Nvidia accelerates development of autonomous systems
While simulation is critical for training, testing and deploying autonomy, achieving real-world fidelity is incredibly challenging.
It requires accurate modeling of the physics and behavior of an autonomous system’s sensors and surroundings.
Designed to address this challenge by delivering large-scale, high-fidelity sensor simulation, Omniverse Cloud APIs, announced recently at Nvidia GTC, are poised to accelerate the path to autonomy. They bring together a rich ecosystem of simulation tools, applications and sensors.
The application programming interfaces address the critical need for high-fidelity sensor simulations to safely explore the myriad real-world scenarios autonomous systems will encounter.
In addition, the Omniverse Cloud platform offers application developers access to a range of powerful Universal Scene Description (OpenUSD), RTX and generative AI-enabled service-level cloud APIs to bring interoperability and physically based rendering to next-generation tools.
Simulation key to unlocking new levels of safety
As demand increases for robots, AVs, and other AI systems, developers are seeking to accelerate their workflows. Sensor data powers these systems’ perception capabilities, enabling them to comprehend their environment and make informed decisions in real time.
Traditionally, developers have used real-world data for training, testing and validation.
However, these methods are limited in covering rare scenarios or data that can’t be captured in the real world. Sensor simulation provides a seamless way to effectively test countless “what if” scenarios and diverse environmental conditions.
With Omniverse Cloud APIs, developers can enhance the workflows they’re already using with high-fidelity sensor simulation to tackle the challenge of developing full-stack autonomy.
This not only streamlines the development process but also lowers the barriers to entry for companies of virtually all sizes developing autonomous machines.
The ecosystem advantage
By bringing together an expansive ecosystem of simulators, verification and validation (V&V) tools, content and sensor developers, the Omniverse Cloud APIs enable a universal environment for AI system development.
Developers and software vendors such as Carla, MathWorks, Mitre, Foretellix and Voxel51 underscore the broad appeal of these APIs in autonomous vehicles.
Carla is an open-source AV simulator used by more than 100,000 developers. With Omniverse Cloud APIs, Carla users can enhance their existing workflows with high-fidelity sensor simulation.
Similarly, Mitre, a nonprofit that operates federally funded R&D centers and is dedicated to improving safety in technology, is building a Digital Proving Ground for the AV industry to validate self-driving solutions.
The DPG will use the Omniverse APIs to enable core sensor simulation capabilities for their developers.
MathWorks and Foretellix provide critical simulation tools for authoring, executing, monitoring, and debugging of testing scenarios.
As the GTC demo showed, combining such simulation and test automation tools with the APIs forms a powerful test environment for AV development. On the showfloor, Foretellix is showing an in-depth look at this solution in Booth 630.
And, by integrating the APIs with Voxel51’s FiftyOne platform, developers can easily visualize and organize ground-truth data generated in simulation for streamlined training and testing.
Leading industrial-sensor solution provider SICK is working on integrating these APIs in its sensor development process to reduce the number of physical prototypes, iterate quickly on design modifications and validate the eventual performance.
These validated sensor models can eventually be used by autonomous systems developers in their applications.
Developers will also have access to sensor models from a variety of manufacturers, including lidar makers Hesai, Innoviz Technologies, Luminar, MicroVision, Robosense, and Seyond, visual sensor suppliers Omnivision, onsemi, and Sony Semiconductor Solutions, and Continental, Forvia Hella, and Arbe for radar.
Additionally, AI/ML developers can call on these APIs to generate large and diverse sets of synthetic data – critical input for training and validating perception models that power these autonomous systems.
Empowering developers and accelerating innovation
By reducing the traditional barriers to high-fidelity sensor simulation, Nvidia Omniverse Cloud APIs empower developers to address complex AI problems without significant infrastructure overhauls.
This democratization of access to advanced simulation tools promises to accelerate innovation, allowing developers to quickly adapt to and integrate the latest technological advancements into their testing and development processes.
