ABSTRACT
This chapter discusses a general methodology to address safety in human-robot interaction. The safety issues are understood as conflicts in the multi-agent system. To resolve the conflicts, the robot’s behavior was constrained according to the “social norm” and the uncertainties it perceived for the other agents (including humans and other robots), which constitutes the safety-oriented behavior design. Two algorithms are discussed under the framework: the safe set algorithm (SSA) and the safe exploration algorithm (SEA). In both algorithms, the robot calculates the optimal action to finish the task while staying safe with respect to the predicted human motion. The difference is that SEA actively tracked the uncertainty levels in the prediction and incorporated that information in robot control, while SSA does not. Both methods have their advantages. A method to combine the two algorithms is discussed to take advantage of both algorithms. Several case studies are presented and demonstrate the effectiveness of the methodology.
