DualAD is an autonomous driving framework that integrates reasoning capabilities with traditional planning modules to handle complex driving scenarios. By leveraging a dual-layer system, DualAD combines rule-based planning with large language models (LLMs) for more human-like cognitive reasoning in critical situations.
DualAD aims to imitate human cognitive processes in autonomous driving by separating planning into two layers:
- Lower Layer – Manages basic driving tasks through a rule-based motion planner.
- Upper Layer – Acts as a reasoning module using LLMs to assess potential dangers and make real-time adjustments in critical scenarios.
- Dual-Layer Architecture: Combines routine driving tasks with high-level reasoning for safety and efficiency.
- Text Encoding: Converts driving scenarios into textual descriptions, enhancing the LLM’s understanding.
- Closed-Loop Simulations: Validates performance in realistic, complex environments, outperforming standard planners.
DualAD demonstrates improved performance in challenging scenarios compared to other planners. Key metrics such as Closed-Loop Score (CLS) and Reactive Closed-Loop Score (R-CLS) showcase DualAD’s ability to outperform rule-based and learning-based models in terms of safety and decision quality.
| Planner | Hard-55 CLS | Super-Hard-24 CLS |
|---|---|---|
| IDM | 50.12 | 34.56 |
| Lattice-IDM | 52.36 | 39.76 |
| DualAD (Lattice-IDM) | 60.25 | 57.31 |
- Python 3.8+
- NuPlan Simulator (for closed-loop simulation)
- Clone the repository:
git clone https://github.com/username/DualAD.git cd DualAD - Install the dependencies:
pip install -r requirements.txt