CN-122008232-A - Multi-mode causal reasoning-based multi-robot control method, device, electronic equipment and computer program product

CN122008232ACN 122008232 ACN122008232 ACN 122008232ACN-122008232-A

Abstract

The application is applicable to the field of robot control, and provides a multi-mode causal reasoning-based multi-robot control method, a multi-mode causal reasoning-based multi-robot control device, electronic equipment and a computer program product, wherein the method comprises the steps of acquiring running state data of a plurality of service dimensions of a plurality of robots and global target quantization data corresponding to a global target; the method comprises the steps of calculating causal influence intensity between running state data of each service dimension of each robot and global target quantized data, screening target running state data of which causal influence intensity meets the set causal influence intensity from the running state data of a plurality of service dimensions of each robot, performing action planning based on the target running state data of the robot and the global target quantized data to obtain local action sequences of the robots, generating action execution instructions of the robots based on the local action sequences, and sending the action execution instructions to the robots. The scheme can improve the suitability of the local action and the global target of the single robot.

Inventors

ZHOU CHUANCHENG
CHEN TAO
LUO QICE
Huang Ganyao
FANG SHUIBO
CHEN ZHILIE

Assignees

广东省工业边缘智能创新中心有限公司

Dates

Publication Date: 20260512
Application Date: 20260324

Claims (10)

1. The multi-mode causal reasoning-based multi-robot control method is characterized by comprising the following steps of: Acquiring running state data of a plurality of service dimensions of each robot and global target quantization data corresponding to a global target, wherein the running state data is planning influence data of local actions; Calculating causal influence intensity between the running state data of each service dimension of each robot and the global target quantized data, wherein the causal influence intensity characterizes the influence degree of the running state data on the global target; Screening out target running state data of which the causal influence intensity meets the set causal influence intensity from the running state data of a plurality of service dimensions of each robot; Performing motion planning based on the target running state data and the global target quantized data of a plurality of robots to obtain local motion sequences of the robots, wherein the local motion sequences comprise at least one local motion which is matched with the global target; and generating an action execution instruction of each robot based on the local action sequence of the robot, and sending the action execution instruction to the robot.
2. The method of claim 1, wherein the operational status data for each business dimension includes operational status sub-data for at least one modality, the calculating causal impact strength between the operational status data and the global target quantization data for each business dimension for each robot comprising: In case the operating state data comprises the operating state sub-data of one modality, calculating the causal influence intensity between the operating state sub-data and the global target quantized data as the causal influence intensity between the operating state data and the global target quantized data, or Under the condition that the running state data comprises running state sub-data of multiple modes, carrying out multi-mode feature fusion processing on the multiple running state sub-data to obtain fusion state data; The causal impact strength between the fusion status data and the global target quantized data is calculated as the causal impact strength between the operational status data and the global target quantized data.
3. The method according to claim 2, wherein the performing multi-modal feature fusion processing on the plurality of operation state sub-data to obtain fusion state data includes: performing cross-modal alignment processing on the plurality of running state sub-data to obtain a plurality of cross-modal alignment data; Based on the global target quantized data, respectively calculating modal influence weights corresponding to the cross-modal alignment data through a cross-modal attention mechanism; And carrying out weighted calculation based on the cross-modal alignment data and the modal influence weights corresponding to the cross-modal alignment data to obtain the fusion state data.
4. The method of claim 1, wherein the screening the target operational state data for which the causal effect intensity satisfies a set causal effect intensity from the operational state data for a plurality of business dimensions for each of the robots further comprises: Taking the running state data of each service dimension of each robot as a local data node, taking the global target quantized data as a global target node, and constructing a causal edge between nodes based on causal influence intensity between the running state data and the global target quantized data to obtain a target causal graph comprising the local data node, the global target node and the causal edge; the screening the target operation state data of which the causal influence intensity meets the set causal influence intensity from the operation state data of a plurality of service dimensions of each robot comprises the following steps: Screening out a target causal edge of which the causal influence intensity meets the set causal influence intensity from the target causal graph; And determining the running state data corresponding to the target local data node associated with the target causal edge as the target running state data.
5. The method according to claim 1, wherein the performing motion planning based on the target operation state data and the global target quantization data of the plurality of robots, to obtain a local motion sequence of each robot, includes: Vectorizing and combining the target running state data of each robot to obtain an initial feature vector of the robot; performing dimension reduction processing on each initial feature vector to obtain a low-dimensional feature vector with set feature dimensions; Inputting the low-dimensional feature vectors, the global target quantized data and the task context data of a plurality of robots into a target motion planning engine to obtain at least one candidate motion planning scheme output by the target motion planning engine, wherein each candidate motion planning scheme comprises candidate motion sequences corresponding to the plurality of robots; And determining a target action planning scheme with global target achievement quantization data closest to the global target quantization data from at least one candidate action planning scheme, wherein the target action planning scheme comprises local action sequences corresponding to the robots.
6. The method of claim 5, wherein said inputting the low-dimensional feature vectors, the global target quantization data, and the task context data for a plurality of the robots into a target motion planning engine, before deriving at least one candidate motion planning solution output by the target motion planning engine, further comprises: detecting potential disturbance factors of the target running state data of a plurality of robots; if no potential disturbance factors exist, determining a first action planning engine as the target action planning engine; if the potential disturbance factors exist, determining a second action planning engine as the target action planning engine, or determining the first action planning engine and the second action planning engine as the target action planning engine; The planning logic of the first action planning engine is a non-degradation reference planning, the planning logic of the second action planning engine is a degradation planning, and a degradation strategy library is embedded in the second action planning engine.
7. The method of claim 6, wherein the first action plan engine outputs a first candidate action plan and the second action plan engine outputs at least one second candidate action plan, wherein the determining a target action plan for which global target achievement quantization data is closest to the global target quantization data from at least one of the candidate action plan comprises: determining the first candidate action plan as the target action plan in the case that the candidate action plan only includes the first candidate action plan; Determining, from at least one of the second candidate action plan, the target action plan for which the global target achievement quantization data is closest to the global target quantization data, by an action utility evaluation function associated with the global target, in the case that the candidate action plan includes only at least one of the second candidate action plan; in the case that the candidate action plan includes the first candidate action plan and at least one of the second candidate action plans, determining, by the action utility evaluation function, the target action plan for which the global target achievement quantization data is closest to the global target quantization data from the first candidate action plan and at least one of the second candidate action plans.
8. A multi-modal causal reasoning based multi-robot control device, comprising: The system comprises an acquisition module, a calculation module and a calculation module, wherein the acquisition module is used for acquiring running state data of a plurality of service dimensions of a plurality of robots and global target quantization data corresponding to a global target, wherein the running state data is planning influence data of local actions; The computing module is used for computing causal influence intensity between the running state data of each service dimension of each robot and the global target quantification data, wherein the causal influence intensity characterizes the influence degree of the running state data on the global target; The screening module is used for screening out target running state data of which the causal influence intensity meets the set causal influence intensity from the running state data of a plurality of service dimensions of each robot; The planning module is used for performing action planning based on the target running state data and the global target quantized data of a plurality of robots to obtain local action sequences of the robots, wherein the local action sequences comprise at least one local action which is matched with the global target; and the generation and transmission module is used for generating an action execution instruction of each robot based on the local action sequence of each robot and transmitting the action execution instruction to the robot.
9. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, which when executed causes the electronic device to implement the method of any one of claims 1 to 7.
10. A computer program product comprising a computer program which, when run, causes the method of any one of claims 1 to 7 to be performed.

Description

Multi-mode causal reasoning-based multi-robot control method, device, electronic equipment and computer program product Technical Field The application belongs to the field of robot control, and particularly relates to a multi-mode causal reasoning-based multi-robot control method, a multi-mode causal reasoning-based multi-robot control device, electronic equipment and a computer program product. Background With the development of intelligent manufacturing, multiple robots are commonly used in the industry to cooperatively process complex tasks, such as material handling, component assembly, order performance, etc. In order to realize ordered cooperation of multiple robots, the prior art provides control methods such as hierarchical architecture scheduling and preset rule driving, however, the methods generally have the following problems that causal quantitative data source modes are single, causal quantitative association between global targets of multiple robots and local actions of single robots is lacking, and the local actions of the single robots are difficult to accurately adapt to the overall targets. Disclosure of Invention The embodiment of the application provides a multi-robot control method, a multi-mode causal reasoning-based multi-robot control device, electronic equipment and a computer program product, which are used for solving the problems that in the prior art, causal quantitative data sources are single in mode, causal quantitative association is lacking between a global target of a plurality of robots and local actions of a single robot, and the local actions of the single robot are difficult to accurately adapt to a total target. A first aspect of an embodiment of the present application provides a multi-robot control method based on multi-modal causal reasoning, including: Acquiring running state data of a plurality of service dimensions of each robot and global target quantization data corresponding to a global target, wherein the running state data is planning influence data of local actions; Calculating causal influence intensity between the running state data of each service dimension of each robot and the global target quantized data, wherein the causal influence intensity characterizes the influence degree of the running state data on the global target; Screening out target running state data of which the causal influence intensity meets the set causal influence intensity from the running state data of a plurality of service dimensions of each robot; Performing motion planning based on the target running state data and the global target quantized data of a plurality of robots to obtain local motion sequences of the robots, wherein the local motion sequences comprise at least one local motion which is matched with the global target; and generating an action execution instruction of each robot based on the local action sequence of the robot, and sending the action execution instruction to the robot. A second aspect of an embodiment of the present application provides a multi-robot control device based on multi-modal causal reasoning, including: The system comprises an acquisition module, a calculation module and a calculation module, wherein the acquisition module is used for acquiring running state data of a plurality of service dimensions of a plurality of robots and global target quantization data corresponding to a global target, wherein the running state data is planning influence data of local actions; The computing module is used for computing causal influence intensity between the running state data of each service dimension of each robot and the global target quantification data, wherein the causal influence intensity characterizes the influence degree of the running state data on the global target; The screening module is used for screening out target running state data of which the causal influence intensity meets the set causal influence intensity from the running state data of a plurality of service dimensions of each robot; The planning module is used for performing action planning based on the target running state data and the global target quantized data of a plurality of robots to obtain local action sequences of the robots, wherein the local action sequences comprise at least one local action which is matched with the global target; and the generation and transmission module is used for generating an action execution instruction of each robot based on the local action sequence of each robot and transmitting the action execution instruction to the robot. A third aspect of an embodiment of the application provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of the method according to the first aspect when executing the computer program. A fourth aspect of embodiments of the application provides a computer program product comprising a compute