Towards Replicable Experiments on Distributed and Adaptive Shared Control Systems
IEEE Robotics & Automation Magazine

ABSTRACT Much work on robots aimed at real-world applications fall in the large segment between teleoperated and fully autonomous systems. Such systems are characterized by the close coupling between the human operator and the robot, in principle allowing the agents to share their particular sensing, adaptation and decision-making capabilities. Replicable experiments can advance the state of the art of such systems, but pose practical and epistemological challenges. For example, the trajectory of the system is governed by the adaptation both in the human and the robot agent. What do we need besides (or instead) of "data sets" for such a system? The degree of similarity between "comparable" experiments and the exact meaning of "replication" need to be clarified. Here we explore replication of a distributed and adaptive shared control for an assistive robot manipulator. We attempt a methodological approach for reporting two virtual human experiments on the system: Modelling the complete human-robot binomial, deriving closed-loop performance metrics from the models, and openly publishing the results and experiment implementations.