Towards Fast Visual Explanations of Local Path Planning with LIME and GAN
As robots become a more significant part of humans’ daily lives, bridging the gap between robot actions and human understanding of what robots do and how they make their decisions becomes challenging. We present an approach to local navigation explanation based on Local Interpretable Model-agnostic Explanations (LIME), a popular approach from the Explainable Artificial Intelligence (XAI) community for explaining individual predictions of black-box models. We show how LIME can be applied to a robot’s local path planner. Moreover, we show how the General Adversarial Network (GAN) can be trained and used for fast explanation generation. We also analyze the quality and runtime of GAN explanations and present a tool for visualizing these explanations online as the robot navigates.