Effects of Incoherence in Multimodal Explanations of Robot Failures

Providing explanations of a robot's behavior is a key enabler of trust in robots. Such explanations should be intuitive to people who are not experts in robotics. Prior research suggests that using multiple modalities to deliver explanations improves clarity. However, current methods for generating multimodal explanations neither assess nor ensure the coherence of the information across modalities. Here, we present an experiment to understand the effect of possible incoherence in multimodal explanations. We perform a user study asking participants to observe a series of robot failures and predict the reason for failure when provided with a controlled variation of multimodal explanations. Specifically, we present a methodology to compare incoherent and coherent explanations, aiming to understand their impact on perceiving robot failures.