Robot platform-based environment sample whole-process automatic collection method and system

Abstract

The application provides a robot platform-based environment sample full-process automatic acquisition method and system, relates to the technical field of data acquisition, and the method comprises the following steps: a robot platform receives a sampling task instruction and analyzes the sampling task instruction; the robot platform is navigated to a target sampling environment according to a task scheduling instruction to perform panoramic scanning mapping; a multi-modal risk feature of an environment point cloud map is identified; a global sampling path is fitted based on multiple environment sample collection points; multiple adaptive standard operation actions are spliced to obtain an automatic operation sequence; and the global sampling path and the automatic operation sequence are used to drive the robot platform to perform automatic environment sample collection. The technical problems of the prior art, such as a complicated manual environment sample acquisition process, low efficiency and poor data comparability, are solved. The technical effects of robot platform full-process automatic sampling, efficient path planning and standardized operation execution are achieved, and the sampling efficiency and data comparability are improved.

Description

Technical Field

[0001] This invention relates to the field of data acquisition technology, specifically to a method and system for the automatic acquisition of environmental samples throughout the entire process based on a robot platform. Background Technology

[0002] Current environmental sampling methods primarily rely on manual operation, typically requiring sampling personnel to carry multiple devices and collect, label, and transport air, surface, and water samples at different times. This approach is cumbersome, time-consuming, and prone to inconsistencies in sampling times, non-standard sampling procedures, and errors in recording and labeling, affecting data comparability and accuracy. Furthermore, manual sampling is significantly influenced by differences in operator technique and experience, leading to substantial fluctuations in results from different personnel in the same environment. Simultaneously, traditional sampling data is often recorded in a scattered manner, making it difficult to bind to spatiotemporal environmental information, lacking structured management and rapid retrieval methods, which hinders large-scale, multi-site, and continuous environmental monitoring, resulting in decreased reliability and accuracy of environmental risk assessments.

[0003] Existing technologies suffer from technical problems such as cumbersome and inefficient artificial environmental sample collection processes and poor data comparability. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for fully automated collection of environmental samples based on a robot platform, in order to solve the technical problems of cumbersome, inefficient and poorly comparable data in the existing manual environmental sample collection process.

[0005] In view of the above problems, this application provides a method and system for automatic collection of environmental samples throughout the entire process based on a robot platform.

[0006] The first aspect of this application provides a fully automated environmental sample collection method based on a robot platform. This method includes: after receiving a sampling task instruction, the robot platform parses and obtains a task scheduling instruction and a media type specification; according to the task scheduling instruction, the robot platform navigates to the target sampling environment and drives a LiDAR and depth camera to perform panoramic scanning and mapping, generating an environmental point cloud map; by performing multimodal risk feature identification on the environmental point cloud map, multiple environmental sample collection points are calibrated; the media type specification is parsed to obtain multiple adaptive specification operation actions for the multiple environmental sample collection points; a global sampling path is fitted based on the multiple environmental sample collection points; the multiple adaptive specification operation actions are concatenated according to the global sampling path to obtain an automated operation sequence; and the global sampling path and the automated operation sequence are used to drive the robot platform to perform automated environmental sample collection.

[0007] Optionally, using the execution similarity of the multiple adaptive normative operation actions as the aggregation weight, a dynamic aggregation strategy is adopted to aggregate the multiple environmental sample collection points into multiple sampling point clusters; based on the multi-degree-of-freedom motion constraints of the robotic arm of the robot platform, multiple sampling displacement paths of the multiple sampling point clusters are fitted; the multiple sampling displacement paths are spatiotemporally stitched together to generate the global sampling path.

[0008] Optionally, multiple media types and multiple sampling operation actions are extracted from the multiple adaptive normed operation actions of the multiple environmental sample collection points; multiple sets of action execution similarities of the multiple environmental sample collection points are calculated based on the multiple sampling operation actions; a feature matrix framework of the multiple environmental sample collection points is constructed, and multiple sampling space coordinates, multiple media types, and multiple sets of action execution similarities of the multiple environmental sample collection points are filled in to obtain a multi-source feature matrix; a weighted similarity distance is calculated based on the multi-source feature matrix, and the multiple environmental sample collection points are aggregated into the multiple sampling point clusters.

[0009] Optionally, the environmental point cloud map is traversed based on a high contact surface recognition model to identify and mark multiple easily polluted areas; multiple forced sampling coordinates are located on the environmental point cloud map based on forced sampling specifications; surface topology analysis is performed on the multiple easily polluted areas to locate multiple non-forced sampling coordinates, wherein the multiple forced sampling coordinates and the multiple non-forced sampling coordinates constitute the multiple environmental sample collection points.

[0010] Optionally, the spatial identifier code and optimal detection pose of the target sampling environment are extracted according to the task scheduling instruction; a digital map reference path is retrieved from the IoT scheduling center according to the spatial identifier code; the robot platform is navigated to the target sampling environment according to the digital map reference path, and the LiDAR and depth camera are driven to perform panoramic scanning simultaneously to record LiDAR point cloud data and depth image data; after registering the LiDAR point cloud data and depth image data through an iterative nearest point algorithm, a statistical outlier removal algorithm is applied to filter out temporary moving object interference to obtain the environmental point cloud map.

[0011] Optionally, by performing semantic segmentation on the environmental point cloud map, multiple device categories of the multiple environmental sample collection points are identified; multiple spatial attribute labels of the multiple environmental sample collection points are extracted from the environmental point cloud map; multiple standard operations are loaded according to the multiple device categories and multiple spatial attribute labels; the multiple device categories and multiple spatial attribute labels are used as dynamic constraints to match multiple supplementary operation requirements in the media type specification; the multiple supplementary operation requirements are dynamically superimposed on the multiple standard operations to generate the multiple adaptive specification operation actions.

[0012] Optionally, the global sampling path, automated operation sequence, and sampling task instructions are encoded to obtain a scene sampling template; the scene sampling template is then updated to the environment sampling execution library.

[0013] A second aspect of this application provides a fully automated environmental sample collection system based on a robot platform. The system includes: a task instruction parsing module, used to parse and obtain task scheduling instructions and media type specifications after the robot platform receives a sampling task instruction; an environment scanning module, used to navigate the robot platform to the target sampling environment according to the task scheduling instructions, and then drive a lidar and depth camera to perform panoramic scanning and mapping to generate an environmental point cloud map; a collection point calibration module, used to calibrate multiple environmental sample collection points by performing multimodal risk feature identification on the environmental point cloud map; a standardized operation acquisition module, used to parse the media type specifications to obtain multiple adaptive standardized operation actions for the multiple environmental sample collection points; a sampling path fitting module, used to fit a global sampling path based on the multiple environmental sample collection points; an operation sequence acquisition module, used to stitch together the multiple adaptive standardized operation actions according to the global sampling path to obtain an automated operation sequence; and an environmental sample collection module, used to drive the robot platform to perform automated environmental sample collection using the global sampling path and the automated operation sequence.

[0014] One or more technical solutions provided in this application have at least the following technical effects or advantages: The method provided in this application involves a robot platform receiving a sampling task instruction, parsing it to obtain a task scheduling instruction and a media type specification; navigating the robot platform to the target sampling environment according to the task scheduling instruction, and then driving a LiDAR and depth camera to perform panoramic scanning and mapping to generate an environmental point cloud map; identifying multiple environmental sample collection points by performing multimodal risk feature recognition on the environmental point cloud map; parsing the media type specification to obtain multiple adaptive standardized operation actions for the multiple environmental sample collection points; fitting a global sampling path based on the multiple environmental sample collection points; stitching the multiple adaptive standardized operation actions according to the global sampling path to obtain an automated operation sequence; and using the global sampling path and the automated operation sequence to drive the robot platform to automatically collect environmental samples. This achieves the technical effects of fully automated sampling, efficient path planning, and standardized operation execution by the robot platform, improving sampling efficiency and data comparability.

[0015] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 A flowchart illustrating the fully automated environmental sample collection method based on a robot platform provided in this application.

[0018] Figure 2 This is a schematic diagram of the fully automated environmental sample collection system based on a robot platform provided in this application.

[0019] Figure labeling: Task instruction parsing module 11, environment scanning module 12, collection point calibration module 13, standardized operation acquisition module 14, sampling path fitting module 15, operation sequence acquisition module 16, environmental sample acquisition module 17. Detailed Implementation

[0020] This application provides a method and system for fully automated environmental sample collection based on a robotic platform, addressing the technical problems of cumbersome, inefficient, and poorly comparable data in existing manual environmental sample collection processes. It achieves fully automated sampling, efficient path planning, and standardized operation execution via a robotic platform, thereby improving sampling efficiency and data comparability.

[0021] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0022] Example 1, as Figure 1 As shown, this application provides a fully automated environmental sample collection method based on a robot platform, which includes: After receiving the sampling task instruction, the robot platform parses it to obtain the task scheduling instruction and media type specification.

[0023] Specifically, the robot platform is equipped with a high-performance communication unit that supports a variety of common communication protocols, such as Wi-Fi, Bluetooth, and 4G / 5G networks, ensuring stable and rapid reception of sampling task instructions sent from the host computer system or remote control terminal. The robot platform also features a retractable multi-joint humanoid robotic arm to meet the sampling height requirements of different areas and media in multiple scenarios. In addition, the robot platform integrates a foldable plate rack to achieve the classification of large batches of samples at one time, distinguished by different partitions and different boxes, such as a clean item storage area before sampling, a specimen storage area after sampling, constant temperature storage, and automatic binding of traceability codes.

[0024] After receiving the sampling task instruction, the robot platform first deconstructs and semantically maps the field information in the instruction, converting it into multi-dimensional structured data. The instruction uses a format containing fixed and expandable fields. Fixed fields describe the sampling medium type, target sampling scene identifier, sampling time window, and sampling priority. Expandable fields describe the sampling quantity, repetition count, and special operational requirements. The sampling medium type field indicates the category of the environmental sample to be collected, such as air, surface, or water. Different medium types correspond to different sampling equipment parameters and operating procedures. The target sampling scene identifier field uniquely identifies the sampling area or scene unit, such as a hospital, office, or laboratory. The sampling time window field limits the earliest execution time and latest completion time of the sampling task. The sampling priority field indicates the execution order of multiple sampling tasks. The robot platform parses the sampling medium type field according to pre-established field mapping rules and maps it to the corresponding medium category label. For example, when the sampling medium type field is air, the sampling flow rate range, sampling duration, and non-contact sampling constraints of the air sample are associated; when the sampling medium type field is an object surface, the surface wiping pressure range, contact angle, and sampling area are associated; and when the sampling medium type field is water, the sampling depth, container type, and cross-contamination prevention constraints are associated.

[0025] Simultaneously, the robot platform jointly parses the target sampling scene identifier and sampling time window, and generates task scheduling instructions to constrain the execution order and timing relationships by combining sampling priority information. For example, in the same sampling scene, air sample collection is prioritized and completed within a specified time window before switching to surface or water sample collection to avoid cross-interference. Through preset task parsing rules, the sampling task instructions are broken down into task scheduling instructions and media type specifications. The task scheduling instructions constrain the robot platform's navigation order, execution time, and task switching logic, guiding the robot to the target sampling environment. The media type specifications limit the execution flow, sampling duration, and contact method of the sampling action, determining subsequent sampling operations.

[0026] After navigating the robot platform to the target sampling environment according to the task scheduling instructions, it drives the lidar and depth camera to perform panoramic scanning and mapping, generating an environmental point cloud map.

[0027] Furthermore, after navigating the robot platform to the target sampling environment according to the task scheduling instructions, the LiDAR and depth camera are driven to perform panoramic scanning and mapping to generate an environmental point cloud map. The method includes: extracting the spatial identifier code and optimal detection pose of the target sampling environment according to the task scheduling instructions; retrieving the digital map reference path from the IoT scheduling center according to the spatial identifier code; navigating the robot platform to the target sampling environment according to the digital map reference path, driving the LiDAR and depth camera to perform panoramic scanning simultaneously, and recording LiDAR point cloud data and depth image data; registering the LiDAR point cloud data and depth image data through an iterative nearest point algorithm, and then applying a statistical outlier removal algorithm to filter out temporary moving object interference to obtain the environmental point cloud map.

[0028] Specifically, after parsing the task scheduling instructions, the robot platform reads the hierarchical location information related to the sampling location from the instructions and maps it to a spatial identification code for the target sampling environment. This spatial identification code is used to uniquely identify the sampling environment and is typically composed of a building number, floor number, functional area type, and room number, such as ICU-2 ward on the 3rd floor of an inpatient building or laboratory room 501 on the 5th floor of Building A in a teaching building, ensuring that the robot platform can accurately distinguish between different sampling environments.

[0029] After determining the spatial identifier code of the target sampling environment, the robot platform associates and calls the set of environmental feature parameters corresponding to the environment type according to the spatial identifier code. The environmental feature parameters include at least the spatial size range, the typical obstacle distribution height, the height range of the main sampling object, and the width information of the passageway. Then, combining the sampling medium type, sampling range requirements, and time constraints given in the task scheduling instructions, the environmental feature parameters are constrained and fused for calculation. When the sampling medium is an air sample, a location within the height range of personnel activity and without obstruction is preferentially selected as the detection reference point. When the sampling medium is an object surface sample, the horizontal distance and pitch angle between the detection pose and the target surface are limited according to the sampling range requirements. When the sampling medium is a water sample, the vertical height and orientation angle of the detection pose are adjusted based on the sampling range requirements and the container placement height to obtain the optimal detection pose. The optimal detection pose is used to describe the recommended observation state of the sensor in the sampling environment, including the robot's docking position, the installation height of the LiDAR and depth camera, the orientation angle, and the scanning rotation range. For example, in a ward scenario, the optimal detection pose is set to an observation posture approximately 1.5 meters from the center of the bed, 1.2 meters high, and with a horizontal rotation angle of 360 degrees, which can cover the main sampling area in the ward while avoiding obstruction.

[0030] Based on the spatial identification code, the robot platform requests a digital map reference path matching the target sampling environment from the IoT dispatch center. The IoT dispatch center stores detailed digital map information for each area within the scene building. The digital map reference path is a pre-constructed navigation reference path, including the spatial coordinate information of corridors, access control points, and key nodes, used to reduce the computational load of real-time mapping and positioning. After obtaining the digital map reference path, the robot platform autonomously navigates to the target sampling environment along the digital map reference path. Upon reaching the designated detection pose, it drives the LiDAR and depth camera to simultaneously perform a panoramic scan. The LiDAR obtains high-precision spatial contour point cloud data by emitting laser beams and measuring reflection time, while the depth camera obtains depth image data containing texture and distance information using binocular vision principles or time-of-flight technology.

[0031] After acquiring point cloud data from the LiDAR and depth image data from the depth camera, the depth value d(u,v) of each pixel in the depth image data is converted into the corresponding 3D point cloud form (Xc,Yc,Zc). The calculation formulas are: Xc=(y-cx)×d(u,v) / fx, Yc=(v-cy)×d(u,v) / fy, Zc=d(u,v), where (u,v) are pixel coordinates, (cx,cy) are the camera principal point coordinates, and fx and fy are the camera focal length parameters. Then, the converted 3D point cloud form is uniformly mapped to the reference coordinate system of the robot platform using a rigid transformation matrix. Using laser point cloud data as the reference point cloud, and the point cloud generated from the depth image as the point cloud to be registered, the two sets of point clouds are spatially registered by an iterative nearest neighbor algorithm. In each iteration, the Euclidean distance from each point in the point cloud to be registered to the nearest neighbor in the reference point cloud is calculated, and the rigid transformation parameters are solved based on minimizing the overall inter-point distance error, thereby gradually correcting the translation and rotation deviations between different sensor coordinate systems until the error converges.

[0032] After point cloud registration, outliers introduced by temporary moving objects such as IV stands and wheelchairs are filtered using a statistical outlier removal algorithm. This is achieved by calculating the average distance of each point's 20 nearest neighbors (e.g., the global average distance). A statistical threshold is then calculated based on the average distance and standard deviation of all points globally, with the threshold being the average distance plus 1.5-2.0 times the standard deviation. If the average distance of a point's neighborhood exceeds this threshold, it is identified as an outlier and removed. This process eliminates temporary interference while preserving the stable characteristics of the environmental structure, ultimately generating a stable, continuous, and complete environmental point cloud map.

[0033] By precisely navigating the robot platform according to task scheduling instructions and driving the LiDAR and depth camera to perform panoramic scanning and mapping simultaneously, the accuracy and completeness of the environmental point cloud map are improved. This enables the spatial location of the sampling points to accurately correspond to the real environment, providing a reliable spatial basis for multimodal risk feature identification, sampling point calibration, and global sampling path planning. It also effectively reduces data errors caused by environmental occlusion or temporary moving objects, and improves the standardization and comparability of the sampling data.

[0034] By performing multimodal risk feature identification on the environmental point cloud map, multiple environmental sample collection points are calibrated.

[0035] Furthermore, by performing multimodal risk feature identification on the environmental point cloud map, multiple environmental sample collection points are calibrated. The method includes: traversing the environmental point cloud map based on a high contact surface identification model to identify and mark multiple easily polluted areas; locating multiple forced sampling coordinates on the environmental point cloud map based on forced sampling specifications; and performing surface topology analysis on the multiple easily polluted areas to locate multiple non-forced sampling coordinates, wherein the multiple forced sampling coordinates and the multiple non-forced sampling coordinates constitute the multiple environmental sample collection points.

[0036] Specifically, when identifying high-contact surfaces in environmental point cloud maps, the environmental point cloud map is first input into a high-contact surface identification model based on a point cloud deep learning network, such as PointNet. The PointNet model consists of a local feature extraction layer, a hierarchical sampling and grouping layer, a global feature aggregation layer, and a prediction output layer. It can simultaneously extract local geometric features and global structural features of the point cloud, thereby predicting surfaces in the point cloud that may be frequently touched by people or equipment. Among them, the hierarchical sampling and grouping layer performs dimensionality reduction processing on the point cloud through multi-scale sampling: first, several center points are selected in the global point cloud according to uniform spacing or the FPS method as the core points of local point blocks. Then, within the neighborhood of each center point, a subset of the point cloud is selected according to a set radius or the number of neighboring points, such as a radius of 0.05-0.2 meters, with 32-128 points sampled in each group to form local point blocks. This ensures that each point block covers the local spatial structure at different scales, preserving local geometric information while reducing computational load. The local feature extraction layer encodes the features of points within each local point block: the 3D coordinates and normal vector information of each point are input into a multilayer perceptron (MLP). The MLP consists of multiple fully connected layers, with three fully connected layers having dimensions of 64, 128, and 256 respectively. Each layer is followed by a ReLU activation function and batch normalization to achieve nonlinear mapping, mapping the low-dimensional geometric information of the points to high-dimensional feature vectors. Subsequently, max pooling is performed on the high-dimensional features of all points within the point block to obtain fixed-dimensional feature vectors for the local point block, used to capture local geometric patterns, such as planes, edges, surfaces, or curvature changes. The global feature aggregation layer propagates the local features to each point in the original point cloud through interpolation, such as distance-based inverse distance weighted interpolation. At the same time, skip connections are used to fuse the low-level features before hierarchical sampling with the interpolated features, so that each point contains both its own local geometric features and global spatial structure information, enhancing the semantic consistency of the point cloud in the overall environment. The output prediction layer inputs the fused features into the fully connected layer and outputs a high contact probability score for each point through a Sigmoid activation function, ranging from 0 to 1, to determine easily contaminated areas. The high contact surface recognition model is constructed through supervised learning on a training point cloud dataset with contact annotations. During training, the model learns the geometric patterns and spatial distribution of high contact surfaces in different environmental scenarios. In the inference phase, the robot platform traverses and predicts each point in the target environment point cloud, outputting a high contact probability score for each point. Then, using a threshold such as 0.5, multiple easily contaminated areas are selected and annotated, such as bed rails, door handles, and desktop control panels. These easily contaminated areas are then marked on the environmental point cloud map to indicate priority sampling areas where high levels of microorganisms or chemical residues may exist.According to mandatory sampling guidelines, multiple mandatory sampling coordinates are located and marked on the environmental point cloud map. These guidelines are pre-defined standards that clearly stipulate sampling must be conducted at certain specific locations. These locations are points that must be monitored for scientific research or fixed locations that must be sampled according to laboratory requirements, such as bedside in a hospital ward, around the toilet in a bathroom, wastewater discharge outlets, or water source sampling points. The robotic platform accurately matches and marks the location descriptions in the mandatory sampling guidelines with the actual coordinates on the environmental point cloud map, based on the specific requirements in the guidelines and the coordinate information.

[0037] For multiple identified easily polluted areas, surface topology analysis is performed. A subset of point clouds corresponding to each easily polluted area is extracted from the environmental point cloud map. A set of three-dimensional point coordinates is obtained through neighborhood filtering or index mapping. Surface reconstruction is then performed on this set of three-dimensional point coordinates using triangular meshing methods, such as Delaunay triangulation or Poisson reconstruction algorithms, to generate a continuous triangular mesh surface model. This ensures smooth connection of the surface in densely populated point areas and preserves local geometric details. After surface construction, the curvature information of each vertex on the surface is calculated using discrete surface geometry methods, including Gaussian curvature and mean curvature. Gaussian curvature can be calculated using the corner defect method, reflecting the degree of local surface curvature at that point. Mean curvature can be approximated using the discrete Laplacian operator, reflecting the smoothness and curvature intensity of the surface at that point. Abrupt curvature changes or extreme points typically correspond to key surface features, such as edge transitions, depressions, or convexities. Next, a topological analysis is performed on the triangular mesh surface. By analyzing the connectivity of the triangular mesh patches, an adjacency graph is constructed, marking the connected regions to which each patch belongs. Then, combining vertex curvature information, regions with positive and large local curvature are marked as convex regions, and regions with negative and large local curvature are marked as concave regions. Simultaneously, surface boundaries are identified by analyzing the adjacency relationships of boundary vertices; for example, vertices with only one adjacent patch edge or region edge vertex are identified to determine surface continuity and boundary features, thus forming a complete annotation of the overall surface shape, concave-convex structure, and key boundaries. Finally, using a sampling strategy, non-forced sampling coordinates are automatically generated at locations with large curvature and obvious geometric features, such as the surface center, high-curvature edge points, concave-convex junctions, and marked key points in concave-convex regions.

[0038] Finally, multiple mandatory sampling coordinates and multiple non-mandatory sampling coordinates are integrated to form multiple environmental sample collection points. Each environmental sample collection point has a clear spatial location label in the three-dimensional coordinate system. These environmental sample collection points encompass representative and risky locations within the environment, comprehensively reflecting the pollution status. For example, in a 50-square-meter point cloud map of a hospital ward, after traversing using a high-contact surface recognition model, eight easily contaminated areas were identified and marked, including two door handles, three switches, two bedside table armrests, and one bathroom door handle. Based on mandatory sampling specifications, three mandatory sampling coordinates were located: the bedside, the area around the toilet, and near the window. After performing surface topology analysis on the eight easily contaminated areas, five non-mandatory sampling coordinates were located, distributed at the center and edges of the easily contaminated areas. Ultimately, the three mandatory sampling coordinates and the five non-mandatory sampling coordinates together constitute the eight environmental sample collection points for the hospital ward.

[0039] By performing multimodal risk feature identification on environmental point cloud maps, the representativeness and relevance of sampling point selection are improved. This enables the robot platform to automatically determine sampling points in complex environments that meet both regulatory requirements and cover potentially high-risk areas. This provides precise spatial targets for the generation of adaptive regulatory operation actions and global sampling path planning, reduces omissions or biases that may be caused by human experience judgment, and further improves the accuracy and reliability of sampling results.

[0040] The media type specification is analyzed to obtain multiple adaptive specification operation actions for the multiple environmental sample collection points.

[0041] Furthermore, the method of parsing the media type specification to obtain multiple adaptive specification operation actions for the multiple environmental sample collection points includes: identifying multiple device categories for the multiple environmental sample collection points by performing semantic segmentation on the environmental point cloud map; extracting multiple spatial attribute labels for the multiple environmental sample collection points from the environmental point cloud map; loading multiple standard operations based on the multiple device categories and multiple spatial attribute labels; using the multiple device categories and multiple spatial attribute labels as dynamic constraints to match multiple supplementary operation requirements in the media type specification; and dynamically superimposing the multiple supplementary operation requirements onto the multiple standard operations to generate the multiple adaptive specification operation actions.

[0042] Specifically, semantic segmentation is performed on the environmental point cloud map to classify the objects or devices associated with each environmental sample collection point, identifying the device categories of multiple environmental sample collection points, such as monitors and infusion pumps in medical equipment, or tables, chairs, doors, and windows in living facilities. Semantic segmentation can employ a point cloud-based deep learning semantic segmentation network, such as PointNet. This network performs nonlinear mapping and feature fusion on the point cloud through multi-layer local and global feature extraction, outputting a corresponding semantic category probability vector for each point after feature encoding. During the training phase of the semantic segmentation network, a point cloud dataset with object category labels is first constructed. Each point cloud sample contains three-dimensional coordinates and a corresponding category label, such as monitor, infusion pump, tables, chairs, doors, and windows. The point cloud samples are divided into training and validation sets, and data augmentation processing is performed on each sample, including rotation, scaling, random noise addition, or point sparsification, to improve the robustness of the semantic segmentation network to different environmental poses and noise levels. Then, the enhanced point cloud is input into a semantic segmentation network based on point cloud deep learning. The semantic segmentation network performs nonlinear mapping on each point through multi-layer local and global feature extraction modules, mapping low-dimensional spatial coordinates to high-dimensional feature vectors, and then performs feature fusion to obtain the semantic representation of the points. During training, a supervised learning strategy is adopted, using the cross-entropy loss function to calculate the error between the class probability vector output by the semantic segmentation network and the true class label. The parameters of the semantic segmentation network are optimized through backpropagation, enabling the semantic segmentation network to gradually learn the geometric features, spatial distribution patterns, and relative positional relationships of different objects. After training, the semantic segmentation network can predict the class of each point in the unknown environment point cloud during the inference stage, achieving automatic semantic segmentation.

[0043] Multiple spatial attribute labels are extracted from the environmental point cloud map for various environmental sample collection points. These spatial attribute labels include, but are not limited to, information such as the location, orientation, height, and size of the accessible surface area of ​​the collection point within the environment, used to describe operational constraints. For example, collection points located at high altitudes require the use of special sampling tools or operational postures, while collection points located in confined spaces require consideration of operational flexibility.

[0044] Based on the identified multiple device categories and extracted spatial attribute tags, multiple supplementary operation requirements are loaded from a pre-set operation library. Standard operations are predefined basic sampling operations for different device categories and common spatial conditions. For example, when sampling the surface of a monitor, standard operations may include using a specific sampling swab, operating according to a certain wiping direction and force, etc. Multiple supplementary operation requirements are dynamically superimposed onto multiple standard operations to ultimately generate adaptive standardized operation actions for each sampling point. These adaptive standardized operation actions can be directly used to drive the robotic arm to perform environmental sample collection. Each action contains complete execution parameters, path constraints, and operation timing.

[0045] For example, in the environmental point cloud map of a hospital ward, there is an environmental sample collection point located on the surface of a monitor on the bedside table. Semantic segmentation technology identifies the device category of this collection point as a monitor. The spatial attribute label of this collection point is extracted from the environmental point cloud map, showing its height as 1.2 meters and its orientation directly towards the bed. Based on the device category (monitor) and the spatial attribute label, a standard operation is loaded from the operation library, such as using a sterile sampling swab to wipe the monitor surface from left to right. Next, the device category and spatial attribute label are used as dynamic constraints to match supplementary operation requirements in the media type specification. It is found that for sampling the surface of medical equipment, simple disinfection of the area around the equipment is required before and after sampling, and the monitor display should not be obstructed during sampling. Finally, these supplementary operation requirements are dynamically overlaid onto the standard operation to generate an adaptive standard operation: first, perform simple disinfection around the monitor; then, use a sterile sampling swab to wipe the monitor surface from left to right, taking care not to obstruct the display during wiping; and finally, perform simple disinfection again around the monitor after sampling.

[0046] The robot dynamically generates adaptive and standardized operating actions based on the equipment category and spatial attribute labels of the sampling points, which improves the automation and standardization of the sampling operation. This allows the robot to adaptively adjust its operating parameters according to the sampling medium type and the spatial attributes of the sampling points, ensuring that the sampling actions for different media and environments comply with experimental or safety specifications. This reduces data inconsistencies caused by differences in human operation and improves sampling efficiency and reliability.

[0047] A global sampling path is fitted based on the multiple environmental sample collection points.

[0048] Furthermore, the method for fitting a global sampling path based on the multiple environmental sample collection points includes: using the execution similarity of the multiple adaptive normative operation actions as the aggregation weight, and employing a dynamic aggregation strategy to aggregate the multiple environmental sample collection points into multiple sampling point clusters; fitting multiple sampling displacement paths of the multiple sampling point clusters based on the multi-degree-of-freedom motion constraints of the robotic arm of the robot platform; and spatiotemporally stitching the multiple sampling displacement paths to generate the global sampling path.

[0049] Furthermore, using the execution similarity of the multiple adaptive normed operation actions as the aggregation weight, a dynamic aggregation strategy is adopted to aggregate the multiple environmental sample collection points into multiple sampling point clusters. The method includes: extracting multiple media types and multiple sampling operation actions from the multiple adaptive normed operation actions; calculating multiple sets of action execution similarities for the multiple environmental sample collection points based on the multiple sampling operation actions; constructing a feature matrix framework for the multiple environmental sample collection points, filling it with multiple sampling space coordinates, multiple media types, and multiple sets of action execution similarities to obtain a multi-source feature matrix; calculating a weighted similarity distance based on the multi-source feature matrix, and aggregating the multiple environmental sample collection points into the multiple sampling point clusters.

[0050] Specifically, multiple environmental sample collection points and their corresponding media types and sampling actions are extracted from multiple adaptive standardized operational actions. Media type refers to the material environment in which the sampled object exists, such as air, liquid, or solid surface. Sampling actions include specific sampling behaviors such as sampling height, contact pressure, wiping, aspiration, scraping, and sampling duration. Based on the extracted sampling actions, the similarity of multiple sets of actions at multiple environmental sample collection points is calculated using a distance metric method between action feature vectors, such as cosine similarity. The sampling actions are transformed into feature vectors, and the cosine value between the vectors is calculated. The closer the cosine value is to 1, the higher the similarity of the two actions.

[0051] A feature matrix framework is constructed using multiple environmental sample collection points. This framework has each environmental sample collection point as a row and three feature dimensions—sampling spatial coordinates, media type, and action execution similarity—as columns. The sampling spatial coordinates, media type, and action execution similarity of each collection point from multiple environmental sample collection points are filled into the feature matrix framework to form a multi-source feature matrix. This multi-source feature matrix contains three sources of data: geometric coordinates, media type, and operation similarity. During the filling process, the sampling spatial coordinates, media type, and action execution similarity are standardized, for example, using z-score or max-min normalization methods, to eliminate the influence of different units and ensure that the multi-dimensional features are within the same numerical range. After obtaining the multi-source feature matrix, weights are assigned to each type of feature based on the actual application scenario and expert experience to reflect the importance of different features to the aggregation. For example, if the sampling operation has a significant impact on the aggregation of sampling points, a higher weight (e.g., 0.6) is assigned to the action execution similarity, while the weights for sampling spatial coordinates and media type can be set to 0.2 each.

[0052] Based on weighted Euclidean distance or weighted cosine similarity, the distances in the three dimensions are multiplied by their corresponding weights and then summed to obtain the weighted similarity distance between each pair of sampling points. A dynamic aggregation strategy or clustering algorithm is then used to aggregate multiple environmental sample collection points into multiple sampling point clusters. For example, using the K-Means clustering algorithm, K sampling points are randomly selected as initial cluster centers. The value of K can be determined based on actual needs and experience, such as the size of the sampling area and the distribution density of the sampling points. Then, the weighted similarity distance from each sampling point to each cluster center is calculated, and the sampling point is assigned to the cluster represented by the nearest cluster center. The cluster center of each cluster is recalculated, i.e., the average value of all sampling points in the cluster across each feature dimension is used as the new cluster center. This process of assigning sampling points and recalculating cluster centers is repeated until the cluster centers no longer change or a preset number of iterations is reached. Finally, each sampling point is assigned to its corresponding sampling point cluster, achieving effective aggregation of multiple environmental sample collection points.

[0053] The three-dimensional spatial coordinates of all sampling points in each sampling point cluster and their corresponding adaptive canonical operation parameters, such as end effector pose, sampling height, sampling direction, and operation duration, are obtained. Combined with the kinematic constraints of the robotic arm, including the number of joint degrees of freedom, upper and lower limits of joint angles, reachable space of the end effector, joint velocity and acceleration limits, and collision constraints between the robotic arm and environmental obstacles, a pose feasible region is constructed. An inverse kinematics algorithm is used to convert the target pose of the end effector at the sampling points into a combination of joint angles. Simultaneously, motion planning algorithms, such as... Within a cluster of sampling points, continuous displacement paths are generated to ensure that each joint movement is within constraints and collisions are avoided. For multiple sampling points within the cluster, the sampling order is first sorted by spatial distance or movement similarity. Then, path optimization is performed between the sampling orders, smoothing the joint trajectory through interpolation or spline curves to generate multiple local sampling displacement paths. Finally, each path is simulated and verified to check whether the robot arm's end effector trajectory, joint angle changes, and velocity acceleration meet safety and operational specifications, ensuring that each path both covers the sampling points and conforms to the robot arm's motion constraints. Based on this, the robot arm can successfully reach each sampling point within each sampling point cluster, execute adaptive and compliant operational actions, and avoid path inaccessibility caused by joint limitations or environmental obstacles, thus improving sampling efficiency and operational safety.

[0054] The fitted local sampling displacement paths within each sampling point cluster are organized according to spatial location and sampling time order. The start point, end point, sampling action duration, and robotic arm motion state information, such as joint angles, end-effector posture, and velocity, are recorded for each path. Based on a spatiotemporal collaborative strategy, the local paths are concatenated: the shortest reachable connection distance between each local path is calculated spatially, and then processed using motion planning algorithms, such as those based on... The system generates safe and continuous connection trajectories while adjusting the sampling order and waiting time in the time dimension to ensure that the timing of different sampling actions meets the sampling time window and operation priority constraints. During path splicing, conflict detection and constraint optimization are employed to prevent the robotic arm's end effector from colliding with environmental obstacles or exceeding joint limits. Finally, the local paths and their connecting trajectories are combined to form a global sampling path covering all sampling points. The operation actions, spatial positions, and temporal arrangements of each sampling point are accurately preserved, enabling the robot to perform efficient, continuous, standardized, and safe fully automated sampling in multi-sampling-point environments.

[0055] By using the similarity of the execution of adaptive standard operation actions as the aggregation weight, a dynamic aggregation strategy is adopted to aggregate sampling points into sampling point clusters. Then, the sampling displacement path is fitted and a global sampling path is generated. This fully considers the similarity of sampling operations and the robot's robotic arm motion constraints, which can generate more reasonable and efficient sampling paths, reduce the robot's movement distance and time, and improve sampling efficiency.

[0056] The automated operation sequence is obtained by concatenating the multiple adaptive standard operation actions according to the global sampling path.

[0057] The global sampling path and automated operation sequence are used to drive the robot platform to automatically collect environmental samples.

[0058] Specifically, the adaptive standardized operation actions of each environmental sample collection point are mapped to the corresponding global sampling path position. The execution parameters of the actions, such as sampling height, contact method, sampling duration, rotation angle, and disinfection requirements, are aligned with the spatial coordinates and time nodes on the path. Through a time-synchronization interpolation algorithm, the continuous pose of the robotic arm's end effector on the global sampling path is fused with the action parameters of each sampling point, forming a list of actions arranged in execution order. Each action includes spatial coordinates, end effector posture, sampling operation, and action timing information. This action list is encoded into an automated operation sequence that can be directly sent to the robot control system.

[0059] The automated operation sequence, along with the global sampling path, is input into the robot platform's control system. This enables the robotic arm to move continuously along the global sampling path and automatically execute the corresponding operation at each sampling point. The control system plans the robotic arm's trajectory based on the global sampling path, driving the robot platform to the first sampling point. Then, following the first operation instruction in the automated operation sequence, it controls the robotic arm and end effector to complete the corresponding sampling operation, such as using a pipette to draw liquid samples. After completing the operation at the first sampling point, the robot platform is again guided to the next sampling point according to the global sampling path, and the corresponding operation is executed. This cycle continues until all sampling points are sampled, achieving automatic environmental sample collection.

[0060] By splicing global sampling paths with adaptive standardized operating actions to generate automated operation sequences, the continuity and executability of sampling operations are improved. This enables robots to complete efficient, accurate, and safe environmental sample collection according to preset paths and action specifications, avoiding differences in human operation and path redundancy, and ensuring the consistency and comparability of sampling data.

[0061] Furthermore, the method also includes: encoding the global sampling path, automated operation sequence, and sampling task instructions to obtain a scene sampling template; and updating the scene sampling template to the environment sampling execution library.

[0062] Specifically, the planned global sampling path, corresponding automated operation sequence, and original sampling task instructions are uniformly encoded to form a reusable data structure. The global sampling path records the continuous motion trajectory and end-effector pose information of the robot platform and robotic arm in the environment. The automated operation sequence records the execution parameters for each sampling point, including the operation action, sampling height, contact method, sampling duration, and disinfection steps. The sampling task instructions include metadata such as target environment identifier, sampling medium type, task priority, and time window. During the encoding process, JSON, XML, or a custom binary format can be used to uniformly store the global sampling path and automated operation sequence according to time series and spatial coordinate mapping, forming a scene sampling template.

[0063] The generated scene sampling templates are uploaded to the environment sampling execution library, and an index is established and bound to the environment identification code. This allows the robot platform to directly call the scene sampling templates when it receives sampling tasks in the same environment in the future, without having to repeat path planning or operation action generation, thereby quickly executing the entire process of automatic sampling. The environment sampling execution library is a structured and manageable digital storage system used to centrally store standardized templates for various environmental sampling scenarios, including global sampling paths, automated operation sequences, and sampling task instructions. Each template is bound to the spatial identification code of a specific target sampling environment.

[0064] By generating scene sampling templates through encoding and updating them to the execution library, knowledge reuse and rapid scheduling of the sampling process are realized, which improves the robot's execution efficiency, reduces redundant planning and calculation, ensures the consistency and comparability of sampling operations under the same target environment, and improves the efficiency and convenience of environmental sampling task management.

[0065] Example 2, based on the same inventive concept as the robotic platform-based automated environmental sample collection method described in the previous examples, such as... Figure 2 As shown, this application provides a fully automated environmental sample collection system based on a robot platform, wherein the fully automated environmental sample collection system based on a robot platform includes: The task instruction parsing module 11 is used to parse the task scheduling instruction and media type specification after the robot platform receives the sampling task instruction; the environment scanning module 12 is used to navigate the robot platform to the target sampling environment according to the task scheduling instruction, and then drive the lidar and depth camera to perform panoramic scanning and mapping to generate an environmental point cloud map; the collection point calibration module 13 is used to calibrate multiple environmental sample collection points by performing multimodal risk feature identification on the environmental point cloud map; the standardized operation acquisition module 14 is used to parse the media type specification to obtain multiple adaptive standardized operation actions for the multiple environmental sample collection points; the sampling path fitting module 15 is used to fit a global sampling path based on the multiple environmental sample collection points; the operation sequence acquisition module 16 is used to stitch the multiple adaptive standardized operation actions according to the global sampling path to obtain an automated operation sequence; and the environmental sample collection module 17 is used to drive the robot platform to perform automatic environmental sample collection using the global sampling path and the automated operation sequence.

[0066] Furthermore, the sampling path fitting module 15 is also used to: use the execution similarity of the multiple adaptive standard operation actions as the aggregation weight, and adopt a dynamic aggregation strategy to aggregate the multiple environmental sample collection points into multiple sampling point clusters; based on the multi-degree-of-freedom motion constraints of the robotic arm of the robot platform, fit multiple sampling displacement paths of the multiple sampling point clusters; and spatiotemporally coordinate the splicing of the multiple sampling displacement paths to generate the global sampling path.

[0067] Furthermore, the sampling path fitting module 15 is also used to: extract multiple media types and multiple sampling operation actions from the multiple adaptive normed operation actions of the multiple environmental sample collection points; calculate multiple sets of action execution similarities of the multiple environmental sample collection points based on the multiple sampling operation actions; construct a feature matrix framework of the multiple environmental sample collection points, fill in multiple sampling space coordinates of the multiple environmental sample collection points, the multiple media types and multiple sets of action execution similarities, to obtain a multi-source feature matrix; calculate a weighted similarity distance based on the multi-source feature matrix, and aggregate the multiple environmental sample collection points into the multiple sampling point clusters.

[0068] Furthermore, the sampling point calibration module 13 is also used to: traverse the environmental point cloud map based on the high contact surface recognition model to identify and mark multiple easily polluted areas; locate multiple forced sampling coordinates on the environmental point cloud map based on the forced sampling specifications; perform surface topology analysis on the multiple easily polluted areas to locate multiple non-forced sampling coordinates, wherein the multiple forced sampling coordinates and the multiple non-forced sampling coordinates constitute the multiple environmental sample sampling points.

[0069] Furthermore, the environment scanning module 12 is also used to: extract the spatial identifier code and optimal detection pose of the target sampling environment according to the task scheduling instruction; retrieve the digital map reference path from the IoT scheduling center according to the spatial identifier code; navigate the robot platform to the target sampling environment according to the digital map reference path, drive the lidar and depth camera to perform panoramic scanning simultaneously, and record the lidar point cloud data and depth image data; after registering the lidar point cloud data and depth image data through the iterative nearest point algorithm, apply the statistical outlier removal algorithm to filter out temporary moving object interference, and obtain the environment point cloud map.

[0070] Furthermore, the standardized operation acquisition module 14 is also used to: identify multiple device categories of the multiple environmental sample collection points by performing semantic segmentation on the environmental point cloud map; extract multiple spatial attribute labels of the multiple environmental sample collection points from the environmental point cloud map; load multiple standard operations according to the multiple device categories and multiple spatial attribute labels; use the multiple device categories and multiple spatial attribute labels as dynamic constraints to match multiple supplementary operation requirements in the media type specification; and dynamically superimpose the multiple supplementary operation requirements onto the multiple standard operations to generate the multiple adaptive standardized operation actions.

[0071] Furthermore, the system is also used to: encode the global sampling path, automated operation sequence, and sampling task instructions to obtain a scene sampling template; and update the scene sampling template to the environment sampling execution library.

[0072] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The fully automated environmental sample collection method and specific examples based on the robot platform in the foregoing Embodiment 1 are also applicable to the fully automated environmental sample collection system based on the robot platform in this embodiment. Through the foregoing detailed description of the fully automated environmental sample collection method based on the robot platform, those skilled in the art can clearly understand the fully automated environmental sample collection system based on the robot platform in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0073] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0074] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for fully automated environmental sample collection based on a robot platform, characterized in that, The method includes: After receiving the sampling task instruction, the robot platform parses it to obtain the task scheduling instruction and media type specification; After navigating the robot platform to the target sampling environment according to the task scheduling instructions, the robot platform is driven to perform panoramic scanning and mapping using lidar and depth camera to generate an environmental point cloud map. By performing multimodal risk feature identification on the environmental point cloud map, multiple environmental sample collection points are calibrated; The media type specification is analyzed to obtain multiple adaptive specification operation actions for the multiple environmental sample collection points; A global sampling path is fitted based on the multiple environmental sample collection points; The automated operation sequence is obtained by concatenating the multiple adaptive standard operation actions according to the global sampling path; The global sampling path and automated operation sequence are used to drive the robot platform to automatically collect environmental samples. The method includes identifying multiple environmental sample collection points by performing multimodal risk feature recognition on the environmental point cloud map. Based on the high contact surface recognition model, the environmental point cloud map is traversed to identify and mark multiple easily polluted areas; Based on the mandatory sampling specifications, multiple mandatory sampling coordinates were located on the environmental point cloud map; Surface topology analysis is performed on the multiple easily polluted areas to locate multiple non-forced sampling coordinates, wherein the multiple forced sampling coordinates and the multiple non-forced sampling coordinates constitute the multiple environmental sample collection points.

2. The fully automated environmental sample collection method based on a robot platform as described in claim 1, characterized in that, The method for fitting a global sampling path based on the multiple environmental sample collection points includes: Using the execution similarity of the multiple adaptive normative operation actions as the aggregation weight, a dynamic aggregation strategy is adopted to aggregate the multiple environmental sample collection points into multiple sampling point clusters; Based on the multi-degree-of-freedom motion constraints of the robotic arm of the robot platform, multiple sampling displacement paths of the multiple sampling point clusters are fitted; The multiple sampling displacement paths are spatiotemporally stitched together to generate the global sampling path.

3. The fully automated environmental sample collection method based on a robot platform as described in claim 2, characterized in that, Using the execution similarity of the multiple adaptive normative operation actions as the aggregation weight, a dynamic aggregation strategy is employed to aggregate the multiple environmental sample collection points into multiple sampling point clusters. The method includes: Extract multiple media types and multiple sampling operations from the multiple adaptive normative operation actions of the multiple environmental sample collection points; Calculate the similarity of multiple sets of action executions at the multiple environmental sample collection points based on the multiple sampling operation actions; Construct a feature matrix framework for the multiple environmental sample collection points, fill in multiple sampling space coordinates of the multiple environmental sample collection points, the multiple medium types and multiple sets of action execution similarities, and obtain a multi-source feature matrix; The weighted similarity distance is calculated based on the multi-source feature matrix, and the multiple environmental sample collection points are aggregated into the multiple sampling point clusters.

4. The fully automated environmental sample collection method based on a robot platform as described in claim 1, characterized in that, After navigating the robot platform to the target sampling environment according to the task scheduling instructions, the method drives the lidar and depth camera to perform panoramic scanning and mapping to generate an environmental point cloud map, the method including: Extract the spatial identification code and optimal detection pose of the target sampling environment according to the task scheduling instructions; The digital map baseline path is retrieved from the IoT dispatch center based on the spatial identification code. The robot platform is navigated to the target sampling environment based on the digital map reference path, and the lidar and depth camera are driven to perform panoramic scanning simultaneously to record lidar point cloud data and depth image data. After registering the laser point cloud data and depth image data using the iterative nearest point algorithm, the statistical outlier removal algorithm is applied to filter out temporary moving object interference, resulting in the environmental point cloud map.

5. The fully automated environmental sample collection method based on a robot platform as described in claim 1, characterized in that, The method for parsing the media type specification to obtain multiple adaptive specification operation actions for the multiple environmental sample collection points includes: By performing semantic segmentation on the environmental point cloud map, multiple device categories of the multiple environmental sample collection points are identified; Extract multiple spatial attribute labels from the multiple environmental sample collection points from the environmental point cloud map; Load multiple standard operations based on the multiple device categories and multiple spatial attribute tags; The multiple device categories and multiple spatial attribute labels are used as dynamic constraints to match multiple supplementary operational requirements in the media type specification. The multiple supplementary operation requirements are dynamically superimposed onto the multiple standard operations to generate the multiple adaptive standard operation actions.

6. The fully automated environmental sample collection method based on a robot platform as described in claim 1, characterized in that, The method further includes: The global sampling path, automated operation sequence, and sampling task instructions are encoded to obtain a scene sampling template; Update the scene sampling template to the environment sampling execution library.

7. A fully automated environmental sample collection system based on a robot platform, characterized in that: The step of implementing the fully automated environmental sample collection method based on a robot platform according to any one of claims 1 to 6, wherein the fully automated environmental sample collection system based on a robot platform comprises: The task instruction parsing module is used by the robot platform to parse the sampled task instructions to obtain the task scheduling instructions and media type specifications after receiving them. The environment scanning module is used to navigate the robot platform to the target sampling environment according to the task scheduling instructions, and then drive the lidar and depth camera to perform panoramic scanning and mapping to generate an environmental point cloud map. The sampling point calibration module is used to calibrate multiple environmental sample sampling points by performing multimodal risk feature identification on the environmental point cloud map; The standardized operation acquisition module is used to parse the medium type specification to obtain multiple adaptive standardized operation actions for the multiple environmental sample collection points; The sampling path fitting module is used to fit a global sampling path based on the multiple environmental sample collection points; An operation sequence acquisition module is used to concatenate the multiple adaptive standard operation actions according to the global sampling path to obtain an automated operation sequence; The environmental sample acquisition module is used to drive the robot platform to automatically acquire environmental samples using the global sampling path and automated operation sequence.

Images