Distributed Quadratic Programming for Safe Human Multirobot Interaction Under Task Constraints

Human multirobot interaction enables humans to work in a shared workspace with multiple robots that conduct cooperative tasks such as cotransporting a rigid object. To guarantee human safety, each robot has to execute unplanned motions via human-in-the-loop control. However, the robot's compliant motions with humans may violate the constraints imposed by the cooperative task with other robots. A distributed controller that guarantees human safety and task constraints is still a challenge. This article proposes a control framework based on distributed quadratic programming (QP). First, safety control sets based on local and global inequality constraints are formulated using control barrier functions to guarantee human safety. Furthermore, task control sets based on local and global equality constraints are formulated to achieve the constraints of the cooperative task. Then, distributed QPs are constructed by separating the global equality and inequality constraints into local constraints with discrepancy variables. Finally, a recurrent neural network converging to the local optimality of a distributed QP is designed, and adaptation laws of discrepancy variables using only exchanged information between adjacent robots are designed to make the distributed QPs reach global optimality. The effectiveness of the proposed control framework is verified in simulation and experiment.