Surgical Task Planning and Learning

This Chapter describes how “top-down” knowledge can be used to plan a robotic task. A crucial concept in modern cognitive systems is deliberation, which is the ability to make decisions that are motivated by reasoning on the available resources, i.e. the capabilities of the robot, the actual description of the environment, and the given mission. Deliberation marks a fundamental difference between automatic and autonomous robotic systems, the former merely executing pre-defined sequences of operations, with possible small adaptations to the sequence, while the latter actively reasoning on their goals and taking actions (possibly within pre-designed bounds, e.g., for safety) accordingly. To this aim, our proposed methodology is based on the use of logic programming to represent human-interpretable concepts and relations about the planning domain. In Chap. 2 we presented the process of extracting knowledge about surgical procedures from textbooks and scientific papers, and in Chap. 4 we show how the task sequence of a plan is extracted from text using techniques of Natural Language Processing (NLP). In this Chapter we present the details of how the deliberation process was carried out and demonstrated in the task shown in Fig. 12. This Chapter is organised in two main sections. First, we introduce action languages as a formal representation method for planning domains. We then show how to express an action language representation in answer set programming, a state-of-the-art logic programming formalism supporting contingent planning and knowledge revision via non-monotonic logics. Then, we consider the problem of planning knowledge extraction from task execution records, combining unsupervised action identification, semantic labeling and inductive logic programming. For clarity, the proposed approaches are exemplified in the context of a paradigmatic surgical pick-and-place training exercise, the ring transfer task, inspired from the peg transfer task described in the Fundamentals of Laparoscopic Surgery. We report preliminary results of our methodology in the end of the Chapter, finally discussing some limitations and future research directions to overcome them.