“Where Does the Devil Lie?”: Multimodal Multitask Collaborative Revision Network for Trusted Road Segmentation

Road segmentation is an essential component of navigation systems. Although recent advancements in road segmentation, the occurrence of failure segmentations remains inevitable. For safety-critical tasks, e.g., navigation, knowing when and where road segmentation fails is crucial. In this paper, we propose a novel trusted road segmentation architecture, namely Multimodal Multitask Collaborative Revision Network (M2CRN), to improve the trust of road segmentation. Our approach incorporates two strategies to predict and rectify segmentation errors. First, a joint learning framework is devised to generate road segmentation results while estimating failure segmentation masks. Second, the road segmentation branch is equipped with an Uncertainty-Aware Revision Module (UARM), which eliminates the error in road segmentation. Additionally, we suppress the response of error regions in the road segmentation branch with an innovative design, called Adaptive Soft Error Suppression (ASES). To validate our methods, extensive experiments are conducted on three benchmark road segmentation datasets. The results demonstrate significant performance improvements with a real-time inference speed of 33.3 FPS, reaffirming the soundness of our revision model.

I. Introduction

Road segmentation, a vital component in safety-critical systems, e.g., autonomous driving [1], [2] and robot-assisted [3], serves to detect freespace. The road segmentation results are then utilized by downstream applications, e.g., trajectory prediction [4], decision making [5], and path planning [6], to guarantee the navigation of safety-critical systems within intricate road environments. Road segmentation holds significant importance, prompting continuous efforts by researchers [7], [8], [9] to improve its efficiency and meet the safety-critical requirements in computer vision.