A control method and system based on a medical transport robot

Abstract

This application relates to the field of medical transport equipment, specifically a control method and system for medical transport robots. It employs a pre-planned global path and a real-time obstacle avoidance strategy to control the robot. The global path planning, based on a 3D grid map, comprehensively considers multiple costs such as distance, congestion, and elevator waiting time to ensure overall optimal task execution. Real-time dynamic perception, based on the fusion of depth vision and radar, accurately identifies target types and motion vectors, enabling various intelligent obstacle avoidance behaviors such as detours, waiting, and proactive obstacle avoidance. This balances traffic efficiency and path continuity, making it particularly suitable for the complex and dynamic scenarios of hospitals with high population density and frequent emergencies. The hierarchical decision-making mechanism avoids the heavy overhead of global planning while optimizing obstacle avoidance paths through a local cost model, enabling the robot to complete medical transport tasks politely, efficiently, and reliably, significantly improving the automation level and emergency response capabilities of hospital logistics.

Description

Technical Field

[0001] This application relates to the field of medical transport equipment, specifically a control method and system based on a medical transport robot. Background Technology

[0002] Medical transport robots are automated transport devices specifically designed for medical environments, primarily used to safely and efficiently transport various items within hospitals, laboratories, and other medical institutions. They integrate technologies such as navigation, obstacle avoidance, the Internet of Things (IoT), and human-computer interaction, making them a crucial component of smart hospitals and medical automation.

[0003] The control logic of medical transport robots differs significantly from that of existing autonomous driving robots. Because medical transport robots operate within hospitals, they frequently encounter scenarios requiring proactive obstacle avoidance. However, current medical transport robots typically employ general obstacle avoidance methods, failing to perform proactive avoidance maneuvers for specific types of obstacles (such as hospital beds and stretchers). Even when an emergency bed is identified, most algorithms simply treat it as a large obstacle, performing a routine stop or detour, rather than the systematic proactive avoidance required by hospital etiquette. The robot may even remain stopped in the middle of the path, still creating an obstruction. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a control method and system based on a medical transport robot to solve the problems in the background art.

[0005] To achieve the above objectives, this application adopts the following technical solution:

[0006] This application discloses a control method based on a medical transport robot, comprising the following steps:

[0007] When a transportation task is received from an external source, preliminary path planning is performed based on the transportation task, a pre-built decision model, and a pre-built 3D gridded map to obtain the target path. The transportation task includes a starting point and a destination, and the decision model includes a distance cost function, a congestion cost function, an elevator waiting cost function, a static rule cost function, and a planner.

[0008] Control the medical transport vehicle to travel along the target path, and collect RGB images, depth images and radar point cloud data of the environment during the journey;

[0009] Dynamic target recognition is performed based on the RGB image, the depth image, and the radar point cloud data to obtain multiple dynamic targets in the environment where the medical transport vehicle is located, the types of multiple dynamic targets, and the motion vectors of multiple dynamic targets.

[0010] Based on the motion vectors of multiple dynamic targets, the types of multiple dynamic targets, and the 3D mesh map, the feasible domain of the medical transport robot at a future time point is selected, and obstacle avoidance decisions are executed based on the feasible domain and the pre-built obstacle avoidance decision model. The obstacle avoidance decisions include detour, waiting in place, and active avoidance. When the obstacle avoidance decision is detour or active avoidance, the obstacle avoidance path is based on the safety cost function, the deviation cost function of the target path, the efficiency cost function, and the planner.

[0011] Upon completion of obstacle avoidance, return to the target path and retrieve the RGB image, depth image, and radar point cloud data of the acquired environment until the destination is reached.

[0012] In one embodiment of this application, the mathematical expression of the decision model is:

[0013]

[0014]

[0015]

[0016]

[0017]

[0018] In the formula, Indicates the total cost. Indicates distance cost. The weights represent the distance cost. This indicates the cost of congestion. The weights representing the costs of congestion This indicates the cost of waiting for the elevator. This represents the weight of the elevator waiting cost. Represents the cost of static rules. The weights are the costs of static rules. This represents the node index in a 3D gridded map. This indicates the total number of nodes in the path. Indicates the distance between adjacent nodes. Indicates the reference maximum distance. Represents a node Reference congestion coefficient, Indicates passing by elevator Average waiting time This indicates the maximum reference waiting time. To represent nodes Is it located in a restricted area? Represents a node The penalty coefficient.

[0019] In one embodiment of this application, dynamic target recognition is performed based on the RGB image, the depth image, and the radar point cloud data to obtain multiple dynamic targets in the environment where the medical transport vehicle is located, the types of the multiple dynamic targets, and the motion vectors of the multiple dynamic targets, including:

[0020] The RGB image is input into a pre-built recognition model to obtain detection boxes and recognition labels for multiple targets; and the radar point cloud data is clustered to obtain multiple point cloud clusters.

[0021] The detection bounding boxes and recognition labels of multiple targets in the RGB image are transferred to the depth image, wherein the RGB image and the depth image are pre-aligned using camera internal parameters;

[0022] Specifically, the detection boxes and labels in the RGB image are converted to the depth image coordinate system through camera calibration parameters and then associated with the depth point cloud data; after clustering the radar point cloud, the point cloud clusters are matched with the depth point cloud clusters to obtain the 3D position of the labeled dynamic target.

[0023] The depth image and the radar point cloud data are preprocessed to obtain preprocessed image data and preprocessed point cloud data, wherein the preprocessing includes spatiotemporal alignment, filtering and coordinate unification;

[0024] Based on the camera's internal parameters, each pixel in the preprocessed image data is converted into a 3D point to obtain depth point cloud data; and the depth point cloud data is clustered to obtain multiple depth point cloud clusters.

[0025] Based on the principle of planar projection, the detection boxes and recognition labels in the depth image are transmitted to multiple depth point cloud clusters; the depth point cloud clusters are matched with the point cloud clusters to obtain multiple data groups with matching positions;

[0026] For any first data group, the detection bounding boxes and recognition labels of the depth point cloud clusters are passed to the point cloud clusters to obtain the detection bounding boxes and recognition labels of multiple point cloud clusters.

[0027] Tracking the center coordinates of each point cloud cluster in multi-frame radar point cloud data And based on the center coordinates of multiple frames of radar point cloud data Constructing motion vectors for multiple dynamic targets ,in, The identifier code for a point cloud cluster.

[0028] In one embodiment of this application, the feasible domain of the medical transport robot at a future time point is selected based on the motion vectors of multiple dynamic targets, the types of multiple dynamic targets, and the 3D meshed map, including:

[0029] Based on the motion vector of each dynamic target and the uniform velocity model, the time point of each dynamic target is calculated. Location Among them, dynamic targets are at a certain point in time. Location The mathematical expression is:

[0030]

[0031] In the formula, Represents dynamic objectives At the present moment Location, Indicates uncertainty;

[0032] The expansion radius is determined based on the type of the dynamic target, and based on the dynamic target at a given time point. Location And expanding the radius to construct dynamic targets at future time points Basic danger domain ;

[0033] Based on the motion vector For the basic hazard domain Targeted expansion is performed to obtain the final danger zone. The final danger zone The mathematical expression is:

[0034]

[0035] In the formula, Minkowski addition is represented. Indicates the expansion factor. Indicates the time step;

[0036] Based on the final danger zone Filtering multiple future time points in the 3D gridded map The feasible domain.

[0037] In one embodiment of this application, based on the final danger zone Filtering multiple future time points in the 3D gridded map The feasible domain includes:

[0038] Based on the final danger zone The 3D meshed map is annotated to obtain hazardous grids and free grids, wherein the hazardous grids are the areas ultimately deemed hazardous. The grid occupied;

[0039] Based on the starting point of the medical transport robot, a connectivity analysis is performed on the free grid to obtain a feasible grid; and a feasible region is constructed based on the feasible grid.

[0040] In one embodiment of this application, performing obstacle avoidance decision based on the feasible region includes:

[0041] The feasible region is divided into multiple connected components, and the minimum channel width of the connected component where the medical transport robot is located is calculated.

[0042] The minimum channel width is compared with a preset width threshold. If the minimum channel width is greater than or equal to the preset width threshold, it is determined that there is a drivable environment at the current moment; otherwise, it is determined that there is no drivable environment at the current moment; and dynamic targets within the target distance range in front of the medical transport robot are acquired.

[0043] If the medical transport robot does not have a drivable environment at the current moment, it will remain in place and wait until it can drivable.

[0044] When there is a dynamic target that meets the target conditions within the target distance range in front of the medical transport robot, and there is a driving environment, active avoidance planning is performed based on the pre-built obstacle avoidance decision model to obtain an active avoidance route, and the medical transport robot is controlled to travel along the active avoidance route. The target conditions include a type label indicating the target type, which will collide with the medical transport robot in the future target time period.

[0045] When a non-target dynamic target exists within the target distance range ahead of the medical transport robot, and a driving environment exists, detour planning is performed based on a pre-built obstacle avoidance decision model to obtain a detour route, and the medical transport robot is controlled to travel along the detour route.

[0046] When there are no dynamic targets within the target distance range ahead of the medical transport robot, and a driving environment exists, the medical transport robot is controlled to continue driving along the target path.

[0047] In one embodiment of this application, the active obstacle avoidance route includes a first obstacle avoidance route and a second obstacle avoidance route, wherein active obstacle avoidance planning is performed based on a pre-built obstacle avoidance decision model to obtain the active obstacle avoidance route, including:

[0048] Obtain the motion vector of a dynamic target of the target type, and perform linear prediction based on the motion vector of the dynamic target of the target type to obtain the predicted route;

[0049] Candidate grids within the target area surrounding the predicted route and located within the feasible region are selected. ;

[0050] Calculate the avoidance evaluation score of multiple candidate grids. The mathematical expression for the avoidance evaluation score is:

[0051]

[0052]

[0053] In the formula, As the first weight, As the second weight, As the third weight, For normalization function, Representation and grid Distance to the nearest dynamic target This indicates that the medical transport robot has moved to the grid. distance, To characterize the mesh Shortest distance from the predicted route Is it greater than the distance threshold? The flag bit;

[0054] Avoidance evaluation score The highest grid is the best avoidance grid. ;

[0055] Starting from the current position of the medical transport robot, the optimal avoidance grid... With the destination as the starting point, path planning is performed using a pre-built obstacle avoidance decision model to obtain the first avoidance route;

[0056] When a dynamic target of the target type leaves, the second avoidance route is obtained by using the optimal avoidance grid as the starting point and any point in the target path as the ending point, combined with the pre-built obstacle avoidance decision model.

[0057] In one embodiment of this application, detour planning is performed based on a pre-built obstacle avoidance decision model to obtain a detour route, including:

[0058] Starting from the current position of the medical transport robot and taking any point after the starting point of the target route as the endpoint, a detour route is obtained by combining the pre-built obstacle avoidance decision model for path planning.

[0059] In one embodiment of this application, the mathematical expression of the obstacle avoidance decision model is:

[0060]

[0061]

[0062]

[0063]

[0064] In the formula, Indicates the cost of security. The weight representing the security cost, Indicates the offset cost. The weights representing the offset cost. Indicates the cost of efficiency. The weights representing the efficiency costs Represents a node The minimum distance to any dynamic target. Represents a node The minimum distance to the target path. Indicates the cumulative path length. Indicates the cumulative curvature of the path. This is the normalization function.

[0065] This application also provides a control system based on a medical transport robot, including:

[0066] The route planning module is used to perform preliminary route planning based on the transportation task, a pre-built decision model, and a pre-built 3D gridded map when a transportation task is received from an external source, so as to obtain the target route. The transportation task includes a starting point and an ending point, and the decision model includes a distance cost function, a congestion cost function, an elevator waiting cost function, a static rule cost function, and a planner.

[0067] The travel and environment perception module is used to control the medical transport vehicle to travel along the target path and to collect RGB images, depth images and radar point cloud data of the environment during the travel process;

[0068] The recognition module is used to perform dynamic target recognition based on the RGB image, the depth image and the radar point cloud data, to obtain multiple dynamic targets in the environment where the medical transport vehicle is located, the types of multiple dynamic targets and the motion vectors of multiple dynamic targets;

[0069] The obstacle avoidance module is used to filter the feasible domain of the medical transport robot at a future time point based on the motion vectors of multiple dynamic targets, the types of multiple dynamic targets, and the 3D mesh map, and to execute obstacle avoidance decisions based on the feasible domain and a pre-built obstacle avoidance decision model. The obstacle avoidance decisions include detour, waiting in place, and active avoidance. When the obstacle avoidance decision is detour or active avoidance, the obstacle avoidance path is based on a safety cost function, a deviation cost function of the target path, an efficiency cost function, and a planner.

[0070] The loop control module is used to return to the target path and the travel and environment perception module when obstacle avoidance is completed, until the destination is reached.

[0071] The beneficial effects of this application are as follows: This application provides a control method and system for medical transport robots, employing a combination of pre-planned global path design and real-time obstacle avoidance strategies to control the medical transport robot. Global path planning, based on a 3D mesh map, comprehensively considers multiple costs such as distance, congestion, and elevator waiting time to ensure overall optimal task execution. Real-time dynamic perception, based on the fusion of depth vision and radar, accurately identifies target types and motion vectors, enabling various intelligent obstacle avoidance behaviors such as detours, waiting, and active obstacle avoidance. This architecture effectively balances traffic efficiency and path continuity while ensuring zero-collision transport safety, making it particularly suitable for complex and dynamic scenarios in hospital environments with high population density and frequent emergencies. The hierarchical decision-making mechanism avoids the heavy overhead of global planning and optimizes obstacle avoidance paths through local cost models, enabling the robot to complete medical transport tasks politely, efficiently, and reliably, significantly improving the automation level and emergency response capabilities of hospital logistics. Attached Figure Description

[0072] The present application will be further described below with reference to the accompanying drawings and embodiments:

[0073] Figure 1 This is a logical framework diagram of a control system based on a medical transport robot shown in one embodiment of this application;

[0074] Figure 2 This is a flowchart illustrating a control method based on a medical transport robot in one embodiment of this application;

[0075] Figure 3 This is a structural diagram of a control system based on a medical transport robot, as shown in one embodiment of this application. Detailed Implementation

[0076] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0077] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the layers related to this application and are not drawn according to the actual number, shape and size of the layers in the actual implementation. In the actual implementation, the form, number and proportion of each layer can be arbitrarily changed, and the layer layout may also be more complex.

[0078] Numerous details are explored in the following description to provide a more thorough explanation of embodiments of this application; however, it will be apparent to those skilled in the art that embodiments of this application may be practiced without these specific details.

[0079] Figure 1 This is a logical framework diagram of a control system based on a medical transport robot shown in one embodiment of this application, such as... Figure 1 As shown, this application includes a cloud portion and an edge computing portion. The cloud portion mainly stores a high-precision map of the hospital and a cloud planner, while the edge computing portion includes a perception module, an execution module, and an edge planner.

[0080] When the robot receives any medical supply transportation task, it analyzes the task locally, extracts task information such as the starting point, destination, type of transported goods, and urgency, and sends the task information to the cloud server to perform path planning. The path planning is output by the cloud planner.

[0081] The edge computing component is deployed inside the robot. The execution module receives the planned path and controls the robot's movement. During the movement, the perception module collects depth images, RGB images, and radar point cloud data. The perception information is input into the edge planner for planning, obstacle avoidance decisions are executed, and the obstacle avoidance strategy is input into the execution module for execution.

[0082] Figure 2 This is a flowchart illustrating a control method based on a medical transport robot in one embodiment of this application, as shown below. Figure 2 As shown, a control method based on a medical transport robot in this application mainly includes the following steps:

[0083] S210, when a transportation task is received from an external source, preliminary path planning is performed based on the transportation task, a pre-built decision model, and a pre-built 3D gridded map to obtain the target path. The transportation task includes a starting point and a destination, and the decision model includes a distance cost function, a congestion cost function, an elevator waiting cost function, a static rule cost function, and a planner.

[0084] The planner in this application uses the A* path planning algorithm, and the mathematical expression of the cost function in the decision model is:

[0085] In one embodiment of this application, the mathematical expression of the decision model is:

[0086]

[0087]

[0088]

[0089]

[0090]

[0091] In the formula, Indicates the total cost. Indicates distance cost. The weights represent the distance cost. This indicates the cost of congestion. The weights representing the costs of congestion This indicates the cost of waiting for the elevator. This represents the weight of the elevator waiting cost. Represents the cost of static rules. The weights are the costs of static rules. This represents the node index in a 3D gridded map. This indicates the total number of nodes in the path. Indicates the distance between adjacent nodes. Indicates the reference maximum distance. Represents a node Reference congestion coefficient, Indicates passing by elevator Average waiting time This indicates the maximum reference waiting time. To represent nodes Is it located in a restricted area? Represents a node The penalty coefficient.

[0092] The total cost mentioned above is obtained by weighted summation of distance cost, congestion waiting time, elevator waiting time cost, and static rule cost.

[0093] Among them, distance cost In, divided by the reference maximum distance This makes the cost dimensionless and confined to a reasonable range, far from the cost. Ensure that the robot does not take unnecessary detours when choosing its route, thus meeting the basic requirements for transportation efficiency.

[0094] Congestion coefficient Reflecting nodes The historical congestion level is based on historical traffic data statistics or real-time pedestrian flow monitoring, and the value ranges from [0,1]. Choosing a route that is generally unobstructed, even if a section is slightly longer, is encouraged as it is smoother overall.

[0095] Elevator waiting costs are calculated based on historical data to account for the unique characteristics of vertical transportation in hospitals, specifically the average waiting time at different times. . The maximum tolerable waiting time for hospital elevators is set, such as 300 seconds.

[0096] Static rule cost In the middle, through a large penalty coefficient Ensure the path does not violate important rules, flags. This indicates whether the node is in a restricted or other restricted area. A value of 0 indicates a passable area, while a value of 1 indicates a restricted area (such as an operating room, ICU, or sterile area).

[0097] S220, control the medical transport vehicle to travel along the target path, and collect RGB images, depth images and radar point cloud data of the environment during the journey;

[0098] S230, based on the RGB image, the depth image and the radar point cloud data, perform dynamic target recognition to obtain multiple dynamic targets in the environment where the medical transport vehicle is located, the types of multiple dynamic targets and the motion vectors of multiple dynamic targets;

[0099] The specific process includes:

[0100] S231, the RGB image is input into a pre-built recognition model to obtain detection boxes of multiple targets and recognition labels of multiple targets; and the radar point cloud data is clustered to obtain multiple point cloud clusters;

[0101] Specifically, deep learning models (such as YOLO and Faster R-CNN) are used to process RGB images and identify various targets within them. These models are trained on large amounts of labeled data and can identify target categories (such as medical staff, patients, and hospital beds) and mark the bounding boxes (2D detection boxes) of the targets on the image. RGB images contain color and texture information, and deep learning models can accurately identify target categories, providing semantic labels for subsequent processing.

[0102] For 3D point cloud data acquired by radar, clustering algorithms (such as DBSCAN and Euclidean clustering) are used to group spatially adjacent points into a class, forming point cloud clusters. Each point cloud cluster represents a potential physical object. Radar point cloud data has precise depth and location information, and after clustering, multiple point cloud clusters are obtained, each cluster corresponding to a physical object, providing a foundation for subsequent matching and tracking.

[0103] S232, the detection boxes and recognition labels of multiple targets in the RGB image are transferred to the depth image, wherein the RGB image and the depth image are pre-aligned using camera internal parameters;

[0104] RGB and depth images are typically acquired by the same device (such as an RGB-D camera). Intrinsic and extrinsic parameters are obtained through camera calibration, allowing the two images to be aligned to the same coordinate system. Using this alignment, 2D bounding boxes and identification labels obtained from the RGB image are mapped to their corresponding positions in the depth image. The depth image contains depth information, which, combined with the bounding boxes, allows for the localization of the target's 3D position, laying the foundation for matching with radar point clouds.

[0105] S233, preprocess the depth image and the radar point cloud data to obtain preprocessed image data and preprocessed point cloud data, wherein the preprocessing includes spatiotemporal alignment, filtering and coordinate unification;

[0106] Since the acquisition frequencies and times of depth cameras and radars may differ, time synchronization and spatial registration are required to ensure that the data correspond to the same time and the same coordinate system.

[0107] Filtering can remove noise points (such as outliers caused by reflection, multipath effects, etc.) from depth images and radar point clouds.

[0108] Finally, the depth image and radar point cloud data are converted to the same coordinate system (such as the robot's body coordinate system) to facilitate subsequent fusion processing.

[0109] S234, Based on the camera's internal parameters, each pixel in the preprocessed image data is converted into a 3D point to obtain depth point cloud data; and the depth point cloud data is clustered to obtain multiple depth point cloud clusters;

[0110] Specifically, using camera intrinsic parameters, the depth value of each pixel in the depth image is converted into a 3D point (x, y, z), generating a depth point cloud. The depth point cloud is then clustered (e.g., using Euclidean clustering) to group points belonging to the same object into a single cluster. Depth point clouds provide the precise location and shape of the target in 3D space, containing more geometric information than 2D images. Both depth point clouds and radar point clouds are 3D point clouds, facilitating matching and fusion.

[0111] S235, Based on the principle of planar projection, the detection box and recognition label in the depth image are transmitted to multiple depth point cloud clusters; the depth point cloud clusters are matched with the point cloud clusters to obtain multiple data groups with position matching;

[0112] For each bounding box in the depth image, find the depth point cloud cluster within the box and assign the recognition label of the bounding box to the corresponding depth point cloud cluster.

[0113] Specifically, the similarity between each depth point cloud cluster and the radar point cloud cluster is calculated (e.g., based on centroid distance, bounding box overlap, etc.), and the best-matching radar point cloud cluster is found to form a matching pair between the depth point cloud cluster and the radar point cloud cluster.

[0114] The depth point cloud cluster acquires recognition labels from the RGB image, containing semantic information; the radar point cloud cluster possesses precise 3D position and velocity information. After matching, the two are correlated to obtain a complete target description with both semantic and geometric information. Through multi-sensor data matching, mutual verification and supplementation can be achieved, reducing false detections or missed detections from a single sensor. After matching, the radar point cloud cluster will acquire recognition labels, facilitating subsequent label-based target tracking.

[0115] S236, For any first data group, pass the detection boxes and recognition labels of the depth point cloud clusters to the point cloud clusters to obtain the detection boxes and recognition labels of multiple point cloud clusters;

[0116] S237, tracking the center coordinates of each point cloud cluster in multi-frame radar point cloud data. And based on the center coordinates of multiple frames of radar point cloud data Constructing motion vectors for multiple dynamic targets ,in, The identifier code for a point cloud cluster.

[0117] Specifically, for each frame of radar point cloud data, each tagged point cloud cluster is identified, and the point cloud clusters of the current frame are associated with those of the previous frame using data association algorithms (such as nearest neighbor or Hungarian algorithm) to form a trajectory. For each tracked point cloud cluster, its motion vector (including velocity magnitude and direction) is calculated using the center coordinates of multiple frames. For example, the instantaneous velocity is obtained by subtracting the center coordinates of the previous frame from the center coordinates of the current frame and dividing by the time interval. The motion vector describes the target's motion state (velocity and direction) and is an important basis for obstacle avoidance decisions.

[0118] In the above process, the advantages of RGB images, depth images, and radar point clouds are fully utilized to improve the accuracy and robustness of target recognition. RGB images provide rich semantic information (target category), while depth images and radar provide precise geometric information (position, shape, velocity). The combination of the two yields a comprehensive understanding of the environment. Hospital environments are dynamic and constantly changing; through multi-sensor fusion and real-time tracking, robots can better cope with dynamic changes such as pedestrian flow and equipment movement.

[0119] S240, based on the motion vectors of multiple dynamic targets, the types of multiple dynamic targets, and the 3D mesh map, the feasible domain of the medical transport robot at a future time point is selected, and obstacle avoidance decision is executed based on the feasible domain and the pre-built obstacle avoidance decision model. The obstacle avoidance decision includes detour, waiting in place, and active avoidance. When the obstacle avoidance decision is detour or active avoidance, the obstacle avoidance path is based on the safety cost function, the deviation cost function of the target path, the efficiency cost function, and the planner.

[0120] Specifically, the feasible region extraction process includes:

[0121] S2401, calculates the motion vector of each dynamic target at each time point based on the uniform velocity model. Location Among them, dynamic targets are at a certain point in time. Location The mathematical expression is:

[0122]

[0123] In the formula, Represents dynamic objectives At the present moment Location, Indicates uncertainty;

[0124] Specifically, based on the known position and motion vector of a dynamic target at the current moment, a uniform motion model is used to predict its future position. This model assumes that the target maintains its current velocity for a short period of time, while introducing an uncertainty term to reflect the increase in prediction error over time. This is a simplified prediction method based on the laws of physical motion.

[0125] Uncertainty Modeled as a Gaussian distribution with variance increasing over time ,in, Related to the type of target (e.g., pedestrians) Larger, stroller (Smaller).

[0126] S2402, determine the expansion radius based on the type of the dynamic target, and based on the dynamic target at a given time point. Location And expanding the radius to construct dynamic targets at future time points Basic danger domain ;

[0127] The safety buffer radius is determined based on the target type, and a spherical or cylindrical basic danger zone is constructed centered on the predicted location. This represents the space occupied by the target body and the basic safety buffer zone. Different types of targets use different radii to match their physical size and behavioral characteristics.

[0128] Furthermore, this application differentiates safety distances by setting different safety boundaries for different targets.

[0129] S2403, based on the motion vector For the basic hazard domain Targeted expansion is performed to obtain the final danger zone. The final danger zone The mathematical expression is:

[0130]

[0131] In the formula, Minkowski addition is represented. Indicates the expansion factor. Indicates the time step;

[0132] Specifically, considering the target's direction of motion, the basic danger zone is expanded along the velocity direction to form a "capsule-shaped" or "spindle-shaped" region. This reflects the need for a larger safety distance in front of the moving object, as the robot requires more reaction time and avoidance space. The faster the speed, the longer the forward expansion, which meets safety requirements.

[0133] S2404, based on the final danger zone The 3D meshed map is annotated to obtain hazardous grids and free grids, wherein the hazardous grids are the areas ultimately deemed hazardous. The grid occupied;

[0134] The continuous hazardous areas are discretized and mapped onto a 3D grid map, with the state of each grid cell marked. This serves as a bridge between the continuous environment and discrete planning, enabling subsequent path planning algorithms to operate on the gridded representation.

[0135] S2405, based on the starting point of the medical transport robot, perform connectivity analysis on the free grid to obtain a feasible grid; and construct a feasible region based on the feasible grid.

[0136] Starting from the robot's current position, a graph search algorithm (such as Breadth-First Search) is used to find all connected components in the free grid. The robot can only move within its own connected component, which is a mathematical guarantee of physical reachability.

[0137] After obtaining the feasible region, the process of making obstacle avoidance decisions includes:

[0138] S2406, the feasible region is divided into multiple connected components, and the minimum channel width of the connected component where the medical transport robot is located is calculated.

[0139] The feasible region is divided into multiple connected components (i.e., isolated free regions), and the minimum pass-through width of the connected component where the robot is located is calculated. This assesses the difficulty of traversing the current region and is a key indicator for determining whether safe passage is possible.

[0140] S2407, compare the minimum channel width with a preset width threshold. If the minimum channel width is greater than or equal to the preset width threshold, determine that there is a drivable environment at the current moment; otherwise, determine that there is no drivable environment at the current moment; and acquire the dynamic target within the target distance range in front of the medical transport robot.

[0141] By comparing the minimum passage width with the robot's required width (plus a safety margin), it is determined whether the current conditions are suitable for safe passage. At the same time, relevant dynamic targets ahead are filtered to avoid unnecessary global detection.

[0142] S2408: When the medical transport robot does not have a drivable environment at the current moment, it shall wait in place until it can drivable.

[0143] When passage is not possible, the robot will automatically stop and wait until the environment improves. This is the most conservative but safest strategy, particularly suitable for avoiding the risks of forcing its way through in a hospital environment.

[0144] S2409, when there is a dynamic target that meets the target conditions within the target distance range in front of the medical transport robot, and there is a driving environment, active avoidance planning is performed based on the pre-built obstacle avoidance decision model to obtain an active avoidance route, and the medical transport robot is controlled to travel along the active avoidance route. The target conditions include a type label as the target type, which will collide with the medical transport robot in the future target time period.

[0145] For high-priority targets (emergency medical personnel, hospital beds, etc.), the robot proactively yields the main passage. A two-stage planning process is employed: first, it avoids the target to a safe location, then waits for the target to pass before returning to its original path. Specifically, the proactive avoidance route generation process includes:

[0146] S24091, Obtain the motion vector of the dynamic target of the target type, and perform linear prediction based on the motion vector of the dynamic target of the target type to obtain the predicted route;

[0147] Assuming the target maintains its current state of motion, predict its future trajectory to obtain the predicted path. ;

[0148]

[0149] S24092, by selecting feasible locations outside the safe distance around the predicted line, candidate grids within the target range around the predicted line and located within the feasible region are filtered out. ;

[0150] S24093, Calculate the avoidance evaluation scores for multiple candidate grids. The mathematical expression for the avoidance evaluation score is:

[0151]

[0152]

[0153] In the formula, As the first weight, As the second weight, As the third weight, For normalization function, Representation and grid Distance to the nearest dynamic target This indicates that the medical transport robot has moved to the grid. distance, To characterize the mesh Shortest distance from the predicted route Is it greater than the distance threshold? The flag bit;

[0154] In the obstacle avoidance evaluation score, This represents the normalized value indicating the distance to the target; the greater the distance, the higher the score (and the safer). It represents the complement of the normalized value of the distance to the avoidance point. The smaller the distance, the higher the score (efficiency). To determine whether to move away from the predicted route (avoidance safety signs).

[0155] S24094, avoidance evaluation score The highest grid is the best avoidance grid. ;

[0156] S24095, taking the current position of the medical transport robot as the starting point, the optimal avoidance grid With the destination as the starting point, path planning is performed using a pre-built obstacle avoidance decision model to obtain the first avoidance route;

[0157] S24096, when a dynamic target of the target type leaves, the second avoidance route is obtained by combining the optimal avoidance grid as the starting point and any point in the target path as the ending point, and the path planning is performed in conjunction with the pre-built obstacle avoidance decision model.

[0158] By filtering through the above evaluation scores, the robot can find the optimal avoidance grid that allows for reasonable avoidance while maintaining a certain distance or close proximity to other targets. This application is based on the optimal avoidance grid. Plan both ends to ensure that the task can be planned smoothly after avoiding obstacles.

[0159] S2410: When there is a non-target type dynamic target within the target distance range in front of the medical transport robot, and a driving environment exists, detour planning is performed based on a pre-built obstacle avoidance decision model to obtain a detour route, and the medical transport robot is controlled to travel along the detour route.

[0160] For non-urgent objectives, the robot plans a route to bypass obstacles and return to the original path, maintaining task continuity without stopping or waiting. Specifically, the process of generating the detour route includes:

[0161] S24101, taking the current position of the medical transport robot as the starting point and any point after the starting point in the target route as the ending point, and combining the pre-built obstacle avoidance decision model to perform path planning, a detour route is obtained.

[0162] After selecting the nearest reachable point on the original path after the obstacle is removed, find the optimal path within the feasible region that avoids all dangerous areas.

[0163] S2411: When there are no dynamic targets within the target distance range ahead of the medical transport robot, and a driving environment exists, control the medical transport robot to continue traveling along the target path. In other words, when there are no obstacles or threats, the robot travels normally along the original path.

[0164] The routes described above are all generated based on the robot's local planner, which also uses the A* path planning algorithm. The corresponding cost function includes:

[0165]

[0166]

[0167]

[0168]

[0169] In the formula, Indicates the cost of security. The weight representing the security cost, Indicates the offset cost. The weights representing the offset cost. Indicates the cost of efficiency. The weights representing the efficiency costs Represents a node The minimum distance to any dynamic target. Represents a node The minimum distance to the target path. Indicates the cumulative path length. Indicates the cumulative curvature of the path. This is the normalization function.

[0170] The aforementioned local path planning employs multi-objective optimization to balance safety, path deviation, and motion efficiency, generating a smooth, safe, and efficient obstacle avoidance path.

[0171] Specifically, security costs Robots are encouraged to maintain a safe distance from moving targets. This indicates that the closer the distance, the greater the cost, encouraging people to stay away from danger.

[0172] Offset cost We encourage that the path should not deviate too much from the original planned path.

[0173] Efficiency Cost Encourage short obstacle avoidance paths and low curvature, and use length weights. curvature weight Balanced values.

[0174] S250, upon completion of obstacle avoidance, returns to the target path and retrieves the RGB image, depth image, and radar point cloud data of the acquired environment until the destination is reached.

[0175] This application presents a control method for a medical transport robot, employing a combination of pre-planned global path design and real-time obstacle avoidance strategies. Global path planning, based on a 3D mesh map, comprehensively considers multiple factors such as distance, congestion, and elevator waiting time to ensure overall optimal task execution. Real-time dynamic perception, fused with depth vision and radar, accurately identifies target types and motion vectors, enabling intelligent obstacle avoidance behaviors such as detours, waiting, and proactive obstacle avoidance. This architecture effectively balances traffic efficiency and path continuity while ensuring zero-collision transport safety, making it particularly suitable for the complex and dynamic scenarios of hospitals with high population density and frequent emergencies. The hierarchical decision-making mechanism avoids the heavy overhead of global planning and optimizes obstacle avoidance paths through local cost models, enabling the robot to complete medical transport tasks politely, efficiently, and reliably, significantly improving the automation level and emergency response capabilities of hospital logistics.

[0176] like Figure 3 As shown, this application also provides a control system based on a medical transport robot, including:

[0177] The route planning module is used to perform preliminary route planning based on the transportation task, a pre-built decision model, and a pre-built 3D gridded map when a transportation task is received from an external source, so as to obtain the target route. The transportation task includes a starting point and an ending point, and the decision model includes a distance cost function, a congestion cost function, an elevator waiting cost function, a static rule cost function, and a planner.

[0178] The travel and environment perception module is used to control the medical transport vehicle to travel along the target path and to collect RGB images, depth images and radar point cloud data of the environment during the travel process;

[0179] The recognition module is used to perform dynamic target recognition based on the RGB image, the depth image and the radar point cloud data, to obtain multiple dynamic targets in the environment where the medical transport vehicle is located, the types of multiple dynamic targets and the motion vectors of multiple dynamic targets;

[0180] The obstacle avoidance module is used to filter the feasible domain of the medical transport robot at a future time point based on the motion vectors of multiple dynamic targets, the types of multiple dynamic targets, and the 3D mesh map, and to execute obstacle avoidance decisions based on the feasible domain and a pre-built obstacle avoidance decision model. The obstacle avoidance decisions include detour, waiting in place, and active avoidance. When the obstacle avoidance decision is detour or active avoidance, the obstacle avoidance path is based on a safety cost function, a deviation cost function of the target path, an efficiency cost function, and a planner.

[0181] The loop control module is used to return to the target path and the travel and environment perception module when obstacle avoidance is completed, until the destination is reached.

[0182] This application discloses a control system for a medical transport robot, employing a pre-planned global path and a real-time obstacle avoidance strategy to control the robot. Global path planning, based on a 3D grid map, comprehensively considers multiple costs such as distance, congestion, and elevator waiting time to ensure overall optimal task execution. Real-time dynamic perception, based on the fusion of depth vision and radar, accurately identifies target types and motion vectors, enabling various intelligent obstacle avoidance behaviors such as detours, waiting, and proactive obstacle avoidance. This architecture effectively balances traffic efficiency and path continuity while ensuring zero-collision transport safety, making it particularly suitable for complex and dynamic scenarios in hospital environments characterized by high population density and frequent emergencies. The hierarchical decision-making mechanism avoids the heavy overhead of global planning and optimizes obstacle avoidance paths through a local cost model, enabling the robot to complete medical transport tasks politely, efficiently, and reliably, significantly improving the automation level and emergency response capabilities of hospital logistics.

[0183] This embodiment also provides an electronic terminal, including: a processor and a memory;

[0184] The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to cause the terminal to perform any of the methods in this embodiment.

[0185] As will be understood by those skilled in the art, the computer-readable storage medium described in this embodiment allows for the implementation of all or part of the steps in the above method embodiments by computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0186] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication between them. The memory is used to store computer programs, the communication interface is used to perform communication, and the processor and the transceiver are used to run the computer programs, so that the electronic terminal performs the steps of the above method.

[0187] In this embodiment, the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0188] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0189] In the above embodiments, although the present application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art based on the foregoing description. The embodiments of the present application are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims.

[0190] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A control method based on a medical transport robot, characterized in that, Including the following steps: When a transportation task is received from an external source, preliminary path planning is performed based on the transportation task, a pre-built decision model, and a pre-built 3D gridded map to obtain the target path. The transportation task includes a starting point and a destination, and the decision model includes a distance cost function, a congestion cost function, an elevator waiting cost function, a static rule cost function, and a planner. Control the medical transport vehicle to travel along the target path, and collect RGB images, depth images and radar point cloud data of the environment during the journey; Dynamic target recognition is performed based on the RGB image, the depth image, and the radar point cloud data to obtain multiple dynamic targets, their types, and their motion vectors in the environment of the medical transport vehicle. This includes: inputting the RGB image into a pre-built recognition model to obtain detection boxes and identification labels for multiple targets; clustering the radar point cloud data to obtain multiple point cloud clusters; transferring the detection boxes and identification labels from the RGB image to the depth image, wherein the RGB image and the depth image are pre-aligned using camera internal parameters; and preprocessing the depth image and the radar point cloud data to obtain preprocessed... Image data and preprocessed point cloud data, wherein the preprocessing includes spatiotemporal alignment, filtering, and coordinate unification; each pixel in the preprocessed image data is converted into a 3D point based on camera intrinsic parameters to obtain depth point cloud data; the depth point cloud data is then clustered to obtain multiple depth point cloud clusters; based on the principle of planar projection, the detection boxes and recognition labels in the depth image are transferred to the multiple depth point cloud clusters; the depth point cloud clusters are matched with the point cloud clusters to obtain multiple data groups with positional matching; for any first data group, the detection boxes and recognition labels of the depth point cloud clusters are transferred to the point cloud clusters to obtain the detection boxes and recognition labels of multiple point cloud clusters; the center coordinates of each point cloud cluster in multiple frames of radar point cloud data are tracked. And based on the center coordinates of multiple frames of radar point cloud data Constructing motion vectors for multiple dynamic targets ,in, The identifier for a point cloud cluster; Based on the motion vectors of multiple dynamic targets, the types of multiple dynamic targets, and the 3D mesh map, the feasible domain of the medical transport robot at a future time point is selected, and obstacle avoidance decisions are executed based on the feasible domain and the pre-built obstacle avoidance decision model. The obstacle avoidance decisions include detour, waiting in place, and active avoidance. When the obstacle avoidance decision is detour or active avoidance, the obstacle avoidance path is based on the safety cost function, the deviation cost function of the target path, the efficiency cost function, and the planner. Upon completion of obstacle avoidance, return to the target path and retrieve the RGB image, depth image, and radar point cloud data of the acquired environment until the destination is reached.

2. The control method based on a medical transport robot according to claim 1, characterized in that, The mathematical expression of the decision model is: In the formula, Indicates the total cost. Indicates distance cost. The weights represent the distance cost. This indicates the cost of congestion. The weights representing the costs of congestion This indicates the cost of waiting for the elevator. This represents the weight of the elevator waiting cost. Represents the cost of static rules. The weights are the costs of static rules. This represents the node index in a 3D gridded map. This indicates the total number of nodes in the path. Indicates the distance between adjacent nodes. Indicates the reference maximum distance. Represents a node Reference congestion coefficient, Indicates passing by elevator Average waiting time This indicates the maximum reference waiting time. To represent nodes Is it located in a restricted area? Represents a node The penalty coefficient.

3. The control method based on a medical transport robot according to claim 1, characterized in that, Based on the motion vectors of multiple dynamic targets, the types of multiple dynamic targets, and the 3D meshed map, the feasible domain of the medical transport robot at future time points is selected, including: Based on the motion vector of each dynamic target and the uniform velocity model, the time point of each dynamic target is calculated. Location Among them, dynamic targets are at a certain point in time. Location The mathematical expression is: In the formula, Represents dynamic objectives At the present moment Location, Indicates uncertainty; The expansion radius is determined based on the type of the dynamic target, and based on the dynamic target at a given time point. Location And expanding the radius to construct dynamic targets at future time points Basic danger domain ; Based on the motion vector For the basic hazard domain Targeted expansion is performed to obtain the final danger zone. The final danger zone The mathematical expression is: In the formula, Minkowski addition is represented. Indicates the expansion factor. Indicates the time step; Based on the final danger zone Filtering multiple future time points in the 3D gridded map The feasible domain.

4. The control method based on a medical transport robot according to claim 3, characterized in that, Based on the final danger zone Filtering multiple future time points in the 3D gridded map The feasible domain includes: Based on the final danger zone The 3D meshed map is annotated to obtain hazardous grids and free grids, wherein the hazardous grids are the areas ultimately deemed hazardous. The grid occupied; Based on the starting point of the medical transport robot, a connectivity analysis is performed on the free grid to obtain a feasible grid; and a feasible region is constructed based on the feasible grid.

5. The control method based on a medical transport robot according to claim 4, characterized in that, Performing obstacle avoidance decisions based on the feasible region includes: The feasible region is divided into multiple connected components, and the minimum channel width of the connected component where the medical transport robot is located is calculated. The minimum channel width is compared with a preset width threshold. If the minimum channel width is greater than or equal to the preset width threshold, it is determined that there is a drivable environment at the current moment; otherwise, it is determined that there is no drivable environment at the current moment; and dynamic targets within the target distance range in front of the medical transport robot are acquired. If the medical transport robot does not have a drivable environment at the current moment, it will remain in place and wait until it can drivable. When there is a dynamic target that meets the target conditions within the target distance range in front of the medical transport robot, and there is a driving environment, active avoidance planning is performed based on the pre-built obstacle avoidance decision model to obtain an active avoidance route, and the medical transport robot is controlled to travel along the active avoidance route. The target conditions include a type label indicating the target type, which will collide with the medical transport robot in the future target time period. When a non-target dynamic target exists within the target distance range ahead of the medical transport robot, and a driving environment exists, detour planning is performed based on a pre-built obstacle avoidance decision model to obtain a detour route, and the medical transport robot is controlled to travel along the detour route. When there are no dynamic targets within the target distance range ahead of the medical transport robot, and a driving environment exists, the medical transport robot is controlled to continue driving along the target path.

6. The control method based on a medical transport robot according to claim 5, characterized in that, The active obstacle avoidance route includes a first obstacle avoidance route and a second obstacle avoidance route. The active obstacle avoidance route is obtained by performing active obstacle avoidance planning based on a pre-built obstacle avoidance decision model, and includes: Obtain the motion vector of a dynamic target of the target type, and perform linear prediction based on the motion vector of the dynamic target of the target type to obtain the predicted route; Candidate grids within the target area surrounding the predicted route and located within the feasible region are selected. ; Calculate the avoidance evaluation score of multiple candidate grids. The mathematical expression for the avoidance evaluation score is: In the formula, As the first weight, As the second weight, As the third weight, For normalization function, Representation and grid Distance to the nearest dynamic target This indicates that the medical transport robot has moved to the grid. distance, To characterize the mesh Shortest distance from the predicted route Is it greater than the distance threshold? The flag bit; Avoidance evaluation score The highest grid is the best avoidance grid. ; Starting from the current position of the medical transport robot, the optimal avoidance grid... With the destination as the starting point, path planning is performed using a pre-built obstacle avoidance decision model to obtain the first avoidance route; When a dynamic target of the target type leaves, the second avoidance route is obtained by using the optimal avoidance grid as the starting point and any point in the target path as the ending point, combined with the pre-built obstacle avoidance decision model.

7. The control method based on a medical transport robot according to claim 5, characterized in that, Based on a pre-built obstacle avoidance decision model, detour planning is performed to obtain detour routes, including: Starting from the current position of the medical transport robot and taking any point after the starting point of the target route as the endpoint, a detour route is obtained by combining the pre-built obstacle avoidance decision model for path planning.

8. The control method based on a medical transport robot according to claim 1, characterized in that, The mathematical expression for the obstacle avoidance decision model is: In the formula, Indicates the cost of security. The weight representing the security cost, Indicates the offset cost. The weights representing the offset cost. Indicates the cost of efficiency. The weights representing the efficiency costs Represents a node The minimum distance to any dynamic target. Represents a node The minimum distance to the target path. Indicates the cumulative path length. Indicates the cumulative curvature of the path. Indicates length weight, Indicates curvature weight, This is the normalization function.

9. A control system based on a medical transport robot, characterized in that, include: The route planning module is used to perform preliminary route planning based on the transportation task, a pre-built decision model, and a pre-built 3D gridded map when a transportation task is received from an external source, so as to obtain the target route. The transportation task includes a starting point and an ending point, and the decision model includes a distance cost function, a congestion cost function, an elevator waiting cost function, a static rule cost function, and a planner. The travel and environment perception module is used to control the medical transport vehicle to travel along the target path and to collect RGB images, depth images and radar point cloud data of the environment during the travel process; The recognition module is used to perform dynamic target recognition based on the RGB image, the depth image, and the radar point cloud data to obtain multiple dynamic targets in the environment where the medical transport vehicle is located, the types of multiple dynamic targets, and the motion vectors of multiple dynamic targets. This includes: inputting the RGB image into a pre-built recognition model to obtain detection boxes and recognition labels for multiple targets; clustering the radar point cloud data to obtain multiple point cloud clusters; transferring the detection boxes and recognition labels of multiple targets in the RGB image to the depth image, wherein the RGB image and the depth image are pre-aligned using camera internal parameters; and preprocessing the depth image and the radar point cloud data to obtain... The process involves preprocessing image data and preprocessing point cloud data, where preprocessing includes spatiotemporal alignment, filtering, and coordinate unification. Each pixel in the preprocessed image data is converted into a 3D point based on camera intrinsic parameters to obtain depth point cloud data. This depth point cloud data is then clustered to obtain multiple depth point cloud clusters. Based on the principle of planar projection, the bounding boxes and identification labels in the depth image are transferred to the multiple depth point cloud clusters. The depth point cloud clusters are matched with the point cloud clusters to obtain multiple data groups with matching positions. For any first data group, the bounding boxes and identification labels of the depth point cloud clusters are transferred to the point cloud clusters to obtain the bounding boxes and identification labels of multiple point cloud clusters. Finally, the center coordinates of each point cloud cluster are tracked within multiple frames of radar point cloud data. And based on the center coordinates of multiple frames of radar point cloud data Constructing motion vectors for multiple dynamic targets ,in, The identifier for a point cloud cluster; The obstacle avoidance module is used to filter the feasible domain of the medical transport robot at a future time point based on the motion vectors of multiple dynamic targets, the types of multiple dynamic targets, and the 3D mesh map, and to execute obstacle avoidance decisions based on the feasible domain and a pre-built obstacle avoidance decision model. The obstacle avoidance decisions include detour, waiting in place, and active avoidance. When the obstacle avoidance decision is detour or active avoidance, the obstacle avoidance path is based on a safety cost function, a deviation cost function of the target path, an efficiency cost function, and a planner. The loop control module is used to return to the target path and the travel and environment perception module when obstacle avoidance is completed, until the destination is reached.

Images