AGENTSCoDRIVER is a novel framework that leverages large language models (LLMs) to enable multi-vehicle collaborative driving with lifelong learning. The framework consists of five modules: observation module, reasoning engine, cognitive memory module, reinforcement reflection module, and communication module. These modules work together to allow vehicles to collaborate, negotiate, and adapt to new environments over time. The cognitive memory module stores and recalls past experiences, while the reinforcement reflection module evaluates and improves decisions. The communication module enables vehicles to exchange information and coordinate actions. The framework is designed to address the limitations of traditional autonomous driving systems, such as lack of interpretability, generalization, and continuous learning. Extensive experiments show that AGENTSCoDRIVER outperforms existing approaches in terms of lifelong learning, collaboration, and negotiation. The framework is capable of handling complex driving scenarios, including highway and intersection scenarios, and demonstrates improved performance with more memory items and communication. However, the framework has limitations, such as not being real-time and lacking the ability to process visual information. Future work includes extending the framework to multi-modal collaborative driving and addressing latency issues when applying LLMs to collaborative driving.AGENTSCoDRIVER is a novel framework that leverages large language models (LLMs) to enable multi-vehicle collaborative driving with lifelong learning. The framework consists of five modules: observation module, reasoning engine, cognitive memory module, reinforcement reflection module, and communication module. These modules work together to allow vehicles to collaborate, negotiate, and adapt to new environments over time. The cognitive memory module stores and recalls past experiences, while the reinforcement reflection module evaluates and improves decisions. The communication module enables vehicles to exchange information and coordinate actions. The framework is designed to address the limitations of traditional autonomous driving systems, such as lack of interpretability, generalization, and continuous learning. Extensive experiments show that AGENTSCoDRIVER outperforms existing approaches in terms of lifelong learning, collaboration, and negotiation. The framework is capable of handling complex driving scenarios, including highway and intersection scenarios, and demonstrates improved performance with more memory items and communication. However, the framework has limitations, such as not being real-time and lacking the ability to process visual information. Future work includes extending the framework to multi-modal collaborative driving and addressing latency issues when applying LLMs to collaborative driving.