LEAD: Learning-Enhanced Adaptive Decision-Making for Autonomous Driving in Dynamic Environments
DOI: 10.1109/tits.2025.3531293
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. Existing methods often struggle with the stochastic nature of human-vehicle interactions and the difficulty of optimizing model parameters for varying environmental conditions. To resolve this, the authors propose LEAD (Learning-Enhanced Adaptive Decision-Making), a framework that combines non-cooperative game theory with maximum entropy inverse reinforcement learning (IRL). The goal is to enable AVs to make rational, safe, and human-like decisions by adapting model parameters in real-time based on environmental variables. The methodology involves a five-step process. First, key environmental variables are extracted from interaction datasets. Second, interactive behaviors are calibrated using ramp vehicle lane-change data. Third, a vehicle interaction behavior model is developed using non-cooperative game theory, defining participants, action spaces (yielding vs. non-yielding), and reward functions that account for safety, efficiency, and traffic constraints. Fourth, model parameters (weight factors in the reward functions) are optimized offline using Maximum Entropy IRL, which aligns the model’s expected features with expert demonstrations. Finally, an adaptive decision-making method is implemented by establishing a mapping model between environmental variables and optimized parameters, allowing for online parameter matching and real-time generation of interactive behavior probabilities. The proposed method was validated using naturalistic driving datasets (highD and exiD) and real-vehicle tests. 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, comprising 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 framework effectively captures the nuances of interactive driving behaviors and adapts to dynamic conditions. The significance of this work lies in its ability to overcome the rigidity of traditional game-theoretic models, which typically require fixed parameters. By integrating offline learning of behavioral characteristics with online identification of optimal parameters, LEAD enables AVs to adapt to diverse traffic situations while maintaining consistency with human logic. This approach enhances the safety and efficiency of autonomous driving in interactive environments, providing a robust solution for handling uncertainties in dynamic traffic.
Key finding
The proposed learning-enhanced adaptive decision-making method achieves high consistency with human driving behavior, demonstrating an average human-like similarity rate of 81.73% in dataset evaluations and 72.73% in real-vehicle tests with no safety violations.
Methodology
mixed_methods
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.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-28 |
| archive | success | canonical_url | — | — | 1 | 2026-06-06 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | success | — | — | — | 1 | 2026-05-28 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 15 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
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.
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Theoretical Contribution: computational model