The autonomous driving software startup RideFlux announced on the 13th that it ranked 3rd in the vision-based end-to-end (E2E) autonomous driving category at the ‘Waymo Open Dataset Challenge’ held during the ‘CVPR 2025’ autonomous driving workshop.

CVPR is a prestigious conference in the field of computer vision, and this year's event is being held from the 11th to the 15th (local time) in Tennessee, USA. In particular, the vision-based E2E autonomous driving category saw fierce competition with 29 autonomous driving research and development teams participating from around the world, including NVIDIA, Xiaomi, the Swiss Federal Institute of Technology Zurich, and Canada’s MILA AI research institute.

RideFlux participated in this challenge in collaboration with the Hanyang University IRCV research team, proposing a model to effectively plan the trajectory of autonomous vehicles based on camera video data released by Waymo.

The ‘Swin-Trajectory’ method they proposed is an algorithm that understands structured vehicle history and the context of images using AI technology to plan future trajectories. Based on effective spatiotemporal modeling, this method not only achieves high accuracy but also operates quickly enough at approximately 14ms on commercial graphic processing units (GPUs), making it suitable for real-time processing, which is crucial in autonomous driving.

Vision-based E2E autonomous driving is considered a global technological trend that has recently garnered attention in the autonomous driving industry. Autonomous driving systems that utilize E2E technology are designed to learn various driving video data through AI models, enabling autonomous vehicles to integrate perception, prediction, judgment, and control in real driving environments.

Professor Hwang Soon-min from Hanyang University’s Department of Future Automotive Engineering stated, “Winning this CVPR 2025 challenge is a testament to the dedicated efforts and skills of our students and exemplifies successful industry-academia collaboration achieved through close cooperation with RideFlux,” adding, “It is meaningful to demonstrate the potential of domestic technology in the autonomous driving sector.”

Park Joong-hee, CEO of RideFlux, said, “This was an opportunity to reaffirm the potential of the technological trend of E2E autonomous driving and the competitiveness of purely domestic technology,” and expressed, “We will continue to strengthen our research and development (R&D) capabilities based on quality data obtained from various autonomous driving services to complete the most reliable autonomous driving.”

Detailed information about the planned model jointly proposed by RideFlux and the Hanyang University IRCV research team can be found on the ‘CVPR 2025 Workshop on Autonomous Driving’ website.

(Source: Edaily, reporter Kim Beom-jun, 25.06.13.)