Florian T. Pokorny

Code-generating agents can now transform a single multimodal surgical demonstration into verified, self-correcting manipulation loops on real da Vinci hardware, clearing viscoelastic tissue fragments with up to 95% success with minimal intervention.

FocusPool is introduced, an attention pooling module that selectively aggregates intermediate visual features according to their relevance to the robot's current proprioceptive context that captures task-progressive, control-relevant local information and is used directly for policy learning.