FarSight Lab
Seeing beyond the moment.
FarSight Lab
Khoury College of Computer Sciences
Northeastern University
We build AI systems that understand video, anticipate what comes next, and plan over long horizons, for first-person assistants, multi-agent strategic settings, and physical AI. Our work is driven by a central idea: that strategic reasoning over dynamic scenes is the next frontier beyond perception.
news
| 2026 | New paper: How You Move Tells What You’ll Do: Trajectory-Conditioned Egocentric Prediction. [project page] |
|---|---|
| 2026 | New paper: RECIPE: Procedural Planning via Grounding in Instructional Video. [project page] |
| 2026 | New paper: EvoGround: Self-Evolving Video Agents for Video Temporal Grounding with Minjoon Jung and Byoung-Tak Zhang. [arXiv] [project page] |
| 2026 | We have several openings for postdocs, visiting researchers, and PhD students to work on embodied AI, video understanding, and multimodal learning. Prospective applicants should contact Lorenzo Torresani with a CV and a one-page research statement. |
| Aug 2025 | Lorenzo appointed President Joseph E. Aoun Chair at Northeastern University. |
| June 2025 | Our state-space video model BIMBA won first place in the EgoSchema Challenge at CVPR 2025. [project page] |
| June 2025 | Three of our papers received Distinguished Paper Awards at the CVPR 2025 EgoVis Workshop: Video ReCap, Ego4D Goal-Step, and HierVL. [awards page] |
| 2025 | PerceptionLM: Open-Access Data and Models for Detailed Visual Understanding accepted to NeurIPS 2025 as spotlight (<3.5%). [arXiv] |
| 2025 | Enrich and Detect: Video Temporal Grounding with Multimodal LLMs accepted to ICCV 2025 as highlight (<2.5%). [project page] |
| 2025 | Two papers accepted at CVPR 2025: BIMBA and ViTED. |