Enabling Robust Human-Robot Cooperation through Flexible Fully Bayesian Shared Sensing

Abstract

Cooperative human-robot sensing has the potential to overcome many hurdles for networked field robotic applications, especially those with significant sensing, computational or communication constraints that make teleoperation or direct supervisory control difficult. The main idea is to augment robotic sensing horizons with the diverse and complementary capabilities of `human sensors' for solving difficult hybrid nonlinear state estimation and perception problems, although robots may still plan and control their own actions autonomously. In this work, we examine two related issues through the use of sketch-based and semantic codebook perceptual interfaces for probabilistic target search problems: (i) how should autonomous machines/robots assess the trustworthiness of human sensors?; (ii) what strategies can be used to generate human-machine `dialog' about complex uncertainties in random variables, without cognitively overburdening humans or undermining human trust in automation?