Abstract
In the realm of autonomous driving, robust perception underout-of-distribution conditions is paramount for the safe deployment ofvehicles. Challenges such as adverse weather, sensor malfunctions, andenvironmental unpredictability can severely impact the performance ofautonomous systems. The 2024 RoboDrive Challenge was crafted to propel thedevelopment of driving perception technologies that can withstand and adapt tothese real-world variabilities. Focusing on four pivotal tasks -- BEVdetection, map segmentation, semantic occupancy prediction, and multi-viewdepth estimation -- the competition laid down a gauntlet to innovate andenhance system resilience against typical and atypical disturbances. Thisyear's challenge consisted of five distinct tracks and attracted 140 registeredteams from 93 institutes across 11 countries, resulting in nearly one thousandsubmissions evaluated through our servers. The competition culminated in 15top-performing solutions, which introduced a range of innovative approachesincluding advanced data augmentation, multi-sensor fusion, self-supervisedlearning for error correction, and new algorithmic strategies to enhance sensorrobustness. These contributions significantly advanced the state of the art,particularly in handling sensor inconsistencies and environmental variability.Participants, through collaborative efforts, pushed the boundaries of currenttechnologies, showcasing their potential in real-world scenarios. Extensiveevaluations and analyses provided insights into the effectiveness of thesesolutions, highlighting key trends and successful strategies for improving theresilience of driving perception systems. This challenge has set a newbenchmark in the field, providing a rich repository of techniques expected toguide future research in this field.