Blind Spots of Objective Measures

Abstract

Tactile Internet (TI) enables the transfer of human skills over the Internet, enabling teleoperation with force feed-back. Advancements are being made rapidly at several fronts to realize a functional TI soon. Generally, TI is expected to faithfully reproduce operator's actions at the other end, where a robotic arm emulates it while providing force feedback to the operator. Performance of TI is usually characterized using objective metrics such as network delay, packet losses, and RMSE. Pari passu, subjective evaluations are used as additional validation, and performance evaluation itself is not primarily based on user experience. Hence objective evaluation, which generally minimizes error (signal mismatch), is oblivious to subjective experience. In this paper, we argue that user-centric designs of TI solutions are necessary. We first consider a few common TI errors and examine their perceivability, The idea is to reduce the impact of perceivable errors and exploit the imperceivable errors to our advantage, while the objective metrics may indicate that the errors are high. To harness the imperceivable errors, we design Adaptive Offset Framework (AOF) to improve the TI signal reconstruction under realistic network settings. We use AOF to highlight the contradictory inferences drawn by objective and subjective evaluations while realizing that subjective evaluations are closer to ground truth. This strongly suggests the existence of 'blind spots of objective measures'. Further, we show that AOF significantly improves the user grade, up to 3 points (on a scale of 10) compared to the standard reconstruction method.