Learning-Based Hierarchical Decision-Making Framework for Automatic Driving in Incompletely Connected Traffic...

Area of Science:

• Robotics
• Computer Science
• Artificial Intelligence

Background:

• End-to-end decision-making algorithms offer strong data processing but suffer from uncertainty.
• Existing learning-based methods require ideal state information, limiting their real-world autonomous driving applicability.
• Autonomous driving in complex environments necessitates robust decision-making with incomplete information.

Purpose of the Study:

• To propose a stable hierarchical decision-making framework for autonomous driving systems.
• To address the limitations of existing algorithms in handling real-world, incomplete information.
• To enable reliable autonomous driving using only image input.

Main Methods:

• A model-based data encoder transforms input images into a universal data format.
• A state machine utilizing a time series Graph Convolutional Network (GCN) classifies driving states.
• Rule-based algorithms are selected for action generation based on classified driving states.

Main Results:

• The proposed framework successfully performs autonomous driving tasks across diverse traffic scenarios.
• The system operates effectively without requiring global network information.
• Comparative experiments validate the efficacy of the hierarchical framework, image data encoder, and time series GCN.

Conclusions:

• The hierarchical decision-making framework provides a stable and effective solution for autonomous driving.
• The model-based image encoder and time series GCN are crucial components for robust state classification and decision-making.
• This approach enhances the reliability of autonomous vehicles in real-world conditions.