Lyft recently released a Level 5 autonomous driving prediction dataset containing more than 1,000 hours of recorded driving. The company also launched an autonomous Driving Prediction Challenge with a $30,000 prize pool.

Lyft released a new data set.

Last July, Lyft released its L5 level Self-driving awareness dataset, which contains more than 55,000 HUMAN-flagged 3D annotated frames. Officials at the time called it the largest publicly available data set of its kind.

Just over a year later, Lyft has released another set of autonomous driving prediction data with an L5 rating.


Application download address: https://www.catalyzex.com/paper/arxiv:2006.14480/dataset

170,000 scenarios, more than 2,500 kilometers of road data

The data set Lyft released this time focuses on sports prediction. One of the problems of long-term research in autonomous driving is creating models that are robust and reliable enough to predict traffic movements, officials said.

The data was collected by a fleet of 23 autonomous vehicles on a fixed route in Palo Alto, California, over a period of four months and includes driving logs of cars, pedestrians and other obstacles encountered.

The data set specifically includes:

  • 1,000 hours: more than 1,000 hours of self-driving car movement recorded;
  • 170,000 scenes: each scene lasts about 25 seconds and includes traffic lights, aerial maps, sidewalks and more;
  • 16,000 miles: 16,000 miles of data from public roads;
  • 15,242 annotated images: includes a high-resolution semantic map of the tagged elements as well as a high-resolution aerial view of the area.
Example of a bird ‘s-eye semantic graph in a data set

The movement data is collected by a fleet of roof-mounted sensors that capture lidar, video cameras and radar data as the vehicle travels tens of thousands of miles.


In the dataset, each scene encodes the state around the vehicle at a given point in time.
Red for self-driving cars and yellow for other vehicles

Together with the toolkit provided, Lyft says the collection makes up the largest, most complete, and most detailed data set to date for developing autonomous driving, machine learning tasks like motion prediction, planning, and simulation.

Currently, only a subset of this dataset is available for download, including:

  • Sample dataset (53 MB)
  • Training data set (in three parts, 69.4GB)
  • Aerial view (2 GB)
  • Semantic graph (2 MB)

Download address:

Prediction

Launch a challenge with a prize pool of $30,000

Lyft, meanwhile, plans to launch a challenge that will begin on Google’s Kaggle platform in August with a total prize of $30,000.

Lyft launched a self-driving 3D object detection contest last year with a $25,000 prize pool

Highlights of this challenge:

  • Competition requirements: Participants predict the movement of vehicles;
  • Preparation: Researchers and engineers will now be able to download training datasets and Python-based packages to experiment with the data. Because test and validation suites will be released as part of the competition;
  • Ultimate goal: To enhance the capacity of the research community and accelerate innovation through data sets and competitions.

Sacha Arnoud, Lyft’s senior director of engineering, and Peter Ondruska, Lyft’s director of audio and video Research, wrote in a blog post, “Data is what drives trying out the latest machine learning technologies, and access to large-scale, high-quality autonomous driving data is limited, But that should not prevent us from experimenting with this research.”

“We believe driverless driving will become a more convenient, safer and sustainable part of the transportation system,” Arnoud and Ondruska said. “By sharing data with the research community, we hope to identify important and unsolved challenges in autonomous driving.”

Visit Hyper-. ai for more quality data sets!

Blog Address:

Medium.com/lyftlevel5/…

Thesis Address:

Arxiv.org/pdf/2006.14…

Making address:

github.com/lyft/l5kit/