Yongxue Xu

Humanoid motion generation and editing can now be precisely controlled through typed kinematic events, enabling robust multi-turn revisions that preserve critical motion states.

Video segmenters suffer from persistent identity drift because they blindly write noisy observations to memory; gating state transitions through propose-verify graph reasoning preserves object history through extended occlusions without altering the underlying base model.

EviGDA is proposed, an Evidence-Augmented Graph Domain Adaptation framework that complements graph representation adaptation with a graph-free local expert, preserving complementary evidence without joint training, learned routing, or target pseudo-labels.