Towards characterizing feasibility of Edge driven split-control in Bilateral Teleoperation of Robots

We report a practical implementation of a 5G Mobile Edge Computing (MEC) system that is used to run analytics-based predictive algorithms to detect and avoid certain unwanted events arising from the large latency between a robot and a remote human controller. This is the split control paradigm. We present a scenario where a haptic glove-controlled robotic hand is used to pick a rigid object. The application running on the edge continuously monitors the kinesthetic data relayed from the robotic hand, uses a standard machine learning algorithm to detect the onset of slip of the object, and then provides appropriate control signals back to the hand in order to avoid slip. We introduce the notion of edge-fiability of an application and demonstrate, by varying the gravitational force applied on the object being lifted, the limits of the edge-based prediction and control approach.