女子和闺蜜到浙江,男同乡尽"地主之谊"热情招待却暗自下药!醒过来在宾馆...
What are Autonomous Vehicles?
Autonomous Vehicles (AVs)—also known as self-driving cars or self-driving vehicles— are vehicles that can navigate and operate safely with little or no human input or intervention. Autonomous vehicles are equipped with autonomous driving systems that use a combination of sensors (such as cameras, radar, and/or lidar), compute, and software to safely perceive their environment and execute driving tasks.
What are the Levels of Autonomous Vehicle Automation?
The Society of Automotive Engineers (SAE) has defined these levels of automation:
- Level 0: No automation (the human driver is fully responsible; includes momentary driver assistance technologies like automatic emergency braking and lane departure warning)
- Level 1: Driver assistance (e.g., cruise control)
- Level 2: Partial automation (e.g., adaptive cruise control, lane-keeping)
- Level 2+: Higher level of partial automation, integrating various advanced driver assistance systems (ADAS)
- Level 3: Conditional automation (the vehicle can handle certain tasks, but the driver must be ready to intervene)
- Level 4: High automation (the vehicle can handle most driving tasks, but may require human intervention in specific scenarios)
- Level 5: Full automation (the vehicle is fully autonomous and can operate without human involvement in all conditions)
How Do Autonomous Vehicles Work?
Autonomous vehicles are powered by physical AI, taking in real-time data from onboard sensors, processing it using end-to-end foundation models running on high-performance compute, and planning and executing driving decisions.
At the heart of autonomous vehicle technology is a revolutionary three-computer framework that enables continuous learning and improvement:
- Data Center Training: The framework starts with NVIDIA DGX? systems for training end-to-end AI models in the data center. Consisting of a comprehensive infrastructure—spanning both hardware and software—DGX systems can be used to run the autonomous vehicles stack through massive amounts of data. This helps it learn how to perceive, plan, and act, helping the autonomous vehicle become safer over time.
- Simulation Environment: NVIDIA Omniverse? with Cosmos? running on NVIDIA OVX? enables rich 3D digital twins that model real-world sensor data, physics, and behavior. These digital twins can generate physically accurate and diverse sensor data to test, train, and validate autonomous vehicles in a safe simulation environment.
- In-Vehicle Compute: NVIDIA DRIVE AGX? acts as the brains of the vehicle, delivering the in-vehicle computing power required for safe, intelligent driving. It accelerates deep learning inference across perception, planning, and control tasks, enabling the vehicle to make real-time decisions based on sensor data. Complementing the hardware is the safety-certified NVIDIA DriveOS? operating system, which provides the real-time processing, security, and system monitoring needed to meet functional safety requirements.
Together, these three computers enable continuous development cycles, speeding improvements in performance and deployment at scale.
As autonomous vehicles drive in the real world, sensor data is collected and sent to the data center. The data is used to refine the autonomous vehicles software stack and add new capabilities. After training, the stack is retested and validated in simulation, then updated on the in-vehicle computer. The cycle then starts again with more driving and data collection.
Autonomous driving systems must be rigorously tested and validated in both simulated and real-world environments to ensure safety and reliability.
What are the Benefits of Autonomous Vehicles?
Autonomous driving offers a range of benefits. One of the most significant advantages is the potential to improve road safety by reducing vehicle collisions, many of which are caused by human error. Autonomous vehicles are designed to follow traffic laws, monitor blind spots, and detect hazards faster than human drivers.
Autonomous vehicles also have the potential to deliver the freedom of mobility for people who are unable to drive by expanding access to transportation options. They can navigate traffic jams and complex urban environments efficiently, which can help reduce traffic congestion and lower emissions—particularly as the transportation industry moves away from internal combustion engines and toward electric vehicles.
Safety in Autonomous Vehicle Development and Deployment
NHTSA and SAE have established rigorous safety standards for autonomous vehicles, including requirements for testing, reporting, and safety assessments.
An autonomous vehicle relies on a suite of sensors, combined with advanced software and compute, to perceive its environment, make decisions, and control the vehicle. As outlined above, the three-computer framework creates a continuous learning and feedback loop that uses data collected by vehicle sensors to improve overall performance—starting with safety.
In essence, autonomous vehicle safety requirements are derived from an analysis of risks. This includes understanding the use-cases based on product definition, identifying potential hazards, and evaluating their risk in terms of exposure (how likely is that a bad outcome happens), severity (how bad is the outcome), and controllability (is the human driver likely to be able to avoid negative outcomes in the case of autonomous vehicles planner failure). It’s quantified by a minimum mean-time-between-failure (MTBF).
NVIDIA Halos is a full-stack safety system that unifies vehicle architecture, AI models, compute, software, tools, and services to ensure the safe development of autonomous vehicles from cloud to car.
Halos provides the critical safety foundation to ensure system-level reliability and iterative improvement for automated driving systems. This includes integration of third party-assessed hardware, software, and processes with a diverse algorithmic architecture and validation pipelines.
Halos complements existing industry-standard safety practices, while introducing unique elements for AI-based end-to-end stacks. This ensures regulatory compliance and advances safe and reliable autonomous vehicle stacks, together with the NVIDIA Halos AI Systems Inspection Lab.
As autonomous vehicle technology continues to advance, ongoing collaboration between manufacturers, regulators, and the public will be essential to address new challenges, such as cybersecurity and data privacy. The ultimate goal is to achieve fully autonomous vehicles that can safely operate on public roads without the need for human intervention, benefiting society by reducing crashes, improving mobility, and creating more efficient transportation systems.
Next Steps
Learn More About Autonomous Vehicles
Deep dive into NVIDIA DRIVE AGX—the hardware and software platform to develop safe autonomous vehicles.
Review the Autonomous Vehicles Safety Report
Safety is imperative to developing autonomous vehicles. Learn how NVIDIA is prioritizing vehicle, occupant, and pedestrian safety in our latest report.
NVIDIA in Automotive
NVIDIA automotive solutions offer the performance and scalability to design, visualize, develop, and simulate the future of driving.