AUTHOR=Shi Lei , Copot Cosmin , Vanlanduit Steve 

TITLE=A Bayesian Deep Neural Network for Safe Visual Servoing in Human–Robot Interaction

JOURNAL=Frontiers in Robotics and AI

VOLUME=Volume 8 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/robotics-and-ai/articles/10.3389/frobt.2021.687031

DOI=10.3389/frobt.2021.687031

ISSN=2296-9144

ABSTRACT=Safety is an important issue in Human-Robot Interaction (HRI) applications. Various researches have focused on different levels of safety in HRI. If a human/obstacle is detected, a repulsive action can be taken to avoid the collision. Common repulsive actions include distance based, potential field based and safety field based. Machine learning based approaches are less explored regarding the selection of the repulsive action. In this paper, we describe a system that can avoid collision with human hands while the robot is executing an Image Based Visual Servoing (IBVS) task. We use a Bayesian Deep Neural Network (DNN) to learn the repulsive pose for hand avoidance. The Bayesian DNN has adequate accuracy.
The Predictive Interval Coverage Probability of the predictions along x, y and z direction are 0.84, 0.95 and 0.96. In the space which is unseen in the training data, the Bayesian DNN is also more robust compared to a regression ResNet. The system is implemented with a UR10 robot. We also compare the repulsive pose inferred by the Bayesian DNN model and the repulsive pose which is in the opposite direction towards the hand. The result shows the inferred repulsive pose can let IBVS converge faster.