How Continual Learning and Generative AI are transforming Robotics
Abstract
Continual Learning (CL) and Generative AI (GenAI) are two of the most transformative developments in modern machine learning. CL equips systems with the ability to learn from sequential data over time without catastrophic forgetting, while GenAI - through Large Language and Vision Models, diffusion models, and other generative architectures - is redefining how robots perceive, reason, and act with minimal supervision. This workshop aims to explore the intersection of CL and GenAI in the context of robotics, focusing on how these paradigms can be effectively integrated to enable long-term autonomy, adaptability, and scalable intelligence. We will also address the key challenges of deploying these technologies in embodied systems, including computational constraints, safety considerations, and the need for robust generalization in the real world. Topics of interest include: (i) continual learning algorithms tailored for robotics, (ii) generative models for perception, planning, and control, (iii) benchmarks and evaluation protocols, and (iv) real-world applications across service, assistive, and industrial robotics. The workshop will feature invited talks, a poster and pitch session, and an open discussion to foster collaboration between the AI and robotics communities.
Full text
I-RIM 3D 2025 - Workshop Proposal How Continual Learning and Generative AI are transforming Robotics Organizers: Enrico Donato∗, Scuola Superiore Sant’Anna, Pisa, enric[email protected] Vincenzo Lomonaco, Universit`a di Pisa, Pisa, vincenzo.lomonac[email protected] Egidio Falotico, Scuola Superiore Sant’Anna, Pisa, egidio.falotic[email protected] ∗corresponding organizer 1. Workshop Format The workshop will feature invited talks and an open discussion, a pitch and a poster sessions for selected abstracts. 2. Workshop Objectives and Main Topics Continual Learning (CL) and Generative AI (GenAI) are reshaping the future of intelligent systems. CL enables robots to adapt continuously to changing environments without forgetting but transferring prior knowledge, while GenAI - through large language and vision models, and generative models as contrastive and diffusion - unlocks new capabilities in perception, reasoning, and behaviour generation. Together, these approaches promise long-term autonomy,adaptability, and scalable intelligence for real-world robotics. However, integrating CL and GenAI into robotic systems remains challenging due to issues such as real-time constraints, embodiment, safety, and transferring models from simulation to physical platforms. This workshop aims to connect the Artificial Intelligence and Robotics communities to: •identify open challenges in deploying CL and GenAI in robotics; •share recent algorithmic and systems-level innovations; •showcase real-world applications of adaptive and generative robotic systems; •promote reproducible benchmarks and cross-disciplinary collaboration. Main topics include: •Continual Learning Algorithms: Online learning, memory systems, and reinforcement learning for resource-constrained robots. •Generative Models in Robotics: Use of LLMs, generative models, and VLMs for planning, control, and interaction. •Evaluation and Safety: Benchmarks for continual and generative learning, domain shifts, uncertainty, and robustness. •Real-World Applications: Deployment in service, industrial, assistive, and field robotics where adaptation is essential. 3. Keywords Continual Learning, Generative AI, Robotics, Lifelong Learning, Online Learning, Foundation Models, Large Language Models, Adaptive Systems, Embodied AI 4. Preferred Date October 17th, 2025 Workshop organizers and confirmed invited speakers can guarantee their presence only on that day, because of other incoming academic events. 5. List of Speakers •Dr. Vincenzo Lomonaco (Universit`a di Pisa) Confirmed Title: The Future of Continual Learning in the Era of Foundation Models: Three Key Directions •Prof. Matteo Saveriano (Universit`a di Trento) Confirmed Title: Towards Safe Continual Learning from Demonstration 1 2025 I-RIM Conference October 17-19, Rome, Italy ISBN: 9788894580570 10.5281/zenodo.17629612 18
•Dr. Egidio Falotico (Scuola Superiore Sant’Anna) Confirmed Title: Continual Learning and Generative AI in Soft Robotics •Prof. Tatiana Tommasi (Politecnico di Torino) Confirmed Title: Generative AI for Robot Grasping •Prof. Loris Roveda (Politecnico di Milano) Confirmed Title: Unleashing Autonomy for Real-World Robots 6. Expected Number of Attendees We anticipate approximately 30+ participants with expertise or interest in Robotics and Artificial Intelligence. 7. Target Audience This workshop targets researchers, engineers, and practitioners from both academia and industry working at the intersection of Artificial Intelligence and Robotics. It is particularly relevant for those developing adaptive robotic systems, focusing on continual learning and generative AI, real-world deployment, safety, and long-term autonomy. The event also welcomes students and early-career researchers interested in these topics. 8. Workshop Outcomes The workshop aims to foster collaboration between the two communities and establish a foundation for ongoing interdisciplinary exchange, from a national to an international viewspan. Expected outcomes: •presentation slides and poster materials shared via the workshop webpage; •asummary report highlighting key discussions, challenges, and future directions; •follow-up activities such as collaborative projects or community-driven benchmark proposals. 9. Plan for Special Issue or Session We plan to propose a special issue in a journal relevant to the I-RIM community (e.g., IEEE Robotics and Autonomous Systems or IEEE Computational Intelligence related journals, Frontiers in Robotics and AI) focused on Continual Learning and Generative AI for Robotics. Workshop speakers and contributors will be invited to submit extended versions of their work. The special issue will encourage submissions that emphasize real-world deployment, empirical evaluation, and interdisciplinary collaboration. Additionally, a focus paper on this topic on a high-impact journal is programmed, highlighting key challenges and future directions on this intersection. 10. Call for Extended Abstracts We will issue a call for extended abstracts (2–4 pages) on relevant topics (e.g. continual robotics learning, generative models for robotics, machine learning for robotics). Submissions will be peer-reviewed by the organizing committee. Selected contributions will be invited to present a short pitch and present a poster during the workshop poster session. The call will be disseminated through academic mailing lists (e.g., I-RIM, academic institutions), social media, and the workshop webpage. Submissions will be handled via a simple platform (e.g., Google Forms). Important dates and instructions will be provided on the workshop webpage. 11. Workshop Webpage The workshop webpage is currently under development and will be hosted as a page of the BRAIR Lab website: https://www.santannapisa.it/en/institute/biorobotics/brair-lab. The page will include the program, speaker details, call for abstracts, submission instructions, and relevant updates. The BRAIR Lab website already includes an example from a previous workshop. 12. Notes on Registration We plan to use the two free one-day registrations to cover the participation of confirmed invited speakers who are not otherwise funded. Additional speakers and attendees will register through the standard I-RIM process, supported by their own funds. We will clearly communicate the registration procedure to all contributors and manage coordination to ensure smooth participation logistics. 19