Intelligent AI full-automatic package storing and taking system
By optimizing the package storage and retrieval system through multimodal perception, graph convolutional networks, and deep reinforcement learning, the problems of low storage space utilization and inefficient path planning in existing technologies are solved, and efficient package storage and retrieval and device coordination are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGSU YOUMAI SPORTS DEV CO LTD
- Filing Date
- 2025-06-25
- Publication Date
- 2026-07-31
AI Technical Summary
Existing parcel storage and retrieval systems lack the ability to accurately perceive and intelligently analyze the physical attributes of parcels, resulting in low storage space utilization and difficulty in coping with complex and ever-changing parcel types and storage and retrieval needs. In particular, path planning is not efficient enough under high-concurrency requests, and the coordination of automated equipment is insufficient.
A multimodal perception acquisition module is used to acquire multi-dimensional information about the package. The package features are analyzed by graph convolutional neural network. A three-dimensional heat map is generated by real-time spatial construction module. A deep reinforcement learning algorithm with spatiotemporal conflict prediction mechanism is embedded to optimize storage location and path planning. Automated storage and retrieval are achieved through a multi-agent collaborative scheduling and execution module.
It enhances the ability to perceive the physical properties of packages, improves space utilization, optimizes path planning efficiency and system throughput, resolves device coordination conflicts in high-concurrency scenarios, and improves overall access efficiency.
Smart Images
Figure CN120646436B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent logistics and automated storage and retrieval technology, and more specifically, to an intelligent AI fully automated package storage and retrieval system. Background Technology
[0002] With the development of intelligent warehousing and automated control technologies, the requirements for the efficiency, intelligence, and space utilization of package storage and retrieval systems are becoming increasingly higher. In scenarios such as logistics, warehousing, and large-scale events, traditional package storage and retrieval methods rely on a fixed storage architecture dominated by manual operation, consisting of static cabinets and manual registration systems. This leads to problems such as excessively long single operation time due to manual sorting and searching, a sharp drop in system throughput during peak periods, and a high accident rate caused by manual operation.
[0003] To overcome the aforementioned shortcomings, existing technologies integrate automated execution technology, introducing automated handling equipment to replace manual physical movement and basic sorting of packages; and establishing a basic digital management system to record simple information about packages and their corresponding fixed storage locations, which to some extent improves storage and retrieval efficiency, reduces some human errors, and enables the system to have a certain processing capacity.
[0004] However, in practical use, it still has some shortcomings, such as the lack of fine perception and intelligent analysis of the physical attributes of packages, as well as the lack of real-time dynamic evaluation of storage space status. Existing technologies are difficult to achieve deep spatial optimization storage based on package characteristics, resulting in room for improvement in storage space utilization. It is also difficult to cope with complex and ever-changing package types and access requirements, especially when handling concurrent requests or non-standard packages, which leads to inefficient path planning and insufficient coordination of automated equipment, thereby affecting overall access efficiency and system throughput. Summary of the Invention
[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides an intelligent AI fully automatic package storage and retrieval system, which solves the problems mentioned in the background art through the following solutions.
[0006] To achieve the above objectives, the present invention provides the following technical solution: A smart AI fully automated package storage and retrieval system includes: Multimodal sensing and acquisition module: used by pre-deployed package acquisition devices to acquire multimodal information of the target package; Dynamic feature recognition module: used to analyze and process the multimodal information using a graph convolutional neural network to obtain the geometric and physical features corresponding to the target package; Real-time space construction module: used to dynamically generate a 3D spatial heat map by fusing storage space data based on voxelized occupiable grid maps; Intelligent decision-making module: a deep reinforcement learning algorithm for embedding a spatiotemporal conflict prediction mechanism, and synchronously optimizes storage location, energy consumption and path conflict probability based on the three-dimensional spatial heat map; Automated scheduling and execution module: Used to decompose concurrent access requests into discrete subtasks using a multi-agent cooperative scheduling algorithm, and drive the robotic arm through digital threads to realize the access of target packages; Data management and monitoring module: used to store and manage the multimodal information, package characteristics, spatial status, users and access requests, and to use an intelligent data bus to realize information interaction between the modules and with external systems.
[0007] Preferably, the multimodal sensing and acquisition module acquires multimodal information about the target package, including the target package's size, shape, weight, center of gravity, surface texture, and physical identifiers.
[0008] Preferably, the dynamic feature recognition module uses a graph convolutional neural network to extract wrapping features from the multimodal information, specifically including: Receive a graph structure composed of preprocessed multimodal information, wherein the node features in the graph structure include the number of nodes and the corresponding feature dimensions; Based on the spatial correlation of sensors provided by the multimodal sensing and acquisition module and the similarity between image features, an adjacency matrix of the graph is dynamically constructed. ; Stack multiple GCN layers, with each GCN layer followed by a ReLU activation function, and perform feature aggregation operations on each GCN layer; The geometric and physical features corresponding to the target package are output separately through parallel branches; The geometric features corresponding to the target package include the package's length, width, height, shape regularity score, and contour curvature features describing the details of the package's surface; The physical feature branches corresponding to the target package include the package's material classification probability distribution, density estimate, and fragility score.
[0009] Preferably, the dynamic feature recognition module dynamically constructs an adjacency matrix of the graph based on the sensor spatial location correlation and the similarity between image features provided by the multimodal perception and acquisition module, specifically including: Based on the spatial locations of the sensors provided by the multimodal sensing and acquisition module, the spatial correlation weights between the sensors are calculated. Specifically, it is expressed as: , in, Represented as a sensor With sensors Spatial correlation weights between them and These are respectively represented as sensors With sensors Coordinates in three-dimensional space Represented as a scale parameter. Represented as Euclidean distance; Based on the features of each node in the graph structure, the similarity weights between image features are calculated. Specifically, it is expressed as: ,
[0010] in, Represented as nodes With nodes Similarity weights between and Represented as nodes With nodes Image feature vectors, Represented as Euclidean distance; The edge weights of the adjacency matrix of the graph Through geometric mean fusion, it is specifically expressed as follows: .
[0011] Preferably, in the intelligent decision-making module, the spatiotemporal conflict prediction mechanism includes storage location conflicts, path conflicts, and sequence conflicts.
[0012] Preferably, the intelligent decision-making module, based on the three-dimensional spatial heat map, synchronously optimizes the storage location, specifically including: The system acquires precise geometric and physical features of packages provided by the dynamic feature recognition module, and combines these with storage density heat, access frequency heat, and device accessibility heat from the three-dimensional spatial heat map to analyze storage requirements and suitable storage area types. The real-time storage space status and the characteristics of the package to be accessed are used as the environmental status. Through the learned decision-making strategy, an execution signal is sent, which determines whether the target package is placed in a specific storage unit. The reward is calculated based on a preset reward function, and the environmental status and reward are fed back.
[0013] The technical effects and advantages of this invention are as follows: 1. This invention improves the fine perception capability of package physical properties through a multimodal sensing and acquisition module, directly alleviating the problem of insufficient spatial optimization caused by the lack of fine perception of package physical properties; 2. This invention models the spatial relationship between sensors and the semantic association between images through a dynamic feature recognition module, which alleviates the robustness problem of feature extraction for irregular packages, directly supports deep space optimization storage based on package features, and improves space utilization. 3. This invention employs a reinforcement learning-based optimal decision-making strategy through an intelligent decision-making module. Combined with spatiotemporal conflict prediction, it avoids device coordination conflicts in advance. When processing concurrent requests, it can dynamically balance storage efficiency and energy consumption, significantly improving path planning efficiency and system throughput in high-concurrency scenarios. Attached Figure Description
[0014] Figure 1 This is a block diagram of an intelligent AI fully automatic package storage and retrieval system provided according to an embodiment of this application. Detailed Implementation
[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0016] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items, and in the description of the embodiments of this application, unless otherwise stated, “a plurality” means two or more.
[0017] As attached Figure 1 The system shown is an intelligent AI fully automated package storage and retrieval system, which includes a multimodal perception and acquisition module, a dynamic feature recognition module, a real-time spatial construction module, an intelligent decision-making module, an automated scheduling and execution module, and a data management and monitoring module.
[0018] The multimodal sensing and acquisition module is used by pre-deployed package acquisition equipment to acquire multimodal information of the target package.
[0019] Specifically, the multimodal perception and acquisition module acquires multi-dimensional physical attributes and identification information of the target package through a variety of pre-deployed sensors, multispectral cameras, and 3D laser scanners. The multimodal information of the target package includes its size, shape, weight, center of gravity, surface texture, and physical identification, wherein the physical identification includes, but is not limited to, barcodes, QR codes, etc.
[0020] In Example 1, four sets of 360° panoramic fisheye cameras were deployed on the dome at the warehouse entrance. These cameras have a high resolution of 3840×2160 and a frame rate of 60fps, forming a wide three-dimensional acquisition area covering a height from 0.2 meters to 3 meters. On both sides of the conveyor belt in the sorting area, linear scanning industrial cameras were installed. These industrial cameras are configured with a pixel size of 3.45μm and capture images of moving packages at a linear rate of up to 2 meters per second, making them suitable for package identification on high-speed assembly lines.
[0021] It should be noted that in Embodiment 1, in addition to the camera, various types of sensors are deployed to acquire physical property information of the package; the sensors include, but are not limited to, piezoelectric weighing platforms, 3D laser scanners, RFID read / write arrays, etc.; wherein the piezoelectric weighing platforms are embedded in key positions of the conveyor belt in an intermittent manner, with a weighing range covering 0.1kg to 50kg and an accuracy of up to ±1g; the 3D laser scanner is deployed at multiple points on the arched gantry above the package channel, scanning the package surface with a high point cloud density of 1000 points / cm² and a Z-axis accuracy of 0.1mm to generate high-precision three-dimensional point cloud data; the RFID read / write array is deployed in a ring at the boundary of the acquisition area, with a reading speed of up to 200 tags / second.
[0022] In Embodiment 2, a dual sensing mechanism of laser light curtain and millimeter-wave radar is used to detect the entry of packages. When any package enters the preset collection area, the dual sensing mechanism immediately triggers a multi-device synchronous start protocol. The multi-device synchronous start protocol instructs all collection devices such as cameras, laser scanners, and weighing platforms to start working synchronously in a very short time. Furthermore, the camera initiates multi-angle shooting, including multiple preset angles such as 0°, 45°, and 90°, to acquire images of different sides of the package and transmit RGB-D image streams. The laser scanner emits scanning beams to scan the surface of the package and uploads the generated point cloud data to the point cloud server. Simultaneously, the piezoelectric weighing platform activates its pressure sensing function, begins collecting the weight data of the package, and submits the weight distribution matrix to the data center. The RFID read / write array is activated synchronously, attempting to read the RFID tag information on the package.
[0023] The dynamic feature recognition module is used to analyze and process the multimodal information using a graph convolutional neural network to obtain the geometric and physical features corresponding to the target package.
[0024] Specifically, the dynamic feature recognition module uses real-time analysis of the multimodal information from the multimodal perception and acquisition module via graph convolutional neural networks to accurately extract the geometric and physical features of the target package. In this embodiment, the packages are classified in multiple dimensions based on the geometric and physical features extracted by the dynamic feature recognition module, and corresponding storage strategies are associated. This includes, but is not limited to, classifying regular cubes and non-fragile items into high-density stacking areas to maximize space utilization; classifying long strips and metal materials into magnetic shelf partitions to utilize special storage structures; classifying irregular shapes and fragile items into independent air-cushioned compartments, with storage and retrieval performed by low-speed robotic arms; and classifying flexible packages and low-density items into compressible cargo compartments to further improve space utilization.
[0025] Furthermore, the multimodal information included in the target is preprocessed. This preprocessing includes, but is not limited to, processing of images, point cloud data, sensor data, and other sensor data. Specifically, image preprocessing includes, but is not limited to, scaling images to a uniform size, pixel value normalization, denoising, and contrast enhancement. Point cloud data preprocessing includes, but is not limited to, point cloud downsampling, denoising, normal vector estimation, and converting point cloud data into a graph structure. Sensor data standardization includes, but is not limited to, standardizing numerical sensor data such as weight and size to achieve zero mean and unit variance, eliminating the influence of data with different dimensions. Identification information processing includes, but is not limited to, verifying and formatting decoded barcodes and QR codes.
[0026] In one possible implementation, constructing a feature extraction model for the target package based on the graph convolutional neural network includes: receiving a graph structure composed of preprocessed multimodal information, wherein the node features in the graph structure include the number of nodes and the corresponding feature dimensions; and dynamically constructing an adjacency matrix of the graph based on the sensor spatial location correlation and the similarity between image features provided by the multimodal perception acquisition module. Multiple GCN layers are stacked, with each GCN layer followed by a ReLU activation function, and feature aggregation is performed in each GCN layer. The geometric and physical features corresponding to the target package are output through parallel branches. The geometric features corresponding to the target package include the length, width, height, shape regularity score, and contour curvature features describing the surface details of the package. The physical features corresponding to the target package include the material classification probability distribution, density estimate, and fragility score of the package.
[0027] It should be noted that the node features include image features, point cloud features, sensor values, etc.; the adjacency matrix describes the connection relationships and weights between nodes; the shape regularity score ranges from 0 to 1, with 1 indicating complete regularity; the material classification probability distribution represents the probability that the target package belongs to categories such as cardboard, plastic, and metal; and the fragility score ranges from 0 to 1, with 1 indicating extreme fragility.
[0028] In one possible implementation, constructing a feature extraction model corresponding to the target package based on the graph convolutional neural network further includes: calculating the spatial correlation weights between sensors based on the sensor spatial locations provided by the multimodal perception acquisition module. Specifically, it is expressed as: ,
[0029] in, Represented as a sensor With sensors Spatial correlation weights between them and These are respectively represented as sensors With sensors Coordinates in three-dimensional space, and , This is expressed as a scale parameter, and in this embodiment, its value is taken as 1 / 2 of the average distance of the sensor network. Represented as Euclidean distance; similarity weights between image features are calculated based on the features of each node in the graph structure. Specifically, it is expressed as: ,
[0030] in, Represented as nodes With nodes Similarity weights between and Represented as nodes With nodes Image feature vectors, Represented as Euclidean distance; the edge weights of the adjacency matrix of the graph. Through geometric mean fusion, it is specifically expressed as follows: .
[0031] It should be noted that the calculated value range of the similarity weights between the image features is as follows: It needs to be mapped to a linear transformation. The elements of the adjacency matrix are configured with a similarity threshold. If the similarity exceeds the preset similarity threshold, it is determined that there is a connection between the packages, and the value is 1; otherwise, it is 0.
[0032] Furthermore, the calculation formula for the feature aggregation operation performed at each GCN layer is specifically expressed as follows: ,
[0033] in, Represented as the first The node feature matrix of the layer, Represented as the first The node feature matrix of the layer, Represented as a preset weight matrix, Represented as an adjacency matrix with self-connections. Represented as The corresponding degree matrix.
[0034] The real-time space construction module is used to dynamically generate a three-dimensional space heat map by fusing storage space data with voxelized occupiable grid maps.
[0035] In Embodiment 2, the voxelized occupancy grid map of the real-time spatial construction module uses 10cm×10cm×10cm as the basic unit and updates the unit occupancy status in real time. The unit occupancy status includes occupied, idle, and unknown.
[0036] Specifically, the three-dimensional spatial heat map is not a single-dimensional representation, but rather a superposition of multiple key heat indicators. Each indicator is derived through a preset calculation formula and is distinguished by different visualization codes. Furthermore, the three-dimensional spatial heat map contains at least three core heat dimensions, including but not limited to storage density heat, access frequency heat, and device accessibility heat.
[0037] In Embodiment 3, the core thermal dimensions of the three-dimensional spatial heatmap include storage density heat, access frequency heat, and device accessibility heat. The storage density heat is calculated by summing the ratios of the total volume of all packages within a preset area to the total volume of voxels in that area, reflecting the degree of filling in the preset area. The storage density heat is visualized using a red gradient, with darker red indicating higher storage density. The access frequency heat represents the total number of times packages within the preset area are accessed per unit time. The access frequency heat is visualized using a blue gradient, with darker blue indicating more frequent package access in that area. The device accessibility heat measures the ease with which automated equipment can reach the preset area. Shorter paths and fewer turns indicate higher accessibility. The device accessibility heat is visualized using a green gradient, with darker green indicating better physical accessibility to the equipment in that area.
[0038] The intelligent decision-making module is used to embed a deep reinforcement learning algorithm for spatiotemporal conflict prediction mechanism, and to simultaneously optimize storage location, energy consumption and path conflict probability based on the three-dimensional spatial heat map.
[0039] Specifically, the intelligent decision-making module receives package features from the dynamic feature recognition module and a three-dimensional spatial heat map provided by the real-time spatial construction module, and simultaneously optimizes the storage location of the package, the operating energy consumption of the automated equipment, and the probability of potential path conflicts based on a deep reinforcement learning algorithm with an embedded spatiotemporal conflict prediction mechanism.
[0040] Furthermore, the implementation of the intelligent decision-making module includes the application of deep reinforcement learning algorithms and spatiotemporal conflict prediction mechanisms, optimization target setting based on three-dimensional spatial heatmaps, storage location planning process, and evaluation and optimization of energy consumption and path conflict probability.
[0041] Furthermore, the application of the deep reinforcement learning algorithm and the spatiotemporal conflict prediction mechanism includes: selecting a deep reinforcement learning algorithm to handle decision-making problems, learning the optimal decision-making strategy through interaction with the environmental state, without the need for pre-setting complex rules; the spatiotemporal conflict prediction mechanism predicts spatiotemporal conflicts that will occur within a preset time based on the real-time poses of all current automated devices, the expected motion trajectories, and the number and type of packages to be processed before planning any storage or retrieval operation.
[0042] It should be noted that the environmental state includes package characteristics, storage space status, device location, etc., and the decision-making strategy is to select the storage location and plan the package retrieval path. Through a large number of trial and error attempts and reward signals, the reward signals include but are not limited to successful storage and retrieval, improved space utilization, reduced energy consumption, and avoidance of conflicts. The spatiotemporal conflicts include storage location conflicts, path conflicts, and sequence conflicts. The location conflict refers to the possibility that multiple packages are planned to be placed in the same storage unit, or the possibility that the grabbing and placement operations of different devices are in the same storage area. The path conflict refers to the possibility that the movement paths of multiple robotic arms intersect or overlap within the same time period, resulting in collisions or deadlocks. The sequence conflict refers to the possibility that the execution order of different tasks may lead to inefficiency or resource contention.
[0043] Furthermore, the optimization objective setting based on the three-dimensional spatial heatmap includes: using the three-dimensional spatial heatmap to guide decision-making and optimization objectives, wherein the optimization objectives include simultaneously optimizing storage location, energy consumption, and path conflict probability; constructing a reward function by weighted combination of optimization objectives; and enabling the deep reinforcement learning algorithm to learn by maximizing the reward function.
[0044] In Example 4, the optimization rewards for storage locations include a positive reward for successfully placing the package in an area with low storage density or a high degree of matching with the package's features; and a penalty for failing to make effective use of space. The optimization rewards for energy consumption include a positive reward for choosing a nearby storage location or path with a smooth movement trajectory; and a penalty for long-distance movement or complex operations. The rewards for avoiding path conflicts include a positive reward for the planned path being predicted to have a low probability of conflict; and a penalty for predicting a high probability of conflict.
[0045] Furthermore, the storage location planning process based on package features and spatial heatmaps includes: acquiring the precise geometric and physical features of the packages provided by the dynamic feature recognition module, combining the storage density heatmap, access frequency heatmap, and device accessibility heatmap from the three-dimensional spatial heatmap, and analyzing storage requirements and suitable storage area types; the storage requirements include, but are not limited to, independent storage, stacking, and proximity to entrances and exits; using the real-time storage space status and the features of the packages to be stored as the environmental status, and sending an execution signal through the learned decision-making strategy, the execution signal determining the placement of the target package in a specific storage unit; calculating the reward according to a preset reward function, and feeding back the environmental status and the reward.
[0046] Furthermore, the assessment and optimization of energy consumption and path conflict probability include: establishing an energy consumption assessment model to quantify the impact of different decision-making schemes on equipment energy consumption. Factors influencing equipment energy consumption include the type of automated equipment, travel distance, acceleration, deceleration, the motion angle and torque of the robotic arm joints, and the complexity of the grasping or placing operation. Combining a spatiotemporal conflict prediction mechanism and path information from a three-dimensional spatial heatmap, the conflict probability of different path planning schemes is assessed. Conflict factors include analyzing the probability of multiple robotic arms traversing the same area within the same time period, path intersections, and possible sequential conflicts.
[0047] The automated scheduling and execution module is used to decompose concurrent access requests into discrete subtasks using a multi-agent cooperative scheduling algorithm, and to drive the robotic arm through digital threads to realize the access of target packages.
[0048] Furthermore, the multi-agent collaborative scheduling algorithm of the automated scheduling execution module includes: the Hungarian algorithm to solve the robotic arm task allocation matrix, the time window constraint planning of the spatiotemporal conflict resolver, and the activation deformation compensation controller for non-standard packages.
[0049] Specifically, the Hungarian algorithm for solving the robotic arm task allocation matrix includes: decomposing multiple concurrent access requests from the intelligent decision-making module into discrete sub-tasks; using the Hungarian algorithm to construct and solve the robotic arm task allocation matrix, where the elements of the task allocation matrix represent the benefits of a specific combination of robotic arms performing sub-tasks; the spatiotemporal conflict resolver employing time window constraint planning includes: using a time window constraint planning method, for each allocated sub-task, planning an initial path based on the device's motion capabilities and the environmental map, identifying potential conflict points through spatiotemporal analysis, and allocating a time window for each device in the conflict area that is only allowed to pass through within a specified time period; the activation of the deformation compensation controller for non-standard packages includes: when packages with high deformation coefficients or packages classified as requiring special handling are present, activating the deformation compensation controller for the robotic arm performing the target task during the task allocation and path planning stages.
[0050] The data management and monitoring module is used to store and manage the multimodal information, package characteristics, spatial status, users and access requests, and uses an intelligent data bus to realize information interaction between the modules and with external systems.
[0051] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. An intelligent AI full-automatic package storage and retrieval system, characterized in that, include: Multimodal sensing and acquisition module: used by pre-deployed package acquisition devices to acquire multimodal information of the target package; Dynamic Feature Recognition Module: Used to analyze and process the multimodal information using a graph convolutional neural network to obtain the geometric and physical features corresponding to the target package. This includes: receiving a graph structure composed of preprocessed multimodal information, where the node features in the graph structure include the number of nodes and their corresponding feature dimensions; and dynamically constructing an adjacency matrix of the graph based on the sensor spatial location correlation and the similarity between image features provided by the multimodal perception and acquisition module. Stack multiple GCN layers, with each GCN layer followed by a ReLU activation function, and perform feature aggregation operations on each GCN layer: in, Represented as the first The node feature matrix of the layer, Represented as the first The node feature matrix of the layer, Represented as a preset weight matrix, Represented as an adjacency matrix with self-connections. Represented as The corresponding degree matrix; the geometric and physical features of the target package are output through parallel branches respectively; the geometric features of the target package include the length, width, height of the package, shape regularity score, and contour curvature features describing the details of the package surface; the physical features of the target package include the material classification probability distribution, density estimate, and fragility score of the package; Real-time space construction module: used to dynamically generate a 3D spatial heat map by fusing storage space data based on voxelized occupiable grid maps; The intelligent decision-making module is used to embed a deep reinforcement learning algorithm for spatiotemporal conflict prediction and to simultaneously optimize storage location, energy consumption, and path conflict probability based on the three-dimensional spatial heat map. Specifically, the optimization of storage location includes: acquiring precise geometric and physical features of the package provided by the dynamic feature recognition module, combining these with storage density heat map, access frequency heat map, and device accessibility heat map to analyze storage needs and suitable storage area types; using the real-time storage space status and the features of the package to be accessed as the environmental state, and sending an execution signal based on the learned decision-making strategy, the execution signal determining where to place the target package in a specific storage unit; calculating the reward according to a preset reward function, and feeding back the environmental state and the reward. Automated scheduling and execution module: Used to decompose concurrent access requests into discrete subtasks using a multi-agent cooperative scheduling algorithm, and drive the robotic arm through digital threads to realize the access of target packages; Data management and monitoring module: used to store and manage the multimodal information, package characteristics, spatial status, users and access requests, and to use an intelligent data bus to realize information interaction between the modules and with external systems.
2. The intelligent AI fully automatic package storage and retrieval system according to claim 1, characterized in that: The multimodal sensing and acquisition module acquires multimodal information about the target package, including the target package's size, shape, weight, center of gravity, surface texture, and physical identifiers.
3. The intelligent AI fully automatic package storage and retrieval system according to claim 1, characterized in that: The dynamic feature recognition module dynamically constructs an adjacency matrix of the graph based on the sensor spatial location correlation and the similarity between image features provided by the multimodal sensing and acquisition module, specifically including: Based on the spatial locations of the sensors provided by the multimodal sensing and acquisition module, the spatial correlation weights between the sensors are calculated. Specifically, it is expressed as: ， in, Represented as a sensor With sensors Spatial correlation weights between them and These are respectively represented as sensors With sensors Coordinates in three-dimensional space Represented as a scale parameter. Represented as Euclidean distance; Based on the features of each node in the graph structure, the similarity weights between image features are calculated. Specifically, it is expressed as: ， in, Represented as nodes With nodes Similarity weights between and Represented as nodes With nodes Image feature vectors, Represented as Euclidean distance; The edge weights of the adjacency matrix of the graph Through geometric mean fusion, it is specifically expressed as follows: 。 4. The intelligent AI fully automatic package storage and retrieval system according to claim 1, characterized in that: The intelligent decision-making module includes a spatiotemporal conflict prediction mechanism, which includes storage location conflicts, path conflicts, and sequence conflicts.