Automatic stacking system and method based on large model
Through an automatic stacking system based on a large model, combined with data collection and analysis of visual sensors, force sensors and distance sensors, a stacking strategy is generated, which solves the intelligence and adaptability problems of traditional automatic stacking systems in complex environments and realizes efficient and safe cargo handling and stacking operations.
Patent Information
- Application Number
- CN202510841944.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-23
- Publication Date
- 2025-09-23
AI Technical Summary
Traditional automatic stacking systems have low intelligence and poor adaptability when faced with complex storage environments and diverse cargo. They find it difficult to quickly and accurately plan grasping and stacking strategies, and are prone to collisions in the presence of dynamic obstacles, affecting operational safety and continuity.
An automatic stacking system based on a large model is adopted, including a perception module, a decision-making and planning module, an execution module and a data storage module. It uses visual sensors, force sensors and distance sensors to collect data in real time, conducts in-depth analysis through the large model processing module, generates a stacking task execution strategy, and executes the task through the execution module, while performing autonomous learning and optimization.
It has significantly improved the efficiency and accuracy of warehousing and logistics operations, reduced operating costs, increased system decision accuracy by 3%, increased operating efficiency by 15%-20%, and reduced operating costs by approximately 22%.
Smart Images
Figure CN120681470A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of automatic stacking, and in particular relates to an automatic stacking system and method based on a large model. Background Art
[0002] In the modern logistics and warehousing industry, automated stacking is widely used for operations such as cargo handling and stacking to improve operational efficiency and reduce labor costs. Traditional automated stacking typically relies on preset programs and simple sensor feedback for operation, and its intelligence level is relatively low. This leads to numerous limitations when dealing with complex and changing warehousing environments and diverse cargo. For example, when handling cargo of varying shapes, sizes, and weights, it is difficult to quickly and accurately plan appropriate grasping and stacking strategies. In the presence of dynamic obstacles in the warehousing environment (such as moving people and other mobile equipment), automated stacking systems are unable to adjust their operating paths in a timely and effective manner, which can easily lead to collisions and affect the safety and continuity of operations. With the rapid development of the logistics industry, higher requirements are being placed on the intelligence, adaptability, and efficiency of automated stacking, and there is an urgent need to introduce advanced technologies to improve its performance. Summary of the Invention
[0003] The purpose of the present invention is to provide an automatic stacking system and method based on a large model to solve the problems of low intelligence and poor adaptability of existing automatic stacking when facing complex storage environments and diversified goods.
[0004] To achieve the above object, the technical solution adopted by the present invention is as follows: An automatic stacking system based on a large model, comprising a perception module, a large model processing module, a decision-making and planning module, an execution module, and a data storage module; The perception module is used to collect real-time image information of the goods, image information of the storage environment, information about the force used to grab the goods, and information about the distance between the goods and surrounding objects; The large model processing module is connected to the perception module and is used to identify the basic information of the goods and the storage environment information based on the information obtained by the large module and the perception module pre-stored therein; The decision-making and planning module is connected to the large model processing module and is used to generate a stacking task execution strategy based on the basic information of the goods and the storage environment information obtained by the large model processing module; The execution module is connected to the decision-making and planning module and is used to perform cargo grabbing, handling and stacking operations according to the stacking task execution strategy of the decision-making and planning module; The data storage module is connected to the perception module, large model processing module, decision-making and planning module, and execution module respectively. It is used to store data collected by the perception module, data identified by the large model processing module, strategy data generated by the decision-making and planning module, and task execution data of the execution module.
[0005] Furthermore, the perception module includes a visual sensor, a force sensor, and a distance sensor; Visual sensors are used to obtain image information of goods and storage environments; The force sensor is installed on the grabbing component to monitor the force when grabbing the goods in real time; Distance sensors are used to detect the distance between goods and surrounding objects.
[0006] Furthermore, visual sensors include cameras and 3D laser scanners.
[0007] Furthermore, the basic information of the goods includes the shape, size, color, weight, location, placement and label information of the goods; The storage environment information includes shelf layout, obstacle distribution and obstacle movement trends in the storage environment.
[0008] Furthermore, the stacking task execution strategy includes the optimal location and method for grabbing goods, planning the operation path in the warehouse environment, and formulating the stacking sequence and layout plan.
[0009] Furthermore, the pre-stored large module in the large model processing module uses the data stored in the data storage module as a training set and performs iterative training at regular intervals.
[0010] The present invention also provides an automatic stacking method based on a large model, which uses the automatic stacking system based on a large model, including the following steps: Step (1), data acquisition: the perception module collects real-time image information of the goods, image information of the storage environment, information on the force used to grab the goods, and information on the distance between the goods and surrounding objects, and transmits these data to the large model processing module; Step (2), data processing and analysis: The large model processing module pre-processes the received data, and then uses the pre-stored large model to deeply analyze and understand the data, thereby identifying the basic information of the acquired goods and the storage environment information; Step (3), decision making: the decision planning module makes the execution decision of the stacking task based on the recognition results of the large model processing module and generates the stacking task execution strategy; Step (4), task execution: the execution module controls the automatic stacking equipment to execute the stacking task according to the stacking task execution strategy generated by the decision-making and planning module; Step (5), learning and optimization: After completing each stacking task, the system records and analyzes the execution-related data of this task, including the data collected by the knowledge module, the data identified by the large model processing module, the strategy data generated by the decision-making planning module, and the task execution data of the execution module, and inputs them as new data samples into the large model of the large model processing module for further training, and continuously optimizes the performance of the large model; at the same time, for cases of unfinished stacking tasks, analyze them, find out the reasons, and make targeted improvements.
[0011] Furthermore, in step (2), the preprocessing includes removing noise and invalid data.
[0012] Furthermore, in step (3), when making the execution decision of the stacking task, the first step is to determine the grabbing strategy of the goods, including selecting the appropriate grabbing point, grabbing method, and grabbing force; then, the optimal path from the current location to the goods storage location is planned, while considering the operation of dynamic obstacles that may appear on the path; finally, the stacking order of the goods and the layout of the goods on the shelves are determined.
[0013] Furthermore, in step (4), when the automatic stacking equipment is controlled to perform the stacking task, it first moves to the cargo location, accurately grabs the cargo through the gripping device, and then moves along the planned path to the designated shelf location, placing the cargo on the shelf according to the predetermined stacking order and layout. During the execution process, the perception module monitors the operating status and changes in the surrounding environment in real time. Once an abnormality is detected, the information is immediately fed back to the decision-making and planning module, which regenerates the decision instructions and adjusts the execution action to ensure the smooth completion of the task; the abnormal conditions include cargo falling and new obstacles appearing on the path.
[0014] The force sensor of the present invention is installed on the grabbing component of the execution module's mechanical arm to monitor the force applied when grabbing goods in real time to prevent damage or dropping of the goods. Distance sensors are used to detect the distance to surrounding objects and provide data support for path planning.
[0015] The large-scale model processing module of this invention is equipped with an advanced multimodal large-scale model trained on a vast amount of warehouse and logistics scenario data, including image data of different goods, layout data of various warehouse environments, and corresponding successful stacking cases and strategies. This large-scale model is able to integrate and deeply understand the multi-source data acquired by the perception module. For example, by analyzing image data collected by visual sensors, it can accurately identify the type, shape, and placement of goods. Combined with data from force sensors and distance sensors, it can comprehensively judge the physical characteristics of the goods and the surrounding environment.
[0016] The decision-making and planning module of the present invention generates a stacking task execution strategy based on the output of the large-scale model processing module. This includes determining the optimal location and method for grabbing goods, planning the path within the warehouse environment, and developing a stacking sequence and layout plan. During path planning, the impact of dynamic obstacles is fully considered, and the large-scale model is used to predict the movement trends of obstacles, allowing for real-time path adjustments to ensure safe and efficient completion of the stacking task. This decision-making and planning module also improves warehouse space utilization and facilitates subsequent cargo retrieval.
[0017] The execution module of the present invention comprises automated stacking equipment, specifically a mobile chassis, a robotic arm mounted on the chassis, and a gripping device mounted on the robotic arm. The execution module precisely executes cargo grabbing, handling, and stacking operations based on instructions generated by the decision-making and planning module. The robotic arm possesses high degrees of freedom and high-precision motion control capabilities, enabling flexible adaptation to diverse stacking requirements. The mobile chassis possesses excellent mobility and stability, enabling rapid and accurate movement to designated locations within a warehouse environment. The gripping device is designed to adapt to a variety of cargo shapes and sizes, achieving precise gripping through feedback from force sensors.
[0018] The data storage module of the present invention is used to store raw data collected by the perception module, intermediate processing data from the large model processing module, policy data generated by the decision-making and planning module, and task execution data from the execution module. On the one hand, this provides the large model processing module with rich training data, allowing it to continuously optimize model parameters and decision-making capabilities as it continuously learns new warehousing and logistics scenario data. On the other hand, it facilitates the system's retrospective analysis of historical task data. When problems such as task execution anomalies or inefficiencies arise, the relevant data in the data storage module can be retrieved to quickly locate the cause of the problem and perform targeted system optimization.
[0019] In the present invention, the large model processing module can identify the type and shape of the goods through image recognition technology, interpret the relevant information labels of the goods using the semantic understanding model, and judge the weight and stability of the goods in combination with the force sensor data.
[0020] In the present invention, the grasping method may include vacuum adsorption and mechanical clamping, but is not limited thereto.
[0021] After each palletizing task, the system records and analyzes relevant data (including perception data, decision data, and execution results). Successful cases and strategies are then fed into the larger model as new data samples for further training, continuously optimizing its performance and decision accuracy. Furthermore, any problems or failures are thoroughly analyzed to identify their causes and to target improvements to the system's parameter settings, algorithm logic, or hardware structure, thereby continuously improving the overall performance and adaptability of the automated palletizing system.
[0022] Compared with the prior art, the present invention has the following beneficial effects: The present invention effectively solves the problems of low intelligence and poor adaptability of traditional stacking, and significantly improves the efficiency and accuracy of warehousing and logistics operations.
[0023] (1) Possessing autonomous learning capabilities: Existing automatic stacking robots lack self-optimization mechanisms and require reprogramming when faced with new scenarios. This invention accumulates operational data through a data storage module and utilizes the autonomous learning capabilities of a large model. For every 1,000 operations completed, the system's decision-making accuracy can be improved by approximately 3%. After long-term operation, the system can continuously optimize stacking strategies, further improving operational efficiency by 15%-20% compared to the initial state, achieving continuous performance improvement.
[0024] (2) Reduced operating costs: Due to the improvement of operating efficiency and accuracy, as well as the reduction of the frequency of failures and manual intervention, this invention can effectively reduce the operating costs of warehousing and logistics. According to calculations, in a warehousing scenario with an annual cargo processing volume of 100,000 pieces, compared with traditional systems, it can save approximately 250,000 yuan in labor maintenance costs each year, reduce equipment maintenance costs by 30%, and reduce overall costs by approximately 22%. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 It is a schematic diagram of the architecture of the automatic stacking system based on a large model of the present invention; Figure 2 It is a flow chart of the automatic stacking method based on a large model of the present invention. DETAILED DESCRIPTION
[0026] The present invention is described in further detail below with reference to the embodiments.
[0027] Those skilled in the art will understand that the following examples are intended to illustrate the present invention only and should not be construed as limiting the scope of the present invention. Where specific techniques or conditions are not specified in the examples, the techniques or conditions described in the literature in the art or in the product specifications were used. Materials or equipment used without manufacturer identification are commercially available conventional products. Example 1
[0028] An automatic stacking system based on a large model, comprising a perception module, a large model processing module, a decision-making and planning module, an execution module, and a data storage module; The perception module is used to collect real-time image information of the goods, image information of the storage environment, information about the force used to grab the goods, and information about the distance between the goods and surrounding objects; The large model processing module is connected to the perception module and is used to identify the basic information of the goods and the storage environment information based on the information obtained by the large module and the perception module pre-stored therein; The decision-making and planning module is connected to the large model processing module and is used to generate a stacking task execution strategy based on the basic information of the goods and the storage environment information obtained by the large model processing module; The execution module is connected to the decision-making and planning module and is used to perform cargo grabbing, handling and stacking operations according to the stacking task execution strategy of the decision-making and planning module; The data storage module is connected to the perception module, large model processing module, decision-making and planning module, and execution module respectively. It is used to store data collected by the perception module, data identified by the large model processing module, strategy data generated by the decision-making and planning module, and task execution data of the execution module.
[0029] An automatic stacking method based on a large model, using the automatic stacking system based on a large model of this embodiment, includes the following steps: Step (1), data acquisition: the perception module collects real-time image information of the goods, image information of the storage environment, information on the force used to grab the goods, and information on the distance between the goods and surrounding objects, and transmits these data to the large model processing module; Step (2), data processing and analysis: The large model processing module pre-processes the received data, and then uses the pre-stored large model to deeply analyze and understand the data, thereby identifying the basic information of the acquired goods and the storage environment information; Step (3), decision making: the decision planning module makes the execution decision of the stacking task based on the recognition results of the large model processing module and generates the stacking task execution strategy; Step (4), task execution: the execution module controls the automatic stacking equipment to execute the stacking task according to the stacking task execution strategy generated by the decision-making and planning module; Step (5), learning and optimization: After completing each stacking task, the system records and analyzes the execution-related data of this task, including the data collected by the knowledge module, the data identified by the large model processing module, the strategy data generated by the decision-making planning module, and the task execution data of the execution module, and inputs them as new data samples into the large model of the large model processing module for further training, and continuously optimizes the performance of the large model; at the same time, for cases of unfinished stacking tasks, analyze them, find out the reasons, and make targeted improvements. Example 2
[0030] An automatic stacking system based on a large model, comprising a perception module, a large model processing module, a decision-making and planning module, an execution module, and a data storage module; The perception module is used to collect real-time image information of the goods, image information of the storage environment, information about the force used to grab the goods, and information about the distance between the goods and surrounding objects; The large model processing module is connected to the perception module and is used to identify the basic information of the goods and the storage environment information based on the information obtained by the large module and the perception module pre-stored therein; The decision-making and planning module is connected to the large model processing module and is used to generate a stacking task execution strategy based on the basic information of the goods and the storage environment information obtained by the large model processing module; The execution module is connected to the decision-making and planning module and is used to perform cargo grabbing, handling and stacking operations according to the stacking task execution strategy of the decision-making and planning module; The data storage module is connected to the perception module, large model processing module, decision-making and planning module, and execution module respectively. It is used to store data collected by the perception module, data identified by the large model processing module, strategy data generated by the decision-making and planning module, and task execution data of the execution module.
[0031] The perception module includes visual sensors, force sensors, and distance sensors; Visual sensors are used to obtain image information of goods and storage environments; The force sensor is installed on the grabbing component to monitor the force when grabbing the goods in real time; Distance sensors are used to detect the distance between goods and surrounding objects.
[0032] Vision sensors include cameras and 3D laser scanners.
[0033] The basic information of the goods includes the shape, size, color, weight, location, placement and label information of the goods; The storage environment information includes shelf layout, obstacle distribution and obstacle movement trends in the storage environment.
[0034] The stacking task execution strategy includes the optimal location and method for grabbing goods, planning the operation path in the storage environment, and formulating the stacking sequence and layout plan.
[0035] The pre-stored large module in the large model processing module uses the data stored in the data storage module as a training set and performs iterative training at regular intervals.
[0036] An automatic stacking method based on a large model, using the automatic stacking system based on a large model of this embodiment, includes the following steps: Step (1), data acquisition: the perception module collects real-time image information of the goods, image information of the storage environment, information on the force used to grab the goods, and information on the distance between the goods and surrounding objects, and transmits these data to the large model processing module; Step (2), data processing and analysis: The large model processing module pre-processes the received data, and then uses the pre-stored large model to deeply analyze and understand the data, thereby identifying the basic information of the acquired goods and the storage environment information; Step (3), decision making: the decision planning module makes the execution decision of the stacking task based on the recognition results of the large model processing module and generates the stacking task execution strategy; Step (4), task execution: the execution module controls the automatic stacking equipment to execute the stacking task according to the stacking task execution strategy generated by the decision-making and planning module; Step (5), learning and optimization: After completing each stacking task, the system records and analyzes the execution-related data of this task, including the data collected by the knowledge module, the data identified by the large model processing module, the strategy data generated by the decision-making planning module, and the task execution data of the execution module, and inputs them as new data samples into the large model of the large model processing module for further training, and continuously optimizes the performance of the large model; at the same time, for cases of unfinished stacking tasks, analyze them, find out the reasons, and make targeted improvements.
[0037] In step (2), the preprocessing includes removing noise and invalid data.
[0038] In step (3), when making the execution decision of the stacking task, the first step is to determine the grabbing strategy of the goods, including selecting the appropriate grabbing point, grabbing method and grabbing force; then plan the optimal path from the current location to the goods storage location, while considering the operation of dynamic obstacles that may appear on the path; finally, determine the stacking order of the goods and the layout of the goods on the shelves.
[0039] In step (4), when the automatic stacking equipment is controlled to perform the stacking task, it first moves to the cargo location, accurately grabs the cargo through the gripping device, and then moves along the planned path to the designated shelf location, placing the cargo on the shelf according to the predetermined stacking order and layout. During the execution process, the perception module monitors the operating status and changes in the surrounding environment in real time. Once an abnormality is detected, the information is immediately fed back to the decision-making and planning module. The decision-making and planning module regenerates the decision instructions and adjusts the execution action to ensure the smooth completion of the task; the abnormal conditions mentioned include cargo falling and new obstacles appearing on the path.
[0040] Application Examples The large-scale model-based automated stacking system of the present invention was deployed in a large logistics warehouse. This warehouse stores a wide variety of goods, including carton-packaged goods of various shapes, sizes, and weights, as well as irregularly shaped parts. Furthermore, the warehouse environment is subject to dynamic obstacles such as frequently moving forklifts and personnel.
[0041] During automated stacking, the perception module's visual sensors capture real-time images of the goods and surroundings. A 3D laser scanner captures three-dimensional spatial data of the goods and their surroundings. Force sensors and distance sensors also collect relevant data simultaneously and transmit this data rapidly to the large-scale model processing module. The large-scale model processing module analyzes the data using a trained multimodal large-scale model to accurately identify the type, shape, size, and placement of the goods, as well as the shelf layout, obstacle locations, and movement trends in the surrounding environment. The decision-making and planning module develops the optimal stacking strategy based on the large-scale model processing module's output. For example, for a rectangular cardboard box, the decision-making and planning module determines a mechanical grip from one corner of the item, setting a moderate grip force to ensure a secure grip without damaging the item. The planned route avoids moving forklifts and personnel, taking the shortest and safest route to the item's storage location. The module also determines where the item is to be placed on a specific shelf level to optimize shelf space utilization and product stability. The execution module follows the instructions generated by the decision-making and planning module, accurately grabbing the goods and moving them along the planned path to the shelf, completing the stacking operation. Throughout this process, the perception module continuously monitors the operating status and surrounding environmental changes. If it detects a worker approaching the operating path, it immediately feeds this information back to the decision-making and planning module, which quickly replans the route to avoid the worker and ensure operational safety.
[0042] Through long-term operation and data accumulation, the system continuously feeds new success stories and strategies into a large model for training and optimization, significantly improving the efficiency and accuracy of automated stacking operations at the warehouse center. Compared to traditional automated stacking systems, the system demonstrates greater adaptability and intelligence when handling complex cargo and environments, effectively improving the overall operational efficiency of warehouse logistics.
[0043] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.
Claims
1. An automatic stacking system based on a large model, characterized in that: It includes perception module, large model processing module, decision planning module, execution module and data storage module; The perception module is used to collect real-time image information of the goods, image information of the storage environment, information about the force used to grab the goods, and information about the distance between the goods and surrounding objects; The large model processing module is connected to the perception module and is used to identify the basic information of the goods and the storage environment information based on the information obtained by the large module and the perception module pre-stored therein; The decision-making and planning module is connected to the large model processing module and is used to generate a stacking task execution strategy based on the basic information of the goods and the storage environment information obtained by the large model processing module; The execution module is connected to the decision-making and planning module and is used to perform cargo grabbing, handling and stacking operations according to the stacking task execution strategy of the decision-making and planning module; The data storage module is connected to the perception module, large model processing module, decision-making and planning module, and execution module respectively. It is used to store data collected by the perception module, data identified by the large model processing module, strategy data generated by the decision-making and planning module, and task execution data of the execution module.
2. The large-scale model-based automatic stacking system according to claim 1, characterized in that: The perception module includes visual sensors, force sensors, and distance sensors; Visual sensors are used to obtain image information of goods and storage environments; The force sensor is installed on the grabbing component to monitor the force when grabbing the goods in real time; Distance sensors are used to detect the distance between goods and surrounding objects.
3. The large-scale model-based automatic stacking system according to claim 2, characterized in that: Vision sensors include cameras and 3D laser scanners.
4. The large-scale model-based automatic stacking system according to claim 1, characterized in that: The basic information of the goods includes the shape, size, color, weight, location, placement and label information of the goods; The storage environment information includes shelf layout, obstacle distribution and obstacle movement trends in the storage environment.
5. The large-scale model-based automatic stacking system according to claim 1, characterized in that: The stacking task execution strategy includes the optimal location and method for grabbing goods, planning the operation path in the storage environment, and formulating the stacking sequence and layout plan.
6. The large-scale model-based automatic stacking system according to claim 1, characterized in that: The pre-stored large module in the large model processing module uses the data stored in the data storage module as a training set and performs iterative training at regular intervals.
7. An automatic stacking method based on a large model, characterized in that: The large-scale model-based automatic stacking system according to any one of claims 1 to 6 comprises the following steps: Step (1), data acquisition: the perception module collects real-time image information of the goods, image information of the storage environment, information on the force used to grab the goods, and information on the distance between the goods and surrounding objects, and transmits these data to the large model processing module; Step (2), data processing and analysis: The large model processing module pre-processes the received data, and then uses the pre-stored large model to deeply analyze and understand the data, thereby identifying the basic information of the acquired goods and the storage environment information; Step (3), decision making: the decision planning module makes the execution decision of the stacking task based on the recognition results of the large model processing module and generates the stacking task execution strategy; Step (4), task execution: the execution module controls the automatic stacking equipment to execute the stacking task according to the stacking task execution strategy generated by the decision-making and planning module; Step (5), learning and optimization: After completing each stacking task, the system records and analyzes the execution-related data of this task, including the data collected by the knowledge module, the data identified by the large model processing module, the strategy data generated by the decision-making planning module, and the task execution data of the execution module, and inputs them as new data samples into the large model of the large model processing module for further training, and continuously optimizes the performance of the large model; at the same time, for cases of unfinished stacking tasks, analyze them, find out the reasons, and make targeted improvements.
8. The large-scale model-based automatic stacking method according to claim 7, characterized in that: In step (2), the preprocessing includes removing noise and invalid data.
9. The large-scale model-based automatic stacking method according to claim 7, characterized in that: In step (3), when making the execution decision of the stacking task, the first step is to determine the grabbing strategy of the goods, including selecting the appropriate grabbing point, grabbing method and grabbing force; then plan the optimal path from the current location to the goods storage location, while considering the operation of dynamic obstacles that may appear on the path; finally, determine the stacking order of the goods and the layout of the goods on the shelves.
10. The large-scale model-based automatic stacking method according to claim 7, characterized in that: In step (4), when the automatic stacking equipment is controlled to perform the stacking task, it first moves to the cargo location, accurately grabs the cargo through the gripping device, and then moves along the planned path to the designated shelf location, placing the cargo on the shelf according to the predetermined stacking order and layout. During the execution process, the perception module monitors the operating status and changes in the surrounding environment in real time. Once an abnormality is detected, the information is immediately fed back to the decision-making and planning module. The decision-making and planning module regenerates the decision instructions and adjusts the execution action to ensure the smooth completion of the task; the abnormal conditions mentioned include cargo falling and new obstacles appearing on the path.