At present, improving the safety of autonomous driving mainly relies on comprehensive, in-depth, and scientific testing and improvement of software and hardware systems; Adopting comprehensive multi-level operational monitoring, safety assessment, early warning, and timely intervention measures; Effectively achieve vehicle road collaboration, expand the perception range of vehicles, improve computing power, and optimize redundant design of software and hardware. Recently, according to foreign media reports, engineers at the Massachusetts Institute of Technology in the United States have proposed introducing a hybrid system to address some of the safety deficiencies currently present in autonomous driving. In this hybrid system, autonomous vehicles can handle simple scene operations on their own, such as cruising on highways, while transferring more complex operations to remote operators. "The safety work of automatic driving is to adopt some safety mechanisms and safety measures technically to gradually reduce the risk of automatic driving to a reasonable and acceptable state," said Xie Hui, director of the Cross Research Center for Unmanned Vehicles of Tianjin University. Currently, safety remains one of the most urgent challenges to be addressed in the field of autonomous driving. Relevant enterprises and research institutions have explored many methods to ensure the safety of autonomous driving, and the establishment of relevant standards and safety systems has also improved the safety of autonomous driving to a certain extent. However, current autonomous driving still cannot allow drivers to release the steering wheel with peace of mind. More exploration is still needed to allow vehicles to roam freely on the road. These technologies provide an escort for automatic driving. Wang Caimei, a doctoral student from the School of Machinery, Tianjin University, introduced that automatic driving vehicles can use inertial navigation, cameras, radars and other sensing devices to monitor and predict the position and posture of the vehicle and its surrounding environment in real time, and evaluate the current safety of the vehicle in real time according to a sound safety index system. When the vehicle's driving safety assessment results in low safety, autonomous vehicles can issue warning signals through sound or vibration to remind passengers. When encountering unexpected situations, the system cannot automatically handle them, ensuring driving safety and requiring manual intervention. "When the auto drive system detects that the current situation is not suitable for automatic driving, or human drivers think that the current driving situation is more complicated and they need to take back the driving right, they can take over the vehicle at any time to ensure the safety of the vehicle," said Wang Caimei. Domain controller software and hardware, perception, decision-making, planning and control algorithms are the core of auto drive system, and its security is particularly important for autonomous vehicles. The vehicle safety system detects existing or potential problems in a timely manner by querying the software and hardware operation status of the domain controller in real-time, evaluating the rationality and safety of each algorithm operation, and ensuring that the "driving brain" is always in a safe operation state. "Perceptual algorithms need to rely on a large amount of data for training and inference, so data security is very important, and it needs to consider whether there are loopholes in data sources, transmission, storage and other links." Guo Fan, a doctor from the School of Mechanics of Tianjin University, introduced that for the loopholes and defects in the perceptual algorithm, monitoring and detection technologies can be used to identify them, or repair and upgrade technologies can be used to improve the algorithm and model. Guo Fan stated that decision-making and planning algorithms should be considered in various traffic scenarios and complex environments