CoLMIN: LLM-based Multi-Decision Path Negotiation for Cooperative Autonomous Driving

CoLMIN proposes a collaborative autonomous driving multi-decision path negotiation framework based on large language models, aiming to achieve stable decision-making consensus through multi-decision path negotiation and reflective reasoning. The framework consists of three core components: (i) the large language model-based multi-intention negotiation module (LMin), which generates candidate driving intentions using the negotiator-evaluator paradigm; (ii) the evaluation-based shallow reflection module (ESRM), which analyzes negotiation results and provides feedback to accelerate consensus formation; (iii) the large language model-based deep reflection module (LDRM), which conducts long-term reflection on negotiation history to mitigate cognitive rigidity. Experiments in the CARLA simulation environment show that CoLMIN outperforms existing methods in challenging interactive driving scenarios.