Lyft Releases a New Autonomous Vehicle Dataset

US-based online taxi service seen as a rival to Uber Lyft, recently released the Perception Dataset and concluded its 2019 object detection competition. The company has now released the Prediction Dataset. In this data set; The company is headquartered in Palo Alto and 23 autonomous from the vehicle It contains records of the movements of cars, pedestrians and other obstacles encountered by the fleet. In parallel, Lyft plans to launch a competition in which participants will predict the movement of traffic elements.

One of the long-standing research problems in the field of autonomous transportation is to create robust and sufficiently reliable models to predict traffic movement. Lyft's data set will be used to focus on movement prediction by incorporating the movement of other types of traffic that its vehicle fleet intertwines with. These other types of traffic include vehicles, cyclists and pedestrians. The data obtained from this movement is obtained through sensors placed on the ceiling of Lyft's vehicles. These sensors LIDAR Or it works like radar.

The data obtained is very important:

Among the data obtained; over 1,000 hours of recording of the movement of traffic elements, each lasting around 25 seconds 170.000 scene, 16,000 miles of public roads (25.749 kilometer) data, 15,000 map descriptions, HD map of the area. Combined with the HD maps the Lyft team has built over Palo Alto, London and Munich, the final part contains the information needed to create the predictive models necessary for vehicles to choose safe routes in given scenarios.

Lyft's Sacha Arnoud and Peter Ondruska point out the importance of data to gain experience with cutting-edge machine learning techniques. The educational dataset and Python-based software kit will be available starting today. Test And verification sets It will be published as part of a competition that will start in August. Competition, Google Kaggle will be on the platform and in total 30.000 dollar reward will be given.

Lyft's second competition; Waymo's driving data set and 110.000 dollar It launches months after the award-winning Waymo Open Dataset competition. The contest winners were awarded online during a workshop at the 2020 Conference on Computer Vision and Pattern Recognition (CVPR) in mid-June. CVPR, was held online this year due to the coronavirus outbreak. You can download the data set published by Lyft here and follow the announcements about the competition here.