Published in ICLR, 2024
InfoCon is a self-supervised method for discovering manipulation concepts from unlabeled robot demonstration trajectories.
Recommended citation: Liu, Ruizhe, Qian Luo, and Yanchao Yang. "InfoCon: Concept Discovery with Generative and Discriminative Informativeness." The Twelfth International Conference on Learning Representations. https://arxiv.org/abs/2404.10606
Published in ICLR, 2025
AutoCGP is a framework for training robots to perform complex manipulation tasks by automatically discovering and using “manipulation concepts” (motor skills like grasping, pushing, etc.) without human annotations.
Recommended citation: Zhou, Pei, et al. "AutoCGP: Closed-Loop Concept-Guided Policies from Unlabeled Demonstrations." The Thirteenth International Conference on Learning Representations. https://openreview.net/pdf?id=9ehJCZz4aM
Published in NeurIPS, 2025
HiMaCon is a self-supervised framework that learns hierarchical “manipulation concepts” from unlabeled multi-modal robot demonstrations without human annotation.
Recommended citation: Liu, Ruizhe, et al. "$\textit {HiMaCon:} $ Discovering Hierarchical Manipulation Concepts from Unlabeled Multi-Modal Data." The Thirty-ninth Annual Conference on Neural Information Processing Systems. https://arxiv.org/pdf/2510.11321
Undergraduate course, HKU, IDS, 2025
