Semi-Autonomous Cooperative Tasks in a Multi-Arm Robotic Surgical Domain

This paper addresses collaborative multi-arm task execution in a robot-aided surgical domain. The proposed approach aims to assist the surgeon by recognizing their intent and autonomously controlling some of the robotic arms. We lean on Learning from Demonstration paradigm, where knowledge is gained by observing humans performing a task. Example cooperative movements are gained in a leader/follower scenario with an operator controlling both arms. Example leader arm movements are encoded in a hierarchical database, which consists of a binary tree enhanced with weighted directed graphs. It is used for online leader movement recognition and prediction. Example follower arm movements are encoded as a set of Dynamic Movement Primitives, which are used for online synthesis of appropriate follower movements, based on recognition and prediction of leader movements. The proposed approach is evaluated on a dual arm robotic surgical system, showing that the operator can perform the cooperative task by only controlling the leader arm. The proposed database structure enables recognition and prediction fast enough for real-time task execution, and the novel approach of speed adaptation ensures both arms are in sync regardless of the operators speed of execution.