A Survey of Continual Learning for Robotics in the Foundation Model Era

The integration of foundation models has revolutionized robotics, endowing agents with unprecedented perception, planning and control capabilities. However, most current robotic models remain static: they are trained offline on fixed datasets and deployed without persistent updates, despite nonstationary environments, shifting task demands, and evolving user needs. Continual learning (CL) offers a principled solution by enabling robots to update from streaming experience while retaining previously acquired capabilities and avoiding catastrophic forgetting. Despite the accelerating progress of CL for robotics, an up-to-date review dedicated to this significant topic remains absent. To bridge this gap, we present the first comprehensive survey of Continual Learning for Robotics in the foundation model era, providing clear task formulations, a taxonomy of techniques, a review of benchmarks and environments, and a discussion of key challenges and future directions.