Creating Safer Streets and Smarter Driverless Vehicles with nuScenes

At Motional, we’re constantly working to make our vehicles safe, smart, and reliable.

Driverless technology is as complex as it is promising, and industry collaboration is key to its success. There’s tremendous intellectual and organizational firepower directed at building a driverless future; thousands of roboticists, engineers, scientists, and researchers, working in automotive and tech companies both large and small. And while competition may be a fact of life, some things are more important—like safety.  

We believe industry-wide collaboration will bring safe driverless vehicles to our streets and communities faster. That’s why we created nuScenes, the industry-leading and publicly-available dataset that teaches driverless vehicles how to safely engage with their ever-changing surroundings.

How nuScenes is building a safe driverless reality

The original nuScenes launched as a collection of 1,000 twenty second-long “street-scenes” in Boston and Singapore. The scenes, composed of millions of photos and data-points collected from our vehicles’ full sensor suites, were then meticulously hand-annotated. They were then made available, free of charge, to the academic and research communities, where they were used to inform and advance machine learning models to build the safest possible self-driving vehicles.   

Pioneering a culture of data-sharing

nuScenes started a movement. Shortly after its release, our competitors followed suit, with at least ten similar datasets now available. nuScenes, however, remains the single most robust dataset of its kind, and has now been accessed by over 8,000 researchers and cited in 250+ scientific papers.

By working together, Motional and other industry players are fostering an environment where collaboration around safety is valued above competition, and we believe this will ultimately bring driverless vehicles to market faster.

An expanded dataset for greater insights and a safer future

Today, we’re announcing a significant expansion of nuScenes: nuScenes-lidarseg, and the brand new nuImages dataset. 

nuScenes-lidarseg is the application of lidar segmentation to the original 1,000 Singapore and Boston driving scenes. It provides a significantly more detailed picture of a vehicle’s surrounding environment than the original nuScenes’ bounding boxes:

nuScenes-lidarseg adds 1.4 billion annotated lidar points, making it the largest publicly available dataset with lidar segmentation.

The all-new nuImages brings in nearly 100,000 annotated, hand-picked images. These were specifically chosen to generate a dataset encompassing a wide scope of challenging aspects of autonomous driving, such as jay-walkers, large intersections, tough weather conditions, and daylight versus nighttime driving.

This expansion of nuScenes represents our continued investment in sharing crucial R&D datasets designed to help create the safest possible self-driving environments, and our commitment to making driverless vehicles a safe and reliable reality.

The road ahead

Driverless technology will, without question, be one of the most transformative, impactful technologies of our lifetime. It’s our responsibility—and that of our industry peers—to ensure that safety is at the core of everything we build. 

We have the power to write our industry’s success story. Our goal is to make this journey one that’s fueled by industry collaboration to develop the safest driverless vehicles possible—and make them an everyday reality for millions. 

nuScenes is free of charge for academic use, and licensing is available for commercial purposes. Learn more about the latest nuScenes expansion by reading the press release. And to learn more about Motional and our mission to change how the world moves, be sure to explore our website.