Intelligent Adaptive Decision-Making for Autonomous Vehicles: A Learning-Enhanced Game-Theoretic Approach in Interactive Scenarios

Huang, Heye; Zheng, Xunjia; Liu, Yicong; Zhao, Shiyue; Wang, Yuning; Wang, Jianqiang · 2023 · Unknown

DOI: 10.1109/dsins60115.2023.10455677

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of adaptive behavioral decision-making for autonomous vehicles (AVs) in complex, dynamic merging scenarios. The primary motivation is to enable AVs to make rational, safe, and human-like decisions amidst stochastic interactions with surrounding vehicles, a task complicated by uncertain motion statuses and unpredictable interaction durations. Existing methods, including learning-based, probabilistic, potential field, and game-theoretic approaches, often suffer from poor adaptability to dynamic environments or difficulties in optimizing multiple model parameters. To overcome these limitations, the authors propose a learning-enhanced game-theoretic framework that combines offline modeling with online parameter adaptation. The methodology employs non-cooperative game theory to model vehicle interactions, defining key elements such as participants, action spaces (yielding vs. non-yielding), and reward functions. The reward functions incorporate safety and efficiency metrics, including predicted distances and speeds, weighted by specific factors. To optimize these weight factors, the authors utilize Maximum Entropy Inverse Reinforcement Learning (IRL). This process involves extracting interaction behavior feature vectors from demonstrative behaviors and maximizing distribution entropy to align the model’s expected features with empirical data. The framework further establishes a mapping model between environmental variables and model parameters, allowing the AV to identify optimal matching parameters in real-time through online learning. The proposed method was validated using naturalistic driving datasets (highD and exiD) and real-vehicle test data. In 188 tested interaction scenarios, the model achieved an average human-like similarity rate of 81.73%, with a specific rate of 83.12% in the highD dataset. In 145 dynamic interactions, the method matched human decisions in 77.12% of cases, resulting in 6,913 consistent instances. Real-vehicle tests demonstrated a 72.73% similarity to human driving behavior with zero safety violations. These results indicate that the model effectively captures the nuances of interactive driving behaviors and adapts to varying traffic conditions. The significance of this work lies in its ability to bridge the gap between static game-theoretic models and dynamic real-world environments. By integrating Maximum Entropy IRL for parameter optimization and establishing a mapping between environmental variables and model parameters, the framework enables AVs to perform adaptive, human-logic-consistent decision-making. This approach addresses the rigidity of traditional game-theoretic models and the interpretability issues of deep learning methods, offering a robust solution for safe and efficient autonomous driving in interactive scenarios.

Key finding

The proposed learning-enhanced game-theoretic approach achieves high consistency with human driving behavior in interactive scenarios, demonstrating an average human-like similarity rate of 81.73% in dataset evaluations and 72.73% in real-vehicle tests without safety violations.

Methodology

mixed_methods

Sample size: 333

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via author_sweep_intake on 2026-05-28.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• situational awareness