Robot Learning via Human Adversarial Games
This addresses the challenge of adversarial human interactions for deployed robotic systems, offering a novel approach to enhance robustness in manipulation tasks.
The paper tackles the problem of human adversarial behavior in robot learning by proposing a physical framework that uses human-applied perturbations to improve model robustness, resulting in significantly improved grasping success compared to self-supervised training.
Much work in robotics has focused on "human-in-the-loop" learning techniques that improve the efficiency of the learning process. However, these algorithms have made the strong assumption of a cooperating human supervisor that assists the robot. In reality, human observers tend to also act in an adversarial manner towards deployed robotic systems. We show that this can in fact improve the robustness of the learned models by proposing a physical framework that leverages perturbations applied by a human adversary, guiding the robot towards more robust models. In a manipulation task, we show that grasping success improves significantly when the robot trains with a human adversary as compared to training in a self-supervised manner.