The FarSight Lab, led by Professor Lorenzo Torresani, develops AI systems for perception, anticipation, and planning in dynamic worlds. Our research spans egocentric video understanding for wearable assistants, large-scale learning from noisy video corpora, and strategic reasoning over multi-agent scenes such as team sports.

A common thread across our work is the move beyond perception toward strategic understanding: building models that not only observe dynamic scenes but reason about cause and consequence, simulate alternative futures, and plan over long horizons. We design models that learn directly from rich signals present in raw video, and benchmarks that test whether models truly understand what they see.