A Scoring Framework for Mode Selection Between High-Accuracy and High-Speed Fusion in Autonomous Driving

In autonomous driving, sensor fusion typically operates under two distinct modes. The high-accuracy fusion mode offers more precise environmental perception, but it requires substantial computational resources and introduces higher latency. In contrast, the high-speed fusion mode responds more quickly, albeit at the expense of some accuracy in specific situations. Balancing these two modes is a central challenge in designing perception fusion systems. Most existing studies focus on improving the fusion algorithms themselves, while relatively few explore a quantitative decision mechanism that selects between fusion modes based on driving scenarios. This paper proposes a mode-selection framework that enables dynamic switching between high-accuracy and high-speed fusion modes based on a multi-metric scoring scheme. The framework takes accuracy, stability, and latency as its core evaluation indicators. By applying normalization and weighted averaging, we establish a unified scoring metric that allows for flexible adjustment of weights according to real-world driving conditions. Using MATLAB simulations, we compare the two fusion modes under different scenarios and weighting configurations. The results show that the proposed scoring framework provides a quantitative and interpretable standard for balancing accuracy and speed, and it can automatically select the fusion mode with the higher score. This study offers a standardized and quantifiable decision-making approach for autonomous driving perception systems. It has the potential to enhance system safety while reducing unnecessary computational overhead.