Online self-supervised learning-based traversability analysis method for off-road autonomous driving

With the development of off-road autonomous driving technology, accurately identifying traversable and non-traversable terrain has become a key challenge. This paper proposes an online self-supervised adaptive clustering method based on the Chinese Restaurant Process (CRP) for dynamically assessing terrain traversability without labeled data. By analyzing the terrain feature of historical trajectories, the clustering process allows to update in real time and provide probabilistic assessments of traversability. This method has strong generalization capabilities and can adapt to changes in different environments. Experimental results demonstrate that the proposed method can provide Accurate traversability assessments with good generalization in complex off-road environments, enhancing the safety and robustness of off-road autonomous driving.