Simulation / Modeling / Design

New DRIVE OS and DriveWorks Updates Enable Streamlined AV Software Development

AI-Generated Summary

  • NVIDIA DRIVE OS and NVIDIA DriveWorks releases are now available together on NVIDIA DRIVE Developer, giving developers simultaneous access to the operating system foundation and autonomous vehicle middleware.
  • DRIVE OS adds the NvMedia Sensor Input Processing Library (SIPL) with safety proxy support for Linux and expands sensor module coverage, while NvStreams improves buffer allocation and synchronization across SoC, dGPU, and CPU engines.
  • The release includes the latest NVIDIA TensorRT™ with dynamic shape, reformat-free I/O, explicit precision, pointwise layer fusion, and shuffle elimination.
  • DriveWorks integrates its Sensor Abstraction Layer with NvMedia SIPL to support additional GMSL cameras such as the Sony IMX390 and ON Semi AR0820, and adds out-of-the-box support for Luminar H3 and Ouster OS2-128 lidars plus the U-blox ZED-F9P GNSS module.
  • A new Time Sensor module in DriveWorks maintains time correlation data and converts between different sensor timestamp clocks, and a DNN tensor module wraps raw tensor data with dimension and layout definitions.
  • New image processing algorithms run on the NVIDIA DRIVE AGX Programmable Vision Accelerator (PVA) in addition to the GPU, expanding acceleration options for perception workloads.

Next Steps

Powered by NVIDIA Nemotron. AI-generated content may summarize information incompletely. Verify important information. Learn more

You asked, we listened: DRIVE OS and DriveWorks releases are now available on NVIDIA DRIVE Developer, providing DRIVE OS users access to DriveWorks middleware and even more updates.

With these releases, developers have access to the latest DRIVE OS and DriveWorks software for autonomous vehicle development, including new features, without having to wait for DRIVE Software updates.

The foundation of the NVIDIA DRIVE software stack, NVIDIA DRIVE OS is designed specifically for accelerated computing and artificial intelligence. It includes NvMedia for sensor input processing, NVIDIA CUDA for efficient parallel computing implementations, NVIDIA TensorRT™ for real-time AI inference, and specialized developer tools and modules that allow developers to access the accelerated hardware engines. 

The NVIDIA DriveWorks SDK provides functionality fundamental to autonomous vehicle development, consisting of a sensor abstraction layer (SAL), sensor plugins, data recorder, vehicle I/O support, and a deep neural network (DNN) framework. It’s modular, open and designed to be compliant with automotive industry software standards.

And now, these key components for autonomous vehicle software development are even more accessible to developers, with frequent updates to unlock performance on the NVIDIA DRIVE AGX platform for greater flexibility and capabilities.

Laying a Foundation with DRIVE OS

DRIVE OS is a robust operating system for autonomous vehicle development, providing access to the underlying compute accelerators in DRIVE AGX Xavier.

New for this release is NvMedia Sensor Input Processing Library (SIPL), an image processing API, targeted for safety. SIPL adds sensor device and query block sources, in addition to source files and libraries to an expanded range of sensor modules. It also delivers safety proxy support for Linux — a mechanism that makes it easier to develop safety applications on non-safety platforms. 

Updated for this release, NvStreams enables efficient allocation, sharing and synchronizing of data buffers across the SoC, dGPU and CPU engine APIs, making it easy for developers to move large data buffers for processing. 

Also included is the latest TensorRT with dynamic shape, reformat-free I/O, explicit precision, pointwise layer fusion and shuffle elimination, as well as new plugins and samples to help developers take advantage of the platform.

Going Further with DriveWorks

On top of DRIVE OS, DriveWorks enables applications to help incorporate software into the vehicle. These include integrating automotive sensors within the software stack, accelerating camera and lidar data processing, interfacing with the vehicle, accelerating inference for perception and calibrating multiple sensor modalities with precision.

Key DriveWorks highlights include the integration of the DriveWorks SAL with NvMedia SIPL, enabling recording of additional GMSL cameras such as the Sony IMX390 and the ON Semi AR0820. Additionally, DriveWorks SAL now supports even more sensors out-of-the-box, such as the Luminar H3 and Ouster OS2-128 lidars as well as the U-blox ZED-F9P GNSS module. 

As always, developers can integrate their own sensors into DriveWorks using the Sensor Plugin Framework. 

Finally, the DriveWorks SAL also now includes a new Time Sensor module for synchronizing timestamps of supported sensors. This module maintains time correlation information correspondence data and supports the conversion between the different clocks used to timestamp sensor data.  

For image processing, the release adds new algorithms to run on the NVIDIA DRIVE AGX Programmable Vision Accelerator (PVA) in addition to the GPU. A new DNN tensor module wraps raw tensor data into a structure, allowing the user to define dimensions and layouts. It also supports the traversal of complex layouts as well as the ability to lock/unlock the data to prevent simultaneous operations. 

By making DRIVE OS and DriveWorks releases available together, developers now have the latest and greatest DRIVE OS features and performance together with seamless autonomous vehicle integration and utilities provided by DriveWorks SDK.

DRIVE AGX developers may access the latest release here.

Discuss (0)

Tags