In this paper, we propose a new paradigm for human-robot collaboration. In this paradigm, the collaboration properly takes advantage of the superior visual performance of the humans and the field exploration capabilities of robots, allowing the robot to only ask humans for help when needed. More specifically, we consider a robotic field exploration and classification task with limited communications with a human operator and under a given energy budget. By learning the visual performance of humans probabilistically, we show how the robot can optimize its path planning, sensing, and communication with humans. More specifically, we show when the robot should ask humans for help, when it should rely on its own judgment and when it should gather more information from the field. In order to show the performance of our framework, we then collect several human data using Amazon Mechanical Turk. Our simulation results with real data then confirm that our approach can save the resources considerably. They further reveal interesting behaviors in terms of when to ask humans for help, which we also mathematically characterize.
展开▼
