METAVerse: Meta-Learning Traversability Cost Map for Off-Road Navigation

• DOI: 10.48550/arxiv.2307.13991
• Publication Date: 2023-01-01
• Authors: Junwon Seo; Taekyung Kim; Seongyong Ahn; Kiho Kwak
• Keywords: FOS: Computer and information sciences; Computer Science - Robotics; Computer Science - Machine Learning; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition; Robotics (cs.RO); Machine Learning (cs.LG)

Autonomous navigation in off-road conditions requires an accurate estimation of terrain traversability. However, traversability unstructured environments is subject to high uncertainty due the variability numerous factors that influence vehicle-terrain interaction. Consequently, it challenging obtain a generalizable model can accurately predict variety environments. This paper presents METAVerse, meta-learning framework for learning global and reliably predicts across diverse We train prediction network generate dense continuous-valued cost map from sparse LiDAR point cloud, leveraging interaction feedback self-supervised manner. Meta-learning utilized with driving data collected multiple environments, effectively minimizing uncertainty. During deployment, online adaptation performed rapidly adapt local environment by exploiting recent experiences. To conduct comprehensive evaluation, we collect various terrains demonstrate our method minimizes Moreover, integrating predictive controller, reduced results safe stable unknown terrains.