospital in the data structure of the map;

[0020] Associate the delivery demand data of each target area inside the hospital with the data structure of the map and record it in the map model;

[0021] Calculate the optimal path of the delivery robot according to the map structure and the delivery demand data;

[0022] Integrate the created map model into the hospital's automatic delivery system.

[0023] Further, the steps of calculating the optimal path of the delivery robot include:

[0024] Let G(V,E) represent the map model, where V is the set of nodes and E is the set of edges;

[0025] Each node v∈V represents a location inside the hospital;

[0026] Each edge e∈E represents a path between two locations, and its weight is the length or cost of the path;

[0027] Mark the starting position as visited and set the distance from the starting position to 0;

[0028] Add the starting position to a priority queue Q, which is sorted according to the distance from the starting position;

[0029] Take out the node u closest to the starting position from the priority queue Q and mark it as visited;

[0030] For all neighbor nodes v of node u, calculate the distance from the starting position to node v and update the distance of node v in the priority queue Q;

[0031] Repeat the above steps until the priority queue Q is empty or the target position is visited;

[0032] According to the calculated shortest path information, reconstruct the best path from the starting position to the target position in reverse.

[0033] Further, before the step of matching in the preset task library according to the priority of the delivery task, it further includes:

[0034] Obtain the historical execution data of all delivery tasks, and any one of the historical execution data includes the best delivery path, the path influence factor, and the actual proportion of each path influence factor;

[0035] According to all the historical execution data of the same delivery task, count the best delivery path with the highest usage frequency and the preset number of path influence factors with the highest occurrence frequency, and determine the preset proportion of each path influence factor;

[0036] After binding the best delivery path, the path influence factor, and the preset proportion of each path influence factor of the same delivery task, save them to the preset task library.

[0037] Further, the step of saving the best delivery path, the path influence factor, and the preset proportion of each path influence factor of the same delivery task to the preset task library after binding includes:

[0038] According to the same best delivery path, configure the scheduling priority of each path influence factor from high to low according to the occurrence frequency of each path influence factor;

[0039] According to the scheduling priority, save the best delivery path, the path influence factor, and the preset proportion of each path influence factor of the same delivery task to the preset task library.

[0040] Further, the step of obtaining the terrain and obstacle information of all target areas and the delivery demand data of each target area includes: detecting the simulation running process in real time, and when the battery power of any one of the delivery robots is lower than the threshold or an unforeseen obstacle is encountered, re-planning the path through a deep learning algorithm.

[0041] Further, before the step of inputting the demand data and priority of the delivery tasks that meet the preset conditions into the trained AI model in sequence according to the scheduling priority to obtain the latest proportion value, it further includes:

[0042] Constructing an AI model through a deep reinforcement learning algorithm;

[0043] Obtaining a plurality of samples, where the samples include historical execution data and corresponding delivery tasks;

[0044] Training the AI model with the plurality of samples to obtain a trained AI model;

[0045] Among them, the step of constructing an AI model through a deep reinforcement learning algorithm includes:

[0046] On the basis of historical execution data, collecting various index data of the internal delivery tasks in the hospital;

[0047] Determining the state space and the action space, where the state space defines the state of the robot at each moment, and the action space defines the actions available to the robot in each state;

[0048] Designing a reward function to motivate the robot to learn the optimal delivery strategy;

[0049] Selecting the Q-learning reinforcement learning algorithm, and training the model with historical execution data and the designed reward function to optimize the model parameters so that it can select the optimal action according to the current state.

[0050] To achieve the above object, the second aspect of the present invention provides a path optimization system for an automatic delivery robot in a hospital based on deep learning, including the following components:

[0051] Terrain and obstacle information acquisition module: used to acquire the terrain and obstacle information of all target areas and collect the delivery demand data of each target area;

[0052] Layout and simulation running module: performing layout in the map model according to the delivery demand data and the terrain and obstacle information of all target areas, and performing simulation running on the map model to simulate the driving route of the delivery robot;

[0053] Data comparison and priority acquisition module: Compare the simulated driving data of all delivery robots obtained during the simulation run with the delivery demand data. If there is simulated driving data of a delivery robot that meets the preset conditions, obtain the priority of this delivery task.

[0054] Task matching and path scheduling module: According to the obtained priority of the delivery task, perform matching in the preset task library to determine the best delivery path, all path influencing factors and their preset weights, and determine the scheduling priority of the delivery task accordingly.

[0055] AI optimization model module: Input the delivery task demand data and priority that meet the preset conditions into the trained AI model to update the weight value of the path influencing factor.

[0056] Path adjustment and optimization module: According to all path influencing factors of the delivery task and the updated weight value, adjust and optimize the best delivery path of the corresponding task to determine the optimal delivery path.

[0057] (III) Beneficial effects

[0058] Compared with the prior art, a method and system for optimizing the path of an automatic delivery robot in a hospital based on deep learning provided by the present invention optimize the delivery path of the automatic delivery robot in the hospital by introducing a deep learning algorithm. First, by scanning the hospital environment in real time and using image recognition technology to update the map information, the accuracy of the delivery path planning is ensured. Then, by collecting the driving data of the delivery robot through simulation operation and comparing it with the delivery demand data, the priority of each delivery task is determined. Then, according to the task priority and historical execution data, the best path and scheduling priority of the delivery task are calculated using the trained AI model. Finally, the best delivery path is adjusted according to the latest weight value of the path influencing factor to ensure the high efficiency and high accuracy of the delivery process. Through this method, the present invention can effectively cope with the complex terrain and obstacle conditions in the hospital, as well as various sudden delivery demands, and solve the problems existing in the traditional method. Brief description of the drawings

[0059] Figure 1 It is a flowchart of a method for optimizing the path of an automatic delivery robot in a hospital based on deep learning disclosed in this application.

[0060] Figure 2 It is a schematic diagram of path optimization disclosed in this application.

[0061] Figure 3 It is a flowchart of simulation operation and data comparison disclosed in this application.

[0062] Figure 4 It is a schematic diagram of AI model creation disclosed in this application.

[0063] Figure 5 Schematic diagram of the framework structure of a path optimization system for hospital automatic delivery robots based on deep learning disclosed in this application.

[0064] Reference numerals shown in the figure: 100, terrain obstacle information acquisition module; 200, layout and simulation operation module; 300, data comparison and priority acquisition module; 400, task matching and path scheduling module; 500, AI optimization model module; 600, path adjustment and optimization module. Specific implementation manners

[0065] The present invention will be described in detail below with reference to the accompanying drawings. The technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0066] As Figure 1 shown, in the first aspect of this embodiment, a path optimization method for hospital automatic delivery robots based on deep learning is provided, and the method includes the following steps:

[0067] Step S100: Obtain the terrain and obstacle information of all target areas, as well as the delivery demand data of each target area.

[0068] First, use a high-precision camera to comprehensively scan the internal and external environments of the hospital, collect the terrain and obstacle information of the target areas, and update these data to the map model in real time through image recognition technology. At the same time, summarize the delivery demand data of each target area, including but not limited to information such as the types, quantities of items to be delivered, and expected delivery times. This process ensures the accuracy and timeliness of the basic data for delivery task planning.

[0069] Step S200: Perform layout in the map model according to the delivery demand data and the terrain obstacle information of all target areas, and perform simulation operation on the map model.

[0070] According to the data obtained in step S100, construct a map model in a computer system that reflects the real hospital environment, and this map model details the positions of various obstacles and the distribution of target areas. Then, perform simulation operation on this map model to simulate the moving path of the delivery robot in the hospital environment to evaluate the feasibility of different delivery plans.

[0071] Step S300: During the simulation operation, obtain the simulated driving data of all the delivery robots.

[0072] During the simulation running phase, the system records the simulated driving data of the delivery robot, such as driving time, route efficiency, etc., providing practical operation data support for subsequent path optimization.

[0073] Step S400: Compare the simulated driving data with the delivery demand data. If there is simulated driving data of the delivery robot that meets the preset conditions, obtain the priority of this delivery task.

[0074] As Figure 3 shown, by comparing the simulated driving data with the delivery demand data, the execution efficiency of each delivery task is evaluated. If the simulation results show that the actual execution indicators (such as delivery time) of a delivery task are lower than or equal to the expected indicators, this task will be assigned a certain priority, and the determination of the priority is based on the preset conditions.

[0075] Step S500: According to the priority of the delivery task, perform a match in the preset task library to obtain the best delivery path corresponding to this task, all path impact factors, and the preset proportion of each path impact factor, and determine the scheduling priority of this delivery task.

[0076] According to the priority of the delivery task determined in the previous step, the system retrieves and matches the best delivery path of each task from the preset task library, and at the same time obtains the relevant path impact factors and their preset proportions. In addition, a scheduling priority will also be determined for each task based on the priority, ensuring that urgent or high-priority tasks can be executed first.

[0077] Step S600: According to the scheduling priority, sequentially input the demand data and priority of the delivery task that meet the preset conditions into the trained AI model to obtain the latest proportion value.

[0078] Input the demand data and determined priority of the delivery task into the trained deep learning AI model. The AI model will recalculate the latest proportion value of each path impact factor based on the current data to adapt to the real-time changing delivery environment and demand.

[0079] Step S700: Adjust the best delivery path corresponding to this task according to all path impact factors of the delivery task and the latest proportion value of each path impact factor to obtain the optimal path.

[0080] Finally, according to the latest proportion value output by the AI model, readjust the best delivery path of each delivery task to determine the optimal path. This step takes into account the latest situations of all relevant path impact factors, ensuring that the delivery tasks can be completed with the highest efficiency and accuracy.

[0081] Through the above steps, this embodiment effectively combines deep learning technology and automatic control, achieving intelligent optimization of the path of hospital automatic delivery robots. This method not only improves the delivery efficiency and accuracy but also has strong adaptability and flexibility, capable of quickly responding to changes in the hospital internal environment and updates in delivery requirements, providing an efficient and intelligent solution for hospital material delivery.

[0082] In the further refined step S100, the delivery requirement data includes the quantity of actual delivery items and the estimated delivery time. These data clarify the specific requirements of each delivery task, providing a basis for efficient delivery. The simulated driving data includes the simulated delivery time and the simulated delivery efficiency, which are obtained through simulated operation and reflect the potential performance of the delivery path. When the simulated delivery time of the delivery robot is lower than the estimated delivery time, or the simulated delivery efficiency is higher than the preset efficiency, the system will automatically obtain the priorities of all delivery tasks, optimize resource allocation, and ensure that high-priority tasks are quickly responded to.

[0083] As Figure 2 shown, in step S200, the creation of the map model is a key step in optimizing the delivery path. First, collect the map data inside the hospital, including building floor plans or three-dimensional structure data. Then, based on these data, establish the data structure of the map and mark the obstacles inside the hospital in it. Associate the delivery requirement data with the data structure of the map and record it in the map model, laying a foundation for the calculation of the delivery path. Finally, according to the map structure and the delivery requirement data, the system calculates the best path for the delivery robot and integrates this map model into the hospital automatic delivery system to achieve intelligent optimization of the delivery path.

[0084] In the specific steps of calculating the best path, first define the map model G(V,E), where V represents the set of nodes and E represents the set of edges. Each node v∈V represents a specific location inside the hospital, and each edge e∈E represents a feasible path between two locations, and its weight can be the length or cost of the path. By setting the starting position and using the priority queue Q to sort by distance, visit and update the distances between nodes in turn until the target position is found. Through this method, the shortest path from the starting position to the target position can be accurately calculated and reconstructed, providing the best navigation path for the delivery robot.

[0085] Before determining the priority of delivery tasks and making matches, that is, before step S500, this embodiment further includes obtaining the historical execution data of all delivery tasks. This data includes the best delivery routes, route impact factors, and their actual proportions. By analyzing all the historical execution data of the same delivery task, the system can count the best delivery route with the highest usage frequency and the most common route impact factors, and determine their preset proportions. In this way, the best delivery route, route impact factors, and their preset proportions of the same delivery task can be bound and saved in the preset task library, providing a reference and optimization basis for future delivery tasks.

[0086] In the process of optimizing the path of the hospital's automatic delivery robot, this embodiment further takes into detailed consideration and handling of unexpected situations that the robot may encounter during the task execution. Specifically, during the simulation run and actual execution, the status of the delivery robot is monitored in real time, such as the battery power and whether there are unforeseen obstacles in its environment. When the battery power of the delivery robot is lower than the set threshold or it encounters an unforeseen obstacle, this system will automatically trigger the deep learning algorithm to re-plan the path. This technology ensures that the delivery robot can complete the delivery task efficiently and safely even in the face of unexpected situations.

[0087] To achieve a more precise and flexible delivery task scheduling, this embodiment also includes the step of constructing and training an AI model through a deep reinforcement learning algorithm, as shown in Figure 4 . The following is a detailed description of this process:

[0088] Constructing the AI model: First, based on the historical execution data, various index data of the internal delivery tasks in the hospital are collected. This data covers multiple aspects of the delivery tasks, providing rich input information for constructing the AI model. Then, the state space and action space of the model are determined. The state space defines the state of the robot at each moment, including position, power, etc.; the action space defines the optional actions of the robot in each state, such as moving forward, turning, stopping, etc.

[0089] Designing the reward function: The design of the reward function aims to encourage the robot to learn the optimal delivery strategy. The reward function is usually designed based on factors such as delivery efficiency and punctuality to ensure that the robot tends to choose those actions that are more efficient and more in line with the delivery requirements during the learning process.

[0090] Model training: Select a reinforcement learning algorithm such as Q-learning to train the AI model. Using the collected historical execution data and the designed reward function, continuously optimize the model parameters. Through training, the AI model can select the optimal action according to the current state, thus realizing the intelligent optimization and adjustment of the delivery robot's path. Taking the Q-learning algorithm as an example, its update formula is as follows:

[0091]

[0092] The meanings of the terms in the formula are as follows:

[0093] : represents the Q-value of executing action a in state s;

[0094] : the learning rate, used to control the influence degree of new information on the Q-value;

[0095] : the immediate reward obtained after executing action a;

[0096] : the discount factor, used to control the importance degree of future rewards;

[0097] : the next state transferred to after executing action a;

[0098] : in the next state select the action with the highest Q-value among all the available actions;

[0099] It can be understood that the Q-value is updated with a new estimated value, and the new estimated value is the weighted sum of the current Q-value and an estimated future benefit. This future benefit is the immediate reward obtained from the current action plus the discounted value (multiplied by γ) of the maximum Q-value in the next state after executing action a. In this way, the Q-learning algorithm learns the optimal action strategy by continuously updating the Q-value in the state-action space.

[0100] Obtaining samples and training: Obtain a large number of samples, including historical execution data and corresponding distribution tasks. These samples are processed and used to train the AI model. Through repeated iterative learning, the AI model gradually improves the accuracy and efficiency of selecting the optimal path for different distribution tasks.

[0101] Through the above process, the constructed and trained AI model can perform more accurate priority sorting and path planning for distribution tasks, enabling the hospital's automatic distribution system to have higher adaptability and efficiency during the execution process. This method based on deep reinforcement learning not only improves the execution quality of distribution tasks but also increases the system's ability to handle complex environments and emergencies.

[0102] In this embodiment, it is necessary to further clarify the synchronous update mechanism of the simulation operation and the real environment response strategy, which specifically includes the following steps:

[0103] Real-time environmental adaptability adjustment: Establish an environmental monitoring module responsible for collecting real-time environmental changes inside and outside the hospital, including but not limited to the emergence of new obstacles, the disappearance or movement of known obstacles, and changes in the environmental layout. When environmental changes are detected, immediately update the obstacle information and terrain data in the map model, and transmit these updates to all delivery robots and the central control system. After receiving the latest map data, the delivery robots recalculate the optimal delivery path using their current location and the destination.

[0104] Dynamic adjustment and learning mechanism for delivery tasks: Monitor the status of delivery robots in real time, including battery power, weight of the transported items, estimated arrival time, etc. When abnormal situations are detected, such as the battery power being below the safety threshold or an estimated late arrival, automatically trigger an emergency response program. This program can re-evaluate the priority of the current delivery task and consider whether it is necessary to reassign the task to other standby robots or adjust the delivery order. Feed back the actual execution data (including delivery path, time taken, energy consumed, etc.) after each delivery task is completed to the AI model for continuous learning and optimization of future delivery strategies.

[0105] Continuous iteration of the deep reinforcement learning model: Set up a regular model evaluation mechanism to evaluate the model performance by comparing the predicted delivery efficiency and the actual delivery efficiency of the model. When the model performance is below the set threshold, or when there are significant environmental changes and task requirement changes, trigger the model retraining process to update the model parameters using the latest historical execution data and environmental data. Introduce the opportunity of external expert knowledge and manual adjustment of parameters to assist the model in quickly adapting to the complex and changing delivery environment and requirements.

[0106] Suppose in a large hospital environment, there are multiple delivery tasks to be completed, including the distribution of drugs to various wards, the transportation of medical devices to the operating room, and the delivery of experimental materials to the laboratory. Use the above deep learning-based path optimization method to ensure that these delivery tasks can be completed efficiently and safely.

[0107] Step S100: Information acquisition

[0108] Use a high-precision camera to comprehensively scan the internal and external environments of the hospital, paying special attention to the entrances and passages of key areas such as wards, operating rooms, and laboratories, as well as possible obstacles (such as temporary construction areas, newly installed equipment, etc.).

[0109] Collect the drug delivery list from the pharmacy to the wards, including the types and quantities of drugs; collect the instrument delivery requirements from the medical device storage room to the operating room; obtain the experimental material requirement list from the supply department to the laboratory, also specifying the types and quantities of each item of supplies.

[0110] Step S200: Simulation run

[0111] Based on the information collected in step S100, a three-dimensional map model reflecting the real hospital environment is constructed in the computer system, with all key areas and obstacle positions marked in detail.

[0112] Simulate the movement path of the delivery robot on this map model to evaluate the feasibility of the delivery plans from the pharmacy to each ward, from the medical equipment storage room to the operating room, and from the supply department to the laboratory.

[0113] Step S300: Obtain simulated driving data

[0114] Record data such as the estimated time and route efficiency of each delivery task during the simulated driving.

[0115] Step S400: Compare and obtain priorities

[0116] Compare the simulated driving data with the actual delivery requirements to determine that tasks such as delivering emergency drugs to the ICU (Intensive Care Unit) have the highest priority.

[0117] Step S500: Match the preset task library

[0118] Match the best path for delivering drugs to the ICU in the preset task library, and obtain the relevant influencing factors and their proportions. At the same time, determine the scheduling priority for this task.

[0119] Step S600: Input to the AI model and update the proportions

[0120] Input the demand data and priority of the drug delivery task into the trained AI model to obtain the latest proportion values of the path influencing factors adjusted for the current environment and requirements.

[0121] Step S700: Adjust and optimize the path

[0122] According to the latest proportion values output by the AI model, readjust the path for delivering drugs to the ICU to ensure the selection of the optimal path.

[0123] Specific example:

[0124] In the morning of a day, the hospital system received three main delivery tasks: quickly delivering urgently needed cardiac surgery drugs from the pharmacy to the ICU, transporting the scheduled surgical instruments from the medical equipment storage room to Operating Room No. 3, and sending the newly arrived PCR reagents from the supply department to the biological laboratory.

[0125] Through the execution of steps S100 to S700, the system first confirmed that the task priority of delivering heart surgery drugs to the ICU is the highest. Then, through simulation and analysis of the AI model, the best path from the pharmacy to the ICU was optimized, considering avoiding the ongoing corridor maintenance area and using elevators instead of stairs to save time. At the same time, the system also planned the corresponding optimal delivery paths for the other two tasks to ensure that all tasks can be completed efficiently.

[0126] In this way, the hospital's automatic delivery robot system not only successfully completed the emergency delivery of heart surgery drugs in the shortest time but also ensured the smooth progress of other delivery tasks, significantly improving the overall efficiency and accuracy of hospital material delivery.

[0127] As Figure 5 shown, the second aspect of this embodiment provides a method for optimizing the path of an automatic delivery robot in a hospital based on deep learning. The system includes the following modules:

[0128] Terrain obstacle information acquisition module 100: used to acquire the terrain and obstacle information of all target areas and collect the delivery demand data of each target area;

[0129] Layout and simulation operation module 200: According to the delivery demand data and the terrain obstacle information of all target areas, perform layout in the map model and execute simulation operation on the map model to simulate the driving route of the delivery robot;

[0130] Data comparison and priority acquisition module 300: Compare all the simulated driving data of the delivery robot obtained during the simulation operation with the delivery demand data. If there is simulated driving data of the delivery robot that meets the preset conditions, obtain the priority of this delivery task;

[0131] Task matching and path scheduling module 400: According to the obtained delivery task priority, perform matching in the preset task library to determine the best delivery path, all path influencing factors and their preset proportions, and accordingly determine the scheduling priority of the delivery task;

[0132] AI optimization model module 500: Input the delivery task demand data and priority that meet the preset conditions into the trained AI model to update the proportion value of the path influencing factor;

[0133] Path adjustment and optimization module 600: According to all the path influencing factors of the delivery task and the updated proportion value, adjust and optimize the best delivery path of the corresponding task to determine the optimal delivery path.

[0134] The above embodiments are only used to illustrate the technical solutions of the present application and not to limit them. Although the present application has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A hospital automatic delivery robot path optimization method based on deep learning, characterized in that: The steps include: Obtain terrain and obstacle information for all target areas, as well as distribution demand data for each target area; Layout is performed in a map model according to the distribution demand data and terrain obstacle information of all target areas, and simulation operation is performed on the map model; During the simulation operation, the simulated driving data of all the delivery robots are obtained; The simulated driving data and the delivery demand data are compared, and if the simulated driving data of the delivery robot satisfies a preset condition, the priority of the delivery task is obtained; According to the delivery task priority, matching is performed in the preset task library to obtain the optimal delivery path corresponding to the task, all path influencing factors and the preset proportion of each path influencing factor, and determine the scheduling priority of the delivery task; According to the scheduling priority, the demand data and the optimal scheduling priority of the delivery tasks that meet the preset conditions are input into the trained AI model in turn to obtain the latest weight value; According to all the path influencing factors of the delivery task and the latest weight value of each path influencing factor, the best delivery path corresponding to the task is adjusted to obtain the optimal path; The creation of a map model includes: Collect map data inside the hospital, including floor plans or three-dimensional structure data; Establish the data structure of the map based on the collected map data; Mark obstacles inside the hospital in the map data structure; The distribution demand data of each target area within the hospital is associated with the data structure of the map and recorded in the map model; Calculate the optimal path for the delivery robot based on the map structure and delivery demand data; Integrate the created map model into the hospital's automatic delivery system; The steps to calculate the optimal path for a delivery robot include: Let G(V,E) represent the map model, where V is the node set and E is the edge set; Each node v∈V represents a location inside the hospital; Each edge e∈E represents a path between two locations, and its weight is the length or cost of the path; Mark the starting position as visited and set the distance from the starting position to 0; Add the starting position to a priority queue Q, sorted by distance from the starting position; Take the node u closest to the starting position from the priority queue Q and mark it as visited; For all neighbor nodes v of node u, calculate the distance from the starting position to node v, and update the distance of node v in the priority queue Q; Repeat the above steps until the priority queue Q is empty or the target location is visited; According to the calculated shortest path information, reversely reconstruct the best path from the starting position to the target position; After the optimal delivery path, path influencing factors, and preset weights of the path influencing factors for the same delivery task are bound, the steps of saving them to the preset task library include: According to the same optimal delivery path, the scheduling priority of each path influencing factor is configured from high to low according to the occurrence frequency of each path influencing factor; According to the scheduling priority, the optimal delivery path, path influencing factors and the preset proportion of each path influencing factor for the same delivery task are saved in the preset task library.

2. According to the deep learning-based hospital automatic delivery robot path optimization method of claim 1, it is characterized in that: The process of collecting terrain and obstacle information and distribution demand data includes using a camera to perform real-time scanning of the environment inside and outside the hospital, and automatically updating the obstacle information in the map model through image recognition technology.

3. According to the deep learning-based hospital automatic delivery robot path optimization method of claim 1, it is characterized in that: The delivery demand data includes the actual delivery quantity and the estimated delivery time, and the simulated driving data includes the simulated delivery time and the simulated delivery efficiency. The simulated driving data and the delivery demand data are compared. If the simulated delivery time of the delivery robot is lower than the estimated delivery time, or the simulated delivery efficiency of the delivery robot is higher than the preset efficiency, the scheduling priority of all delivery tasks is obtained.

4. According to the deep learning-based hospital automatic delivery robot path optimization method of claim 1, it is characterized in that: Before the step of matching in the preset task library according to the delivery task priority, the method further includes: Obtaining historical execution data of all delivery tasks, any of which includes an optimal delivery path, a path influencing factor, and an actual weight of each path influencing factor; According to all historical execution data of the same delivery task, the best delivery path with the highest number of uses and a preset number of path influencing factors with the highest frequency of occurrence are counted, and the preset weight of each of the path influencing factors is determined; The optimal delivery path, path influencing factors and preset weights of each path influencing factor for the same delivery task are bound and saved in the preset task library.

5. The method for optimizing the path of a hospital automatic delivery robot based on deep learning according to claim 1, characterized in that: The step of obtaining the terrain and obstacle information of all target areas and the delivery demand data of each target area includes: real-time detection of the simulation operation process, and when the battery power of any delivery robot is lower than a threshold or encounters an unforeseen obstacle, the path is replanned through a deep learning algorithm.

6. The method for optimizing the path of a hospital automatic delivery robot based on deep learning according to claim 1, characterized in that: Before the step of inputting the demand data and priority of the delivery tasks that meet the preset conditions into the trained AI model in turn according to the scheduling priority to obtain the latest weight value, the method further includes: Build AI models through deep reinforcement learning algorithms; Acquire multiple samples, the samples including historical execution data and corresponding delivery tasks; Training the AI ​​model using the multiple samples to obtain a trained AI model; The steps to build an AI model through deep reinforcement learning algorithms include: Based on historical execution data, collect various indicator data of internal hospital distribution tasks; Determine the state space and action space, where the state space defines the state of the robot at each moment, and the action space defines the optional actions of the robot in each state; Design a reward function to motivate the robot to learn the optimal delivery strategy; Select the Q-learning reinforcement learning algorithm, and use historical execution data and the designed reward function to train the model and optimize the model parameters so that it can select the optimal action based on the current state.

7. A hospital automatic delivery robot path optimization system based on deep learning, characterized in that: The optimization system includes the following modules: Terrain obstacle information acquisition module: used to obtain terrain and obstacle information of all target areas, and collect distribution demand data of each target area; Layout and simulation operation module: layout is performed in the map model according to the delivery demand data and the terrain obstacle information of all target areas, and simulation operation is performed on the map model to simulate the driving route of the delivery robot; Data comparison and priority acquisition module: compares the simulated driving data of all delivery robots obtained during the simulation process with the delivery demand data. If there is simulated driving data of the delivery robot that meets the preset conditions, the priority of the delivery task is obtained; Task matching and route scheduling module: According to the acquired delivery task priority, it matches in the preset task library to determine the best delivery route, all route influencing factors and their preset weights, and accordingly determine the scheduling priority of the delivery task; AI optimization model module: inputs the delivery task demand data and scheduling priorities that meet the preset conditions into the trained AI model to update the weight value of the path influencing factor; Path adjustment and optimization module: According to all the path influencing factors of the delivery task and the updated weight values, the best delivery path of the corresponding task is adjusted and optimized to determine the optimal delivery path; The creation of a map model includes: Collect map data inside the hospital, including floor plans or three-dimensional structure data; Establish the data structure of the map based on the collected map data; Mark obstacles inside the hospital in the map data structure; The distribution demand data of each target area within the hospital is associated with the data structure of the map and recorded in the map model; Calculate the optimal path for the delivery robot based on the map structure and delivery demand data; Integrate the created map model into the hospital's automatic delivery system; The steps to calculate the optimal path for a delivery robot include: Let G(V,E) represent the map model, where V is the node set and E is the edge set; Each node v∈V represents a location inside the hospital; Each edge e∈E represents a path between two locations, and its weight is the length or cost of the path; Mark the starting position as visited and set the distance from the starting position to 0; Add the starting position to a priority queue Q, sorted by distance from the starting position; Take the node u closest to the starting position from the priority queue Q and mark it as visited; For all neighbor nodes v of node u, calculate the distance from the starting position to node v, and update the distance of node v in the priority queue Q; Repeat the above steps until the priority queue Q is empty or the target location is visited; According to the calculated shortest path information, reversely reconstruct the best path from the starting position to the target position; After the optimal delivery path, path influencing factors, and preset weights of the path influencing factors for the same delivery task are bound, the steps of saving them to the preset task library include: According to the same optimal delivery path, the scheduling priority of each path influencing factor is configured from high to low according to the occurrence frequency of each path influencing factor; According to the scheduling priority, the optimal delivery path, path influencing factors and the preset proportion of each path influencing factor for the same delivery task are saved in the preset task library.

Images