In recent years, self-driving vehicles have gradually been appearing on the road, but society has also begun to worry about the possibility of accidents caused by the anomaly self-driving system. Many researchers have begun to study the anomaly detection of self-driving vehicles, and each has proposed different detection algorithms. However, since self-driving vehicles are not yet popular, how to collect data, simulate attacks, and verify and compare multiple algorithms is a major obstacle to research. In this regard, we built an Internet of Vehicles platform, VADtalk, that facilitate anomaly detection modeling and deployment for self-driving vehicles. VADtalk contains programs such as anomaly detection model training and vehicle connection. When developers complete model uploading and setting through the GUI, the platform will automatically collect self-driving data, train the model, and even verify the operation of the model using a self-driving simulator, and then provide the results to the developer. After the developer determines the model, VADtalk can connect the trained model with the self-driving vehicle to actually perform real-time anomaly detection on it.