SiaSearch, a Berlin-based automated driving data handling platform, has partnered with Motional to provide engineers enhanced access to the nuScenes dataset. Then uScenes public dataset, which Motional debuted in 2019, teaches autonomous vehicles how to navigate changing road environments. Earlier this year, Motional announced an expansion for nuScenes that now includes 1.4 billion annotated LiDAR points and nearly 100,000 annotated 2D images.
The partnership between SiaSearch and Motional enables nuScenes users to access, explore, and understand the data in a more streamlined fashion. SiaSearch automatically extracts granular metadata from raw multimodal automated driving sensor data. It then makes the data fully searchable and accessible for evaluation through its GUI and API.
Needle in a Haystack
With the new Sia Interval Query Engine feature, which includes a catalog of 50 plus driving events and attributes, users can ask any questions about the raw data. According to SiaSearch, the goal of the partnership with Motional is to prevent engineers from feeling as if they are looking for a “needle in a haystack.”
“With SiaSearch, the way engineers interact with driving data changes completely,” explained Holger Caesar, Team Lead Data-Curation at Motional and nuScenes Project Lead. “SiaSearch automatically turns unstructured data into structured data and allows us, engineers, to efficiently interact with it, reducing the time spent on data selection.”
“Having spoken to many researchers, one of the key challenges with large driving data sets is that they are difficult to explore and access, especially if like most, you’re looking for specific situations to train or validate models,” said Uriel Eldan, Vice President of Business Development for SiaSearch. “The datasets lack semantic searchability. We’ve started SiaSearch to enable this searchability on very large sensor datasets through an automatic scene and event detection.”
SiaSearch automatically tags nuScenes with driving maneuvers, interactions with other traffic agents, and infrastructural and environmental attributes. With the tagging, researchers can use nuScenes to find any driving scenario they need almost immediately. As a result, researchers spend more time on the training models and less time manually reviewing the dataset.
“In the past 18 months, we have been developing and deploying this technology with leading automotive OEMs, Tier 1s, and full-stack companies,” Eldan said. “We hope making nuScenes available in a fully searchable format will help researchers in the field, but also inspire many more engineers to work on automated driving.”
How to Access nuScenes
nuScenes is free for academic use, and licensing is available for commercial purposes. The enhancements to the nuScenes dataset outlined above can be accessed through the SiaSearch website. Engineers, researchers, and other industry experts can register now and will receive free access following a brief approval process. Along with launching nuScenes last year, Motional also contributed to the Safety First for Automated Driving whitepaper.