Vector Institute

“SeededGrasp: Language-Guided Grasping in Complex Scenes with Multiple Embodiments”

“The result: 72% success in simulation and 78% in real-world experiments, with the VLM used zero-shot without fine-tuning or few-shot examples (Section 4.3).”

“Maria Attarian, University of Toronto / Vector Institute / Google DeepMind — Co-author with a dual academic-industry affiliation. Notable because she is also a co-author on GeoMatch (reference [10]), the multi-embodiment grasping method that SeededGrasp builds upon and outperforms.”

“Igor Gilitschenski, University of Toronto / Vector Institute — Senior/corresponding author. His lab (UofT-ISC) produced this work and hosts the project page. Also co-authored GeoMatch [10] and GeoMatch++ [30], establishing a sustained research program in multi-embodiment grasping that this paper extends.”

“Researchers have demonstrated a working, self-replicating AI worm that uses stolen GPU compute to run local LLMs, plan attacks, and spread autonomously — with no dependency on external APIs. The worm achieves an overall end-to-end attack success rate of ~37%, with 88% self-replication success once a foothold is established.”