Continual SLAM: Beyond Lifelong Simultaneous Localization and Mapping Through Continual Learning

Robots operating in the open world encounter various different environments that can substantially differ from each other. This domain gap also poses a challenge for Simultaneous Localization and Mapping (SLAM) being one of fundamental tasks navigation. In particular, learning-based SLAM methods are known to generalize poorly unseen hindering their general adoption. this work, we introduce novel task continual extending concept lifelong single dynamically changing environment sequential deployments several drastically differing environments. To address task, propose CL-SLAM leveraging dual-network architecture both adapt new retain knowledge with respect previously visited We compare as well classical show advantages online data. extensively evaluate on three datasets demonstrate it outperforms baselines inspired by existing visual odometry methods. make code our work publicly available at http://continual-slam.cs.uni-freiburg.de .