Trust-Driven Privacy in Human-Robot Interactions

In this paper we present a trust-driven differential privacy implementation for private trajectory sharing in human-robot interactions. While differential privacy implementations depend on a privacy parameter that is typically set before runtime, there are a number of applications in which human users may not have any information about their robot interaction partners a priori, making it difficult to determine a reasonable privacy level for information sharing. To enable collaboration in scenarios with unfamiliar robots, we dynamically adapt a human user's privacy level when sending information to a robot by using a quantitative measure of trust. We develop a trust model that reflects a robot's level of cooperation over time and captures key features of trust from both the psychological and human-robot interaction communities. To characterize our framework and its performance, we quantify the amount of information a robot can gain as a function of its cooperation, and we present bounds on the level of cooperation needed to attain a desired level of trust (and therefore privacy) over time. Simulation results are provided to illustrate this trust-driven private information sharing scheme.