Planning and Learning for Multi-Robot Coordination

The decreasing cost and increasing sophistication of robot hardware has created many new opportunities for teams of robots to be deployed to solve real-world problems. Although many algorithms have been proposed for multi-robot domains, the vast majority are specialized to match specific team or problem characteristics. Ideally, more general methods would exist for controlling multi-robot teams in a wide range of domains. In this talk, I will discuss such a general model for multi-robot coordination as well as planning and learning methods for automatically generating solutions. These methods provide principled solutions that optimize control and communication decisions while considering uncertainty in outcomes, sensors and the communication channel. To demonstrate the scalability and effectiveness of these methods, I will show results from a cooperative beer delivery problem with heterogenous ground robots and a package delivery problem with teams of aerial robots.