Towards Surgical Task Planning from Text

This Chapter aims to introduce a use case and show how, after having defined some constraints on the language, the SRL method, together with other rule-based methods, can help the engineer write the logic rules needed to define the plan for the robot. In particular, we propose AUTOMATE (lAngUage To lOgic teMplATEs), a pipeline that helps the translation of natural language instructions to linear temporal logic (LTL). This chapter describes the use of a controlled language and a general English language model: state of the art in autonomous surgical robotics mainly uses two tasks as benchmarks, i.e., peg transfer and tissue retraction, presented later. Although performed with surgical robots, they are still simplified tasks whose description does not require particularly complex surgical expressions. Nonetheless, as the surgical robotics community moves to more realistic and complex benchmarks requiring more specialized surgical language, the models developed in the previous chapters will allow for more in-depth language understanding. This Chapter aims to empirically show that SRL technology is a fundamental tool for extracting logical entities from procedural natural language text and to highlight the technological deficiencies for achieving a completely automated translation.