Xicheng Gong

DexEyeGrasp: Learning Dexterous Grasping in Clutter with FingerEye Observations

Co-first Author

Dexterous grasping in dense clutter using fingertip-camera observations, teacher–student learning, self-supervised geometric prediction, and residual visual adaptation for sim-to-real transfer.

Vision-Language Models Are Joint Denoisers in Robotic Action Generation

JADE brings the vision-language model into the action denoising loop, using noisy action latents, gated cross-attention, and clean-latent supervision to jointly refine robot actions.

UniWAM: Unified World-Action Model

Core Contributor

A unified world-action model combining physical reasoning, future visual generation, and action prediction through a tri-modal architecture.

RotVLA: Rotational Latent Action for Vision-Language-Action Model

Qiwei Li, Xicheng Gong, Xinghang Li, Peiyan Li, Quanyun Zhou, Hangjun Ye, Jiahuan Zhou, Yadong Mu

Continuous rotational latent actions in SO(n) for compositional action representations and robotic manipulation.

Extending Embodied Question Answering from Perception to Decision

Xicheng Gong, Qiwei Li, Peiran Xu, Yadong Mu

A large-scale embodied question-answering dataset and benchmark for perception, reasoning, and decision-making in dynamic environments.

RePlan-Bot: Multi-Level Replanning for Embodied Instruction Following

Xicheng Gong, Guozheng Sun, Peiran Xu, Yadong Mu

Multi-level replanning for embodied instruction following through online goal adjustment, structured exploration, and action correction.

NavQ: Learning a Q-Model for Foresighted Vision-and-Language Navigation

Peiran Xu, Xicheng Gong, Yadong Mu

A Q-model for vision-and-language navigation that captures long-horizon future semantics to support goal-oriented decisions.