NVIDIA DriveWorks

NVIDIA DriveWorks
Name	NVIDIA DriveWorks
Title	NVIDIA DriveWorks
Developer	NVIDIA Corporation
Initial release	2015
Latest release	2019
Programming language	C++
Operating system	Linux (operating system)
License	Proprietary

NVIDIA DriveWorks is a software development kit and middleware suite for autonomous vehicle and advanced driver-assistance systems development produced by NVIDIA Corporation. It provides libraries, APIs, tools, and sample applications intended to accelerate sensor processing, perception, localization, and simulation workflows for automotive and robotics research. DriveWorks integrates with hardware platforms and ecosystems used in autonomous driving projects, linking compute, sensor, and simulation stacks.

Overview

DriveWorks was introduced by NVIDIA Corporation to complement the NVIDIA DRIVE hardware platform and to support ecosystem partners such as Bosch (company), Continental AG, Aptiv PLC, ZF Friedrichshafen AG, and Valeo. It targets companies and institutions including automotive OEMs like Tesla, Inc., General Motors, Toyota Motor Corporation, Ford Motor Company, and research organizations such as Carnegie Mellon University, Massachusetts Institute of Technology, and Stanford University. DriveWorks provides modular components for perception, sensor fusion, calibration, and simulation that interoperate with frameworks and standards adopted by entities like ROS, OpenCV, CUDA, TensorRT, and OpenGL.

Architecture and Components

The DriveWorks architecture layers middleware, perception modules, sensor drivers, and simulation tools atop NVIDIA compute platforms such as NVIDIA DRIVE AGX and NVIDIA Tegra. Core components include sensor interface modules for LiDAR vendors like Velodyne Lidar and Ouster (company), camera and radar drivers compatible with suppliers such as Mobileye and Bosch (company), and middleware for data buses including CAN bus used broadly by automotive suppliers such as Denso Corporation and Magneti Marelli. The stack integrates compute-acceleration technologies from CUDA, inference engines like NVIDIA TensorRT, and image processing from OpenCV. It also provides simulation interfaces to platforms such as CARLA (simulator), LGSVL Simulator, and visualization tools tied to Omniverse and Gazebo.

Features and Functionality

DriveWorks offers functionality for sensor calibration, time synchronization, and real-time data pipelines leveraging NVIDIA GPU acceleration. Perception modules encompass object detection and classification, semantic segmentation, and tracking using deep learning models trained with frameworks like PyTorch and TensorFlow. Localization and mapping components interface with algorithms like Simultaneous localization and mapping implementations used in projects at ETH Zurich and University of Oxford. Additional features include motion estimation, ego-motion computation, sensor fusion for multi-modal inputs (cameras, LiDAR, radar), and runtime telemetry compatible with enterprise tools from Siemens, ABB, and Rockwell Automation for vehicle integration.

Development and Integration

DriveWorks SDK provides APIs exposed in C++ and integrates with development toolchains used by partners such as Arm Holdings for embedded systems and Intel for data center workflows. The SDK is designed to interoperate with middleware from AUTOSAR suppliers and with standards promoted by organizations including SAE International and ISO. Development workflows typically use continuous integration systems like those by Jenkins (software), code repositories from GitHub, and dataset annotation tools from vendors such as Scale AI and Labelbox. Integration practices draw on mapping and HD map providers like Here Technologies, TomTom, and Mapbox to support localization and routing.

Use Cases and Deployment

DriveWorks is applied in autonomous driving levels by companies pursuing SAE J3016 classifications, including pilot projects for autonomous taxis at firms like Waymo LLC and Cruise (company), deployed research platforms at Uber Advanced Technologies Group, and production ADAS features by automakers including BMW, Mercedes-Benz Group, and Volkswagen Group. Use cases include perception stacks for urban driving tested in cities such as Palo Alto, California, Munich, and Shenzhen, simulation-driven testing in facilities like those run by NASA research partners, and fleet validation programs managed by providers like Uber Freight and logistics divisions of DHL.

Compatibility and Platforms

The DriveWorks SDK targets embedded NVIDIA platforms including NVIDIA DRIVE Xavier and NVIDIA DRIVE AGX Pegasus, and supports host development on distributions of Ubuntu (operating system) favored by research labs at University of California, Berkeley and industrial teams at Toyota Research Institute. It interoperates with middleware and tools across ecosystems such as ROS 2, machine learning toolchains from Google DeepMind-affiliated projects, and cloud platforms provided by Amazon Web Services, Microsoft Azure, and Google Cloud Platform for training and simulation scale-out. Hardware compatibility extends to sensor manufacturers including Velodyne Lidar, Ouster (company), Quanergy Systems, and camera suppliers like OmniVision Technologies.

History and Versioning

DriveWorks was announced by NVIDIA Corporation following the introduction of the DRIVE platform; early releases coincided with partnerships showcased at events like CES and Mobile World Congress. Versioning aligned with DRIVE hardware generation updates and with ecosystem milestones from partners such as Audi, Hyundai Motor Company, and NIO Inc.. Over time, DriveWorks evolved alongside NVIDIA’s software portfolio including CUDA Toolkit, DeepStream SDK, and TensorRT releases, with notable integration announcements at developer conferences such as NVIDIA GTC and industry gatherings like Automotive World. The public-facing SDK releases tapered as NVIDIA consolidated autonomous driving efforts into broader platforms and collaborations with suppliers and research institutions.

Category:Autonomous vehicle software Category:NVIDIA software