Human2LocoMan: Learning Versatile Quadrupedal Manipulation with Human Pretraining

Yaru Niu, Yunzhe Zhang, Mingyang Yu, Changyi Lin, Chenhao Li, Yikai Wang, Yuxiang Yang, Wenhao Yu, Tingnan Zhang, Zhenzhen Li, Jonathan Francis, Bingqing Chen, Jie Tan, Ding Zhao

Workshop Statement: Our work leverages intuitive human behaviors and egocentric demonstrations to teach robots complex tasks in a scalable and flexible manner. Instead of designing robot-specific policies from scratch, we treat human input as a rich, structured source of prior knowledge to pretrain a cross-embodiment policy. This framework not only improves learning efficiency and robustness but also enables generalization to novel and out-of-distribution tasks—highlighting the potential of human-robot co-learning systems for real-world robotic deployment. Furthermore, our system equips conventional quadrupedal robots with versatile manipulation capabilities through the use of lightweight loco-manipulators, significantly extending their functionality beyond locomotion. This advancement supports the development of compact, user-friendly quadrupeds capable of meaningful manipulation in everyday human-centered environments.

Keywords: Quadrupedal Manipulation, Cross-Embodiment Learning, Teleoperation and Data Collection

Abstract: Quadrupedal robots have demonstrated impressive locomotion capabilities in complex environments, but equipping them with autonomous versatile manipulation skills in a scalable way remains a significant challenge. In this work, we introduce a system that integrates data collection and imitation learning from both humans and LocoMan, a quadrupedal robot with multiple manipulation modes. Specifically, we introduce a teleoperation and data collection pipeline, supported by dedicated hardware, which unifies and modularizes the observation and action spaces of the human and the robot. To effectively leverage the collected data, we propose an efficient learning architecture that supports co-training and pre-training with multimodal data across different embodiments. Additionally, we construct the first manipulation dataset for the LocoMan robot, covering various household tasks in both single-gripper and bimanual modes, supplemented by a corresponding human dataset. Experimental results demonstrate that our data collection and training framework significantly improves the efficiency and effectiveness of imitation learning, enabling more versatile quadrupedal manipulation capabilities. Our hardware, data, and code will be open-sourced at: https://human2bots.github.io.

Submission Number: 31