With the rapid advancement of technology and the gradual establishment of standardized guidelines, the autonomous driving industry is focused on ensuring the safety of autonomous vehicles (AVs) and their interactions with human-driven vehicles. Recently, at the NVIDIA AI Summit held in Washington, industry experts from regulatory and technical fields engaged in in-depth discussions on this critical topic.

NVIDIA’s Vice President of Automotive, Danny Shapiro, former National Highway Traffic Safety Administration (NHTSA) Director Mark Roskinde, and NVIDIA AV Research Director Marco Pavone participated in the discussion. Shapiro began by highlighting the urgent need to enhance road safety, using the high frequency of global traffic accidents and significant casualties as a starting point. He noted that NVIDIA has been collaborating with the automotive industry for over two decades, dedicating efforts to the development of advanced driver-assistance systems and fully autonomous driving technologies.

NVIDIA's AV development strategy focuses on the integration of three computers: one dedicated to AI training, another for simulation testing and validation, and a third installed in the vehicle to process sensor data in real-time for making safe driving decisions. This framework establishes a continuous development loop, consistently advancing the performance and safety of AV software.

As an authority in automotive safety, Roskinde analyzed the complexity of current U.S. regulations, pointing out that federal agencies focus on the vehicles themselves, while individual states emphasize the operators, including driver education, insurance, and licensing aspects. Pavone shared how emerging technologies, particularly in generative AI and neural rendering, are prompting researchers and developers to rethink AV development methodologies. These new technologies not only enhance simulation capabilities by generating complex scenarios for safety testing but also leverage foundational models, such as visual language models, to aid developers in building more robust autonomous software.

During the discussion, a collaborative initiative announced by MITRE at the AI Summit emerged as a relevant and timely topic. MITRE will partner with Michigan University's Mcity to develop virtual and physical AV verification platforms for industry deployment. This platform will integrate Mcity's simulation tools with digital twins of their testing facilities, providing physics-based sensor simulation capabilities through the NVIDIA Omniverse Cloud Sensor RTX API. This collaboration will enable developers to conduct thorough testing in simulated environments, ensuring vehicle safety before deploying AVs in the real world.

Roskinde commented that MITRE's announcement provides a reliable source capable of creating an independent and neutral environment for testing safety assurances, a practice that has been successful in fields such as aviation. Pavone added that simulation plays a crucial role in these efforts, allowing for the testing of extreme hazardous conditions in repeatable and diverse ways, thereby enabling large-scale simulation of various scenarios.

Shapiro also emphasized the advantages of simulation: "Simulations are repeatable and controllable. We can control the weather and time within simulations, manage all scenarios, and introduce various hazard factors. Once a simulation is set up, we can run it repeatedly, ensuring that as our software evolves, we are addressing issues and making adjustments as needed."

The forum concluded by emphasizing the ultimate goal of autonomous driving—reducing deaths and injuries on the roads—which is a shared objective of both the industry and regulatory agencies.