Call for Papers

Topics of Interest

The Continual RL workshop covers topics broadly spanning foundations, theory, algorithms, experimental protocols, and applications of CRL*. Topics include, but are not limited to:

• Foundations and Formulations: CRL problem framing, non-stationarity characterization, theoretical guarantees, and limitations.

• Stability-Plasticity Trade-off: brain-inspired and neuro-inspired mechanisms to balance stability and plasticity, etc.

• Training Pathologies: plasticity loss, dormant neurons, and other pathologies due to non-stationary data distribution in (C)RL.

• Test-Time Adaptation: in-context learning, meta-learning, planning, exploration, and test-time training.

• Transfer Learning: multi-task settings, domain adaptation, sim-to-real, offline-to-online, etc.

• Open-Endedness: training protocol design, including (automatic) curriculum learning, procedural environment design, etc.

• Multi-Agent RL: unknown co-agents, dynamic population, ad-hoc teamwork, etc.

• Hierarchical RL: non-stationarity in hierarchical RL agents, and option/skill discovery for open-ended environments.

• Continual RL for Foundation Models: RL post-training and test-time scaling, addressing catastrophic forgetting and adaptation.

• Evaluation Protocols and Benchmarks: metrics, benchmarks, and experimental protocols for measuring CRL performance.

• Other Related Areas: online learning, active learning, CRL applications, etc.

*CRL was first explored in the dissertation by Ring [1]. Continual learning [2] was studied alongside closely related themes that have been appearing under different names, including lifelong learning [3,4], never-ending learning [5,6], and open-ended learning [7], with slightly different formulations and emphases depending on context.

Review and Selection Process

• All submissions will be evaluated by the organizers and invited reviewers in accordance with a strict conflict-of-interest policy. Reviewers will not assess submissions from collaborators or individuals from the same institution; such papers will be reassigned to non-conflicted reviewers.

• All submissions will undergo a desk reject phase for completeness, length compliance, and relevance to the workshop scope.

• Withdrawn or rejected submissions will not be made public.

• Accepted papers will be presented as posters, with a subset selected for contributed talks.

• At least one co-author of each accepted paper is expected to register for RLC 2026 and present the work at the poster session.