Learning to Plan with Personalized Preferences

ICLR 2025 Conference Submission5283 Authors

Keywords: Embodied AI, Personalized Preference, Human-AI interaction

Abstract:

Understanding and adapting to human preferences is essential for the effective integration of artificial agents into daily human life, particularly as AI becomes increasingly involved in collaboration and assistance roles. Previous studies on preference recognition in embodied intelligence have largely adopted a generalized yet non-personalized approach. To fill in this gap, our research focuses on
empowering embodied agents to learn and adapt to individual preferences, a task complicated by the challenges of inferring these preferences from minimal observations and requiring robust few-shot generalization. To facilitate future study, we introduce PbP, an embodied environment that supports hundreds of diverse preferences ranging from complex action sequences to specific sub-actions. Our
experiments on PbP reveal that while symbol-based approaches show promise in terms of effectiveness and scalability, accurately inferring implicit preferences and planning adaptive actions from limited data remain challenging. Nevertheless, preference serves as a valuable abstraction of human behaviors, and incorporating preference as a key intermediary step in planning can significantly enhance the personalization and adaptability of AI agents. We hope our findings can pave the way for future research on more efficient preference learning and personalized planning in dynamic environments.

Primary Area: datasets and benchmarks

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 5283