Reproduce “PADriver: Towards Personalized Autonomous Driving” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Genghua Kou, Fan Jia, Weixin Mao, Yingfei Liu, Yucheng Zhao, Ziheng Zhang, Osamu Yoshie, Tiancai Wang and 2 more
PADriver introduces a Multi-modal Large Language Model (MLLM) framework for personalized autonomous driving, enabling adaptation to user preferences through natural language prompts, such as switching between 'slow,' 'normal,' and 'fast' driving modes. The system leverages explicit danger level modeling and a chain-of-thought reasoning process, demonstrating distinct behavioral patterns and improved computational efficiency compared to existing LLM-based driving methods.
Key insights
- Explicit Danger Level Modeling: The system assigns numerical risk scores to potential actions, providing quantitative references for decision-making and allowing adaptable risk tolerance.
- Mode-Specific Behavior Patterns: Distinct driving modes (slow, normal, fast) are achieved, each exhibiting unique trade-offs in speed, safety, and comfort based on user prompts.
- Chain-of-Thought Reasoning: A structured output generation process (Description Danger Level Action) enhances interpretability and reduces hallucination by mimicking human cognitive decision-making.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.