Search papers, labs, and topics across Lattice.
Peking University, Affiliation:
5
0
4
Explicit 3D skeletal representations unlock parameter-efficient robotic manipulation, achieving 85.9% success on LIBERO-Plus with half the parameters of prior methods.
STAR achieves superior interaction recognition by seamlessly integrating skeletal and visual data, enabling efficient inference on skeletons alone.
VideoLLMs may excel at recognizing actions but often hallucinate human motions, revealing a critical flaw in their design.
Visual goal prototypes can boost robot manipulation success rates by up to 17% compared to text-based instructions, revealing the power of leveraging action-free demonstrations.
Unlock the power of MLLMs for structured data like human skeletons with a differentiable rendering approach that allows end-to-end training.
