Online Formal Verification of Robot Trajectories for Guaranteed Safety of Humans

Guaranteeing safe behaviour of robots is not easy, especially in the presence of humans, whose behaviour is always unpredictable. We show how a frequently-updating, continually-verifying controller can guarantee safety of nearby humans in a formal way while maintaining efficiency of the robot. Two different models of the human behaviour are presented: a model based on extreme human motion as specified in ISO standards, and another, kinematic model parametrised by test data from a range of human subjects performing extreme movements. We implement the controller in an experimental setup, and find that both models allow the robot to work efficiently when the human is safe and stop when the human is in danger of colliding with the robot; the ISO-based model is less cautious and sometimes does not account for fast human motion. The kinematic model, which accounts for all motion, is more restrictive towards robot motion but nevertheless allows the robot to operate when the human is not in danger.