CN-121979219-A - Task self-adaptive incremental cognitive map construction method and device for mobile operation robot, electronic equipment and storage medium

CN121979219ACN 121979219 ACN121979219 ACN 121979219ACN-121979219-A

Abstract

The application relates to the technical field of robot control, in particular to a method and a device for constructing a task self-adaptive incremental cognitive map of a mobile operation robot, electronic equipment and a storage medium. According to the method provided by the application, through fusing real-time multi-mode perception and semantic understanding, a rasterized semantic cognitive map fusing geometric, semantic and dynamic attributes is constructed and updated in an incremental mode, the attention weight of the attributes is dynamically adjusted based on the task types, and finally, closed-loop optimization is formed through task execution feedback, so that autonomous, reliable and efficient body-building intelligent operation of the robot in a complex dynamic scene is realized.

Inventors

WEI KUN
ZHANG JIANZHENG
ZOU JINPEI
DONG YI

Assignees

上海飒智智能科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260205

Claims (10)

1. The method for constructing the task self-adaptive incremental cognitive map of the mobile operation robot is characterized by comprising the following steps of: acquiring real-time multi-mode sensing data of an environment where the robot is located, and processing the sensing data to obtain an environment understanding result with a semantic tag; based on the environment understanding result and the real-time pose information of the robot, using grids as construction units, and adopting a confidence evaluation mechanism to carry out probabilistic dynamic update on each grid attribute so as to incrementally construct a semantic cognition map; According to the type of the task currently executed by the robot, dynamically adjusting the attention weight of each grid attribute in the semantic cognitive map, and generating a control instruction based on the semantic cognitive map after adjustment; and updating the semantic cognition map based on a task execution result obtained after the robot executes the control instruction and new environment perception data to obtain a global semantic cognition map.
2. The method of claim 1, wherein the step of processing the sensory data to obtain the context understanding result with semantic tags comprises: performing data preprocessing on the real-time multi-mode sensing data to obtain preprocessed sensing data; Extracting feature information related to the environment geometric structure, the object semantics and the dynamic change from the preprocessed perception data to obtain environment feature information; and carrying out semantic understanding on the environment characteristic information to obtain the environment understanding result with the semantic tag.
3. The method of claim 1, wherein the step of constructing the semantic cognitive map in an incremental manner based on the environmental understanding result and the real-time pose information by using grids as construction units and using a confidence evaluation mechanism to perform probabilistic dynamic update on each grid attribute comprises: dividing an environment space where the robot is located into a plurality of grids; Initializing geometric occupation attributes, semantic category attributes and dynamic attributes of the grids aiming at each grid to obtain a multidimensional state of the grids; Based on the environment understanding result and the real-time pose information, carrying out probabilistic updating on the multi-dimensional state of the grid through a confidence evaluation mechanism to obtain a target multi-dimensional state; Combining all grids and the corresponding target multidimensional states to construct a semantic cognition map; when the robot moves to an unobserved area, the semantic cognitive map is incrementally expanded to cover a new area, and initializing and probabilistic updating based on the confidence assessment mechanism is performed for grids in the new area.
4. The method of claim 3, wherein the step of probabilistically updating each attribute of the grid by a confidence assessment mechanism based on the environmental understanding result and the real-time pose information to obtain an updated multidimensional state further comprises: Judging whether attribute contradictions of the grids occur at a semantic level or a geometric level for each grid; If yes, based on attribute confidence, context semantic information or space consistency principle output by the confidence evaluation mechanism, the detected attribute contradiction is resolved to maintain consistency of the semantic cognitive map.
5. The method according to claim 1, wherein the step of probabilistically updating each attribute of the grid through a confidence evaluation mechanism based on the environmental understanding result and the real-time pose information to obtain an updated multidimensional state comprises the steps of updating geometric occupation attributes of the grid based on a sensor observation update model and fusing sensor observation data; Based on a multi-source clue evidence fusion strategy, fusing semantic recognition results and context information, and updating semantic category probability distribution of the grid; Based on time sequence analysis and motion modeling, the historical observation state and the dynamic object track information are fused to update the dynamic confidence coefficient of the grid; taking the space-time memory trace intensity and the prediction consistency score as evaluation factors, and carrying out comprehensive confidence evaluation on each attribute of the grid according to the confidence evaluation mechanism; The prediction consistency score is determined based on the consistency of the current state of the attribute and the prediction result of the object model of the current state to the future state and the actual observation result.
6. The method of claim 1, wherein the step of dynamically adjusting the attention weight for each grid attribute in the semantic cognitive map according to the type of task currently being performed by the robot and generating control instructions based on the adjusted map awareness comprises: Determining the attention weight of geometric occupation attributes, semantic category attributes and dynamic attributes of each grid in the semantic cognitive map based on the type of task currently executed by the robot; Carrying out weighted fusion on the grid attribute confidence coefficient in the semantic cognition map according to the attention weight to generate task-oriented map cognition; and carrying out path planning and operation decision-making based on the task-oriented map cognition, and generating a motion control instruction of the robot.
7. The method according to claim 1, wherein the step of updating the semantic cognitive map based on the task execution result obtained after the robot executes the generated control instruction and new environment awareness data to obtain a global semantic cognitive map comprises: Taking the task execution result and the new environment perception data as feedback information to generate new environment characteristic information and semantic understanding result; And using the feedback information, the new environmental characteristic information and the semantic understanding result to perform probabilistic incremental updating on the attribute of the related grid in the semantic cognitive map through the confidence evaluation mechanism to obtain the global semantic cognitive map.
8. A mobile manipulator robot task adaptive incremental cognitive map building device, the device comprising: the acquisition module is used for acquiring real-time multi-mode sensing data of the environment where the robot is located and processing the sensing data to obtain an environment understanding result with a semantic tag; The construction module is used for carrying out probabilistic dynamic update on each grid attribute by taking the grids as construction units and adopting a confidence evaluation mechanism based on the environment understanding result and the real-time pose information of the robot so as to incrementally construct a semantic cognition map; the generation module is used for dynamically adjusting the attention weight of each grid attribute in the semantic cognitive map according to the type of the task currently executed by the robot and generating a control instruction based on the semantic cognitive map after adjustment; and the updating module is used for updating the semantic cognition map based on a task execution result obtained after the robot executes the control instruction and new environment perception data to obtain a global semantic cognition map.
9. An electronic device comprising a memory for storing a computer program and a processor that runs the computer program to cause the electronic device to perform the method of any one of claims 1 to 7.
10. A storage medium having stored therein computer program instructions which, when read and executed by a processor, perform the method of any of claims 1 to 7.

Description

Task self-adaptive incremental cognitive map construction method and device for mobile operation robot, electronic equipment and storage medium Technical Field The invention relates to the technical field of robot control, in particular to a method and a device for constructing a task self-adaptive incremental cognitive map of a mobile operation robot, electronic equipment and a storage medium. Background The traditional mobile operation robot is mainly based on synchronous positioning and map construction (SLAM) technology, an accurate geometric map is constructed in advance in a structured or semi-structured industrial environment (such as a fixed production line and a warehouse system), and task execution is realized through a preset program or offline semantic annotation. The method depends on a priori models of static environments, focuses on the integrity and consistency of geometric information, and can achieve higher positioning and navigation accuracy in known environments. However, the mapping process is usually one-time or batch update, and lacks real-time response capability to environment dynamic changes and semantic information. Along with the expansion of the application scene of the robot to a dynamic, open and unstructured industrial environment, the traditional method faces the limitation that on one hand, the existing incremental map updating focuses on local correction of a geometric layer, semantic understanding and dynamic perception cannot be fused deeply, so that the map lacks high-level environment cognition, and on the other hand, most semantic map construction methods depend on a large amount of offline training data and closed class definitions, and are difficult to adapt to the actual situations of various object classes and frequent state changes in the industrial field. In addition, the existing system still has defects in the aspects of dynamic obstacle processing, environment change self-adaption, task-oriented map optimization and the like, and the autonomous operation capability of the robot in a complex scene is restricted. Disclosure of Invention In view of the above, the present invention aims to provide a method and a device for constructing a task adaptive incremental cognitive map of a mobile operation robot, an electronic device and a storage medium, so as to realize autonomous, reliable and efficient self-intelligent operation of the robot in a complex dynamic scene. In a first aspect, an embodiment of the present invention provides a method for constructing a task adaptive incremental cognitive map of a mobile operation robot, where the method includes: acquiring real-time multi-mode sensing data of the environment where the robot is located, and processing the sensing data to obtain an environment understanding result with a semantic tag; Based on the environment understanding result and the real-time pose information of the robot, using grids as construction units and adopting a confidence evaluation mechanism to carry out probabilistic dynamic update on the grid attributes so as to maintain incremental construction of a semantic cognitive map; According to the type of the task currently executed by the robot, dynamically adjusting the attention weight of each grid attribute in the semantic cognitive map, and generating a control instruction based on the adjusted semantic cognitive map; And updating the semantic cognition map based on a task execution result obtained after the robot responds to the execution control instruction and the current new environment perception data to obtain a global semantic cognition map. With reference to the first aspect, the step of processing the perception data to obtain an environmental understanding result with a semantic tag includes: Performing data preprocessing on the real-time multi-mode sensing data to obtain preprocessed sensing data; Extracting feature information related to the environment geometric structure, the object semantics and the dynamic change from the preprocessed perception data to obtain environment feature information; and carrying out semantic understanding on the environment characteristic information to obtain an environment understanding result with semantic tags. In combination with the first aspect, based on the environmental understanding result and the real-time pose information, the step of using the grids as the construction units and adopting the confidence evaluation mechanism to carry out probabilistic dynamic update on the grid attributes to construct the semantic cognitive map in an incremental manner comprises the following steps: Dividing an environment space where the robot is located into a plurality of grids; initializing geometric occupation attributes, semantic category attributes and dynamic attributes of the grids aiming at each grid to obtain a multidimensional state of the grids; Based on the environment understanding result and the real-time pose information, carrying out proba