SurgLAT: AI Framework for Autonomous Laparoscopic Camera Control

A novel AI framework named Surgical Latent Attention Tracking (SurgLAT) has been developed for the autonomous control of laparoscopic cameras in robotic surgery. This framework, outlined in a paper on arXiv (2608.07876), tackles the difficulty of interpreting the surgeon's intent in fast-changing surgical environments, where the focus is not a fixed object but a fluctuating attention state. SurgLAT utilizes a causal online architecture that incorporates a frozen DINOv3 encoder alongside a state-conditioned spatial token mixer to gather operative insights based on a memory-informed spatial prior. A selective causal latent memory module captures both immediate motion continuity and long-term surgical intent changes by dynamically accessing current, recent, and past latent states. The decoded latent attention state generates a probabilistic attention heatmap and identifies the operative area for subsequent endoscope navigation. This advancement signifies progress toward more autonomous and context-sensitive robotic surgery, which may enhance accuracy and lessen the workload for surgeons. The paper was submitted as a new entry on arXiv, identified as 2608.07876v1.