Description
Collaboration between humans and autonomous agents requires the ability to infer and adapt to other agents’ plans while effectively conveying one’s own intent. In some cases, our teammates’ actions early on can give us a clear idea of what the remainder of their plan is, that is, what action sequence we should expect; in others, they might leave us less confident, confused or even lead us to the wrong conclusion.
In this work, we use a Bayesian model of how people make such predictions in order to facilitate the interpretation of robot plans by human collaborators. We subsequently propose the concept of t-predictability to quantitatively describe an action sequence in terms of its easiness for expressing the entire plan. A t-predictable planner is then developed to generate action sequences that purposefully maximize the expected accuracy and confidence with which human observers can predict the overall plan from only the initial few actions. Through an online experiment and an in-person user study with physical robots, we find that t-predictable planner outperforms a traditional optimal planner in objective and subjective collaboration metrics. We believe that t-predictability will play a significant role for improving human-robot collaboration.