Article Open Access http://dx.doi.org/10.26855/acc.2025.04.005
Research on Intelligent Decision-making Methods for Complex Scenarios Based on Multimodal Large Models
Lili Song
Shenzhen Nuoyi Fluid Technology Co., LTD., Shenzhen 518107, Guangdong, China.
*Corresponding author: Lili Song
Published: May 13,2025
Abstract
With the gradual development of technology, traditional intelligent decision algorithms face major challenges in a complex environment in which different data from multiple sources, great uncertainty, and dynamics coexist. This paper deals with the powerful capabilities of multimodal large models (MLMs) in the processing of multimodal data such as text, images, audio, etc. First, the basic nature of MLMs and the characteristics of complex scenarios are discussed, and existing MLMs-based decision-making training methods and their limitations regarding data generalization and interpretation are also discussed. At the same time, a new framework is proposed that contains three key components: first, it uses pre-treatment techniques such as noise reduction and intermodal alignment to solve the noise problem when multiple modes of transport are used; second, it combines early and late coupling with attention mechanisms to achieve mixed low-level signal Fusion. trait correlation and high-level semantic interdependence; third, it uses advanced learning optimization strategies to adapt to dynamic environmental feedback. The study used the nuScenes autonomous vehicle dataset and data from the field of Medicine and wellness mimic-3 for evaluation. The accuracy rate of decision-making in autonomous vehicles reached 92.5% and in medical services 89.3%, which is 4-5% higher than in the conventional method of combining data. This study demonstrates the importance of adaptive modal weighing and iterative strategy optimization, as well as that the data privacy model has urgent issues that require a solution, such as improved interpretability and extensibility. This is relevant to study.
References
[1] Guo C. Research on multi-modal trajectory prediction method for intelligent vehicles based on deep inverse reinforcement learning [dissertation]. Shaanxi: Chang'an University; 2023.
[2] Song D. Research on intelligent vehicle anthropomorphic decision-making and planning strategy based on traffic situation cognition [dissertation]. Jilin University; 2024.
[3] Fu W. Research on intelligent decision making of autonomous driving based on multimodal fusion [dissertation]. Heilongjiang: Harbin Institute of Technology; 2023.
[4] Wang T. Research on multi-modal trajectory generation method based on flow model [dissertation]. Liaoning: Dalian Maritime University; 2022.
[5] Xu Y. Research on multi-modal adaptive gait generation method for bipedal robot based on deep reinforcement learning [dissertation]. Shanghai: Shanghai Normal University; 2024.
[6] Wang X. Research on multi-modal fusion 3D multi-target tracking method for intelligent vehicle environment perception [dissertation]. Chongqing: Chongqing University; 2023.
[7] Zhong Z. Research on decision planning method of autonomous vehicles based on dynamic interaction [dissertation]. Guangdong: South China University of Technology; 2023.
[8] Luo F. Research on multi-modal trajectory prediction method based on attention mechanism [dissertation]. Shandong Institute of Business and Technology; 2024.
[9] Cheng H. Research on multi-modal disaster information detection method based on deep learning [dissertation]. Qilu University of Technology; 2023.
[10] Deng W. Research on autonomous navigation system of mobile robot based on multimodal perception fusion in indoor environment [dissertation]. Shanghai: Shanghai Normal University; 2024.
How to cite this paper
Research on Intelligent Decision-making Methods for Complex Scenarios Based on Multimodal Large Models
How to cite this paper: Lili Song. (2025) Research on Intelligent Decision-making Methods for Complex Scenarios Based on Multimodal Large Models. Advances in Computer and Communication, 6(2), 81-86.
DOI: http://dx.doi.org/10.26855/acc.2025.04.005