The report focuses on AI safety and R&D toolchains for intelligent vehicles. Drawing on the team’s extensive research in the system safety of connected and autonomous vehicles (CAVs) and autonomous driving control, it analyzes failure risks in AI perception and decision-making algorithms, as well as functional safety and data security challenges. It further elaborates on a comprehensive safety protection framework tailored for full-scenario autonomous driving. Addressing key industry pain points, the report introduces an end-to-end R&D toolchain for intelligent driving, covering simulation testing and electronic control development platforms, while demonstrating their practical applications in algorithm verification and safety assessment. Finally, by integrating theoretical insights with engineering practice, it explores iterative pathways for advancing AI safety technologies and outlines future directions for domestically developed tools, offering actionable strategies to drive technological upgrades and real-world deployment across the industry.

A 22-DOF vehicle dynamics model is developed in Modelica and implemented in MWorks Sysplorer 2025a.The generalized coordinates comprise six sprung-body motions, twelve translational motions of the four unsprung masses, and four wheel-spin motions. The physical plant is formulated as a coupled differential-algebraic equation system: the body, suspension, unsprung mass, wheel, tire, and road components are connected through custom acausal physical connectors, whereas steering and wheel-torque commands are retained only as external causal boundaries. The model is assessed using acceleration, braking, and steering-step datasets from MWorks, a 22-DOF Simulink model, and road tests. The threesource comparison demonstrates consistent overall dynamic trends and close agreement in key performance indicators, particularly in braking-distance prediction and steady steering response, thereby validating the effectiveness of the proposed acausal 22-DOF modeling framework.