A robot on the shoulder: Coordinated human-wearable robot control using Coloured Petri Nets and Partial Least Squares predictions

Abstract

A wearable robot secured on the shoulder of a human is developed for assisting its wearer in the execution of tasks in the overhead workspace. Installing a ceiling panel is an example of such a task. During this task the robot can hold the panel or collaborate actively with the human in order to fix the panel with the appropriate equipment. This wearable robot, termed "Supernumerary Robotic Limbs", works closely with the human to streamline the operation and reduce the human workload. First, the design concept of the Robot-on-the-Shoulder system is described, and a new approach to the coordinated control between the wearable robot and the human is presented. A graphical task process representation based on Coloured Petri Nets (CPN) is used to model the concurrent and distributed nature of the human-robot system, which comprises two human hands and two robot hands. The CPN framework is then extended to a type of hybrid control system by imbedding local dynamic controllers in the Transition nodes of the CPN model. Each local dynamic controller collects sensor signals relevant to the target transition and makes a predictive control decision. This allows the robot on the shoulder to take a proactive and preemptive action as well as to confirm a successful Transition. The control parameters for these algorithms are tuned based on "teaching-by-showing" techniques using human demonstration data. Partial Least Squares is used for extracting significant sensor signals from high-dimensional sensor data in order to do real time predictions and control. A prototype robot-on-the-shoulder system is built, and the CPN hybrid control is implemented and tested for a ceiling panel installation task.