Fu-En Yang

Continuous video understanding doesn't require constant multimodal LLM compute: selectively waking a heavyweight reasoner via lightweight spatial filtering delivers real-time, causal anomaly reasoning without diluting rare events.

Online 3D reconstruction failure on long videos is not a representation collapse but an artifact of single-anchor pose extrapolation—and querying relative poses across multiple keyframes fixes it with just 1% parameter overhead.

PhysCaP enables robots to actively infer hidden physical properties, achieving superior manipulation efficiency without additional sensors.