ECHO-G: Embodied Co-speech Humanoid mOtion Generation
Published in arXiv preprint arXiv:2609.39575, 2026
Recommended citation: Y. Li*, P. Gao*, M. Wang, S. Shen, S. Yang and H. Xu. ECHO-G: Embodied Co-speech Humanoid mOtion Generation. arXiv preprint arXiv:2609.39575, 2026. (* equal contribution) https://arxiv.org/abs/2609.39575
ECHO-G lets humanoid robots gesture while they talk. It jointly uses text and speech latents to generate semantically rich full-body motion that lands on the beat, runs end-to-end whole-body control, and fits on an embedded Jetson Orin AGX. Joint work with Mondo Robotics and the HKUST Aerial Robotics Group.
Project page · Code · Video