We win the 1st Place in RobotDrive Challenge (ICRA 2024)
Celebrating Success: First Place in the RoboDrive Challenge ICRA 2024
We are thrilled to announce that our research group has secured the 1st Place in the highly competitive RoboDrive Challenge at ICRA 2024, specifically excelling in the Track-1 Robust BEV Detection.
This challenge, hosted at the 41st IEEE Conference on Robotics and Automation (ICRA) in Yokohama, Japan, plays a pivotal role in advancing the robustness of autonomous driving perception technologies. Our team’s entry focused on enhancing the accuracy and resilience of Bird’s Eye View (BEV) detection systems against a variety of real-world corruptions and sensor failures.
Our technique, TSMA-BEV, integrates AugFFT—an augmentation module leveraging fast Fourier transformation for robust image modifications—and SeqMixAug, a sequential consistency strategy that significantly enhances the stability of detections across diverse environmental challenges. This methodology not only supports robust perception across multiple corruptions but also aligns with real-world applications where conditions are unpredictable and varied.
Key to our success was the challenge’s focus on real-world application scenarios such as severe weather and sensor anomalies, conditions in which our TSMA-BEV framework excelled by maintaining high accuracy and stability. Our approach was tested against a competitive field of international teams, where it demonstrated superior capability to handle out-of-domain data and various visual corruptions effectively.
This victory is not just a testament to our team’s hard work and innovative spirit but also aligns with our ongoing commitment to contribute to the safety and reliability of autonomous driving technologies. Our technical report details the methodologies and technologies developed, highlighting our novel solutions to overcome the obstacles posed by the challenge scenarios.
Please see more information as follows.
Authors:
- Xu Cao, My student from HKUST(GZ)
- Hao Lu, My student from HKUST(GZ)
- Ying-Cong Chen, HKUST(GZ)
