A collaborative large model decision method and device for a complex dynamic environment

By aligning multi-source heterogeneous data through contrastive learning and cross-modal attention mechanisms, and combining adversarial incremental learning and thought chain reasoning of large language models, the problem of data fusion and decision-making for UAVs in complex environments is solved, enabling efficient and safe collaborative operation of UAV swarms.

CN122884137APending Publication Date: 2026-10-09HEFEI DUNYI TECHNOLOGY CO LTD
0 Cites 0 Cited by

Patent Information

Application Number
CN202611053450.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-15
Publication Date
2026-10-09

AI Technical Summary

Technical Problem

In complex and dynamic environments, UAVs struggle to achieve efficient and secure fusion and decision-making of multi-source heterogeneous data. Traditional methods have limitations in terms of rapid convergence and multi-task adaptability accuracy, and their collaborative effects are difficult to guarantee when faced with sudden interference.

Method used

We employ contrastive learning and cross-modal attention mechanisms for data alignment and fusion, combined with adversarial incremental learning and thought chain reasoning from a large language model, to construct short-term and long-term storage systems, thereby achieving global situational representation and decision optimization.

Benefits of technology

It improves the perception accuracy and dynamic adaptability of UAV swarms in complex environments, shortens strategy generation time, enhances the interpretability and security of the decision-making process, and improves collaborative efficiency.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The application discloses a complex dynamic environment collaborative large model decision method, comprising: based on contrast learning and cross-modal attention mechanism, cross-modal alignment and deep fusion are carried out on multi-source heterogeneous sensing data received by a UAV to obtain global situation representation; then the global situation representation is input into a short-term and long-term double-layer storage system constructed based on an adversarial incremental learning; and based on a thinking chain reasoning process of a large language model, a fusion global situation feature is converted into a reasoning chain and a corresponding output probability is determined in combination with a current multi-modal sensing state and a task instruction, so that a decision result is determined based on the output probability. The implementation of the application significantly improves the sensing accuracy, continuous learning ability and strategy interpretability of the UAV system under extreme working conditions, and meets the needs of safe and efficient collaborative operation of the unmanned cluster in a complex environment.
Need to check novelty before this filing date? Find Prior Art