NVIDIA (ahem, one of Udacity’s partners in developing the Udacity Self-Driving Car Engineer Nanodegree Program) announced their new DRIVE Constellation Autonomous Vehicle Simulator a few weeks ago at their annual GPU Technology Conference.
NVIDIA is taking the next step toward the holy grail of autonomous vehicle simulation: training autonomous driving software in the simulator and deploying it straight to the real world.
We’re not there yet. NVIDIA executives subtly make the point that DRIVE Sim is for “testing”:
“Deploying production self-driving cars requires a solution for testing and validating on billions of driving miles to achieve the safety and reliability needed for customers,” said Rob Csongor, vice president and general manager of Automotive at NVIDIA. “With DRIVE Constellation, we’ve accomplished that by combining our expertise in visual computing and datacenters. With virtual simulation, we can increase the robustness of our algorithms by testing on billions of miles of custom scenarios and rare corner cases, all in a fraction of the time and cost it would take to do so on physical roads.”
Testing and validation is hugely important to the autonomous vehicle industry. Every time a software developer tweaks a model or a the autonomous vehicle source code, the developer needs to verify that nothing else broke. The fastest way to do this is testing in simulation. NVIDIA is releasing what looks like the world’s best simulation platform for testing.
The biggest leap, though, is training in simulation. It’s one thing to write code, put it in a car, collect data about different scenarios, and then later test in simulation that the your self-driving software still handles those scenarios correctly.
It’s a whole different animal to be able to train self-driving software in simulation, without ever putting a car out on the road. Whoever cracks that nut will have a huge leg up in the autonomous vehicle race. Keep an eye on NVIDIA.