**DriveCoT: Integrating Chain-of-Thought Reasoning with End-to-End Driving** This paper introduces DriveCoT, a comprehensive end-to-end driving dataset that includes chain-of-thought (CoT) labels and diverse challenging driving scenarios. The dataset is collected using the CARLA simulator, leveraging its leaderboard 2.0 framework. It contains sensor data, control decisions, and CoT labels to indicate the reasoning process. The authors propose a rule-based expert policy to control the vehicle and generate ground truth labels for different driving aspects and final decisions. The dataset serves as an open-loop end-to-end driving benchmark, enabling the evaluation of accuracy in various CoT aspects and final decisions. Additionally, the paper presents DriveCoT-Agent, a baseline model trained on the DriveCoT dataset. This model generates chain-of-thought predictions and final driving decisions, demonstrating strong performance in both open-loop and closed-loop evaluations. The model's ability to handle high-speed driving and lane-changing scenarios, along with its interpretability, makes it a significant contribution to the field of end-to-end autonomous driving. The main contributions of the paper are: 1. Introduction of DriveCoT, the first end-to-end driving dataset with CoT labels and diverse challenging scenarios. 2. Development of a rule-based expert policy to handle challenging scenarios in CARLA leaderboard 2.0. 3. Proposal of DriveCoT-Agent, a baseline model trained on DriveCoT that generates chain-of-thought predictions and final driving decisions, showing strong performance in both open-loop and closed-loop evaluations. The paper also includes a detailed description of the dataset collection process, the expert policy, and the model architecture. Experimental results demonstrate the effectiveness of the proposed dataset and model in various driving scenarios, highlighting the benefits of integrating CoT with end-to-end driving.**DriveCoT: Integrating Chain-of-Thought Reasoning with End-to-End Driving** This paper introduces DriveCoT, a comprehensive end-to-end driving dataset that includes chain-of-thought (CoT) labels and diverse challenging driving scenarios. The dataset is collected using the CARLA simulator, leveraging its leaderboard 2.0 framework. It contains sensor data, control decisions, and CoT labels to indicate the reasoning process. The authors propose a rule-based expert policy to control the vehicle and generate ground truth labels for different driving aspects and final decisions. The dataset serves as an open-loop end-to-end driving benchmark, enabling the evaluation of accuracy in various CoT aspects and final decisions. Additionally, the paper presents DriveCoT-Agent, a baseline model trained on the DriveCoT dataset. This model generates chain-of-thought predictions and final driving decisions, demonstrating strong performance in both open-loop and closed-loop evaluations. The model's ability to handle high-speed driving and lane-changing scenarios, along with its interpretability, makes it a significant contribution to the field of end-to-end autonomous driving. The main contributions of the paper are: 1. Introduction of DriveCoT, the first end-to-end driving dataset with CoT labels and diverse challenging scenarios. 2. Development of a rule-based expert policy to handle challenging scenarios in CARLA leaderboard 2.0. 3. Proposal of DriveCoT-Agent, a baseline model trained on DriveCoT that generates chain-of-thought predictions and final driving decisions, showing strong performance in both open-loop and closed-loop evaluations. The paper also includes a detailed description of the dataset collection process, the expert policy, and the model architecture. Experimental results demonstrate the effectiveness of the proposed dataset and model in various driving scenarios, highlighting the benefits of integrating CoT with end-to-end driving.