Xinyu Liu

Embedding reference tokens at semantic positions allows for unprecedented precision in multi-reference video editing, setting a new benchmark for instruction quality.

Grasp datasets can revolutionize robotic dexterity, enabling significant improvements in articulated tool use performance.

AERIS enables real-time, role-driven intelligence for aerial robots, adapting dynamically to resource constraints while maintaining high-performance navigation capabilities.

FlowTrain redefines VLM training efficiency, achieving up to 1.7x throughput improvements by decoupling execution and optimizing resource allocation.

SurgVista achieves unprecedented visual fidelity and interaction accuracy in surgical simulations, outperforming state-of-the-art models as prediction horizons extend.

Ling-2.6 and Ring-2.6 achieve unprecedented efficiency in agentic intelligence, enabling instant responses and deep reasoning at trillion-parameter scale.

Achieve high-fidelity video generation without compromising reasoning by progressively handing off generation from a lightweight, reasoning-aligned generator to a high-capacity pretrained generator in a shared latent space.

Subject-specific variability in biomedical time-series can be mitigated by explicitly aligning spectral structure, leading to a 6% F1-score improvement over existing methods.

Expert-level video aesthetics can be captured and improved using a hierarchical rubric and reward models trained with a progressive learning scheme.