SurgLAT for Latent Attention Tracking in Robotic Surgery

SurgLAT uses a frozen DINOv3 encoder and a causal latent memory module to model a surgeon's evolving operative intent during laparoscopic procedures. It transforms latent attention states into probabilistic heatmaps to guide autonomous endoscope control in dynamic surgical environments.