Cong Huang

SIEVE reveals that leveraging reusable structures in demonstration data can lead to more efficient and effective imitation learning, outperforming full-data training with significantly less input.

Ink3D achieves a breakthrough in 3D asset creation, enabling the generation of complex textures that were previously unattainable with conventional methods.

Human-as-Humanoid achieves a staggering 4.8–7.2x increase in demonstration throughput, transforming how humanoid robots learn from human actions.

EgoPriMo enables humanoid robots to generate and forecast complex motions interactively using just egocentric observations and high-level language prompts.

Many VLA models fail to translate semantic understanding into accurate action selection, often performing at near-random levels despite successful grasping.

Robots get a spatial-temporal reasoning boost with STARRY, a world model that aligns future predictions with action generation, leading to a significant jump in manipulation success.

Onboard perception and pre-trained whole-body control can enable humanoid robots to perform dynamic racket sports like tennis without external motion capture or task-specific retraining.

Achieve robust humanoid task execution in complex environments by turning high-level language instructions into verifiable, geometrically-grounded task programs that can recover from failures.

Achieve >97.5% of full-data VIT performance with only 16% of the data using ScalSelect, a surprisingly effective and scalable training-free data selection method.