Toward Lifelong Autonomy: Advancing Continual Robot Learning for Adaptable Intelligent Systems

Continual robot learning is an emerging interdisciplinary field that integrates advances from machine learning, robotics, and cognitive science to build agents capable of accumulating knowledge and skills throughout their lifetime. Unlike traditional training paradigms that assume static objectives, continual learning examines how robots can learn new tasks drawn from an evolving, potentially unbounded distribution. The key challenge lies in balancing stability and plasticity—learning new behaviours efficiently while retaining previously acquired competencies. Foundations from continual learning research have introduced three main families of methods: (1) regularization-based strategies that constrain updates to preserve past knowledge, (2) replay-based approaches that revisit experience memories to prevent forgetting, and (3) dynamic architectures that allocate specific model capacity to emerging tasks. Within robotics, these concepts have been reinterpreted to address domain-specific requirements such as sensorimotor grounding, interaction with physical environments, and autonomous skill composition. Recent breakthroughs include the emergence of self-supervised world modeling and unsupervised skill discovery frameworks that enable autonomous adaptation with minimal supervision. Increasing attention is also being directed toward structured memory systems, uncertainty-aware estimation, and model-based representations that support scalable knowledge retention, transfer, and decision-making in open-ended learning settings.This Research Topic aims to advance the understanding and application of continual robot learning by bringing together contributions that span theory, methodology, and real-world deployment. It seeks to identify scalable methods that endow robots with the ability to autonomously acquire, refine, and reuse knowledge across diverse environments and interaction contexts. The ultimate goal is to move toward systems that can learn continuously without catastrophic forgetting, handle noisy or overlapping tasks, and dynamically adapt to non-stationary dynamics. Particular emphasis is placed on safety, robustness, and reliability during continual adaptation, ensuring that evolving policies remain predictable and trustworthy when deployed on physical robotic platforms. By exploring learning paradigms that combine joint perception, planning, and control with continual adaptation, the Research Topic will help bridge the gap between simulation-driven advances and the challenges posed by real-world robotics. It also encourages discussions on cognitive and computational mechanisms underlying open-ended learning systems and their implications for interactive intelligence.To gather further insights into scalable lifelong learning frameworks applicable to robotics and embodied agents, we welcome articles addressing, but not limited to, the following themes:- Continual reinforcement learning for robotic control and decision-making- Structured memory, world models, and uncertainty-aware learning for lifelong adaptation- Continual learning from demonstration, skill transfer, and policy composition- Autonomous skill discovery and self-supervised learning in robots and embodied agents- Safe and reliable continual learning under non-stationary environments and dynamics- Continual adaptation of foundation and pre-trained models in robotics Topic editors: Muhammad Burhan Hafez, Ali Ayub, Zirui Li Source: https://www.frontiersin.org/research-topics/80846/toward-lifelong-autonomy-advancing-continual-robot-learning-for-adaptable-intelligent-systems