How Ford Uses AI to Create Diverse Driving Data

When it comes to teaching autonomous vehicles, you can never have too much training data. That’s where Nikita Jaipuria and Rohan Bhasin come in.

Jaipuria, a research scientist in computer vision and machine learning, and Bhasin, a research engineer, spoke with AI Podcast host Noah Kravitz about their roles at Ford Motor Company, as well as their GTC Digital session.

They explained that autonomous vehicles need to be trained on a diverse dataset that represents any possible situation, but it can be difficult to collect all of that data in the real world.

So Ford is using a combination of gaming engine-based simulations and generative adversarial networks (GANs) to fill the gap in real-world datasets.

Key Points From This Episode:

Using gaming engines, Ford can create synthetic data. Using GANs on these simulations increases their photorealism. They can deploy this sim-to-real pipeline on HPC clusters and quickly generate large amounts of data.
Jaipuria and Bhasin note significant success with datasets that were originally very small or heavily biased — meaning that there weren’t enough scenarios represented in the dataset.
This sim-to-real method reduces the time and cost it takes to collect and annotate real-world data, and could be broadly applicable to other perception features.

Tweetables:

“GANs make it much harder to identify which is real and which is not.” — Rohan Bhasin [9:10]

“When you start mixing synthetic data with real data, that’s when you start seeing boosts in performance, especially to unseen scenarios and datasets.” — Nikita Jaipuria [13:29]

NVIDIA’s Neda Cvijetic Explains the Science Behind Self-Driving Cars

What John Madden was to pro football, Neda Cvijetic is to autonomous vehicles. No one’s better at explaining the action, in real time, than Cvijetic. Cvijetic, senior manager of autonomous vehicles at NVIDIA, drives our NVIDIA DRIVE Labs series of videos and blogs breaking down the science behind autonomous vehicles.

Talking Trucks with TuSimple’s Chuck Price

Self-driving trucks are on the horizon, and Chuck Price, chief product officer of TuSimple, talks about how the company is developing commercially ready level four fully autonomous driving solutions for the logistics industry.

AI’s Mild Ride: RoadBotics Puts AI on Pothole Patrol

Thanks to AI, one startup is working to pave the way to a smoother ride. Benjamin Schmidt, president and co-founder of RoadBotics, talks about how his company is improving roads by identifying areas at risk of potholes so governments can improve roads proactively.