A continuous pipeline, or data factory, is needed for the training, testing, and validation of autonomous vehicles in order to add new scenarios and hone deep neural networks.
The simulation stage of this process is crucial. With high-fidelity, physically based simulation, AV developers can test a virtually infinite number of scenarios repeatedly and in large quantities. Simulation is constantly evolving and getting better, getting closer and closer to bridging the gap between the real and virtual worlds, just like much of the AI-related technology.
A virtual testing and validation environment for AV is provided by NVIDIA DRIVE Sim, which was built on Omniverse. It is a highly accurate simulation platform that can support ground-breaking tools like neural reconstruction and the generation of synthetic data, allowing for the creation of digital twins of driving environments and scenarios.
The AI Podcast was joined by Matt Cragun, senior product manager for AV simulation at NVIDIA, to discuss the evolution of simulation for self-driving technology. He went into detail about the beginnings and inner workings of DRIVE Sim.
In addition, he gave a sneak preview of the research frontiers being investigated for this crucial testing and validation technology.
