A Model-based Approach to Supervisory Automata for Multiple-Controller Autonomous Systems

Anthony Rodriguez (University of Arizona), Amanda Pyryt (University of Maryland, Baltimore County)

Robust controller design is a fundamental necessity for the design of autonomous driving agents. Piloting a vehicle presents a unique control problem because of the difficulty in designing a control strategy that can perform optimally in a diverse array of situations. To combat this problem, a supervisory agent is normally introduced to select the optimal controller for a situation from a set of candidate controllers, allowing a control system to adapt to changing environments and objectives. Although these designs are effective in expanding the capability of an autonomous driving system, they often require long development periods and arduous effort in defining sound state transition systems for new maneuvers while doing little to address the difficulty of accomplishing disparate tasks with the same controller. We propose a solution to this issue through design of a modeling language used for describing and implementing switched-control systems. We will show that low-level code for a finite automata multi-controller system implemented with any software framework can easily be generated from high-level, graphical representations with the help of a modeling tool such as Generic Modeling Environment (GME). We show that an autonomous control system can be easily field-reconfigured to accomplish new tasks by performing a system reconfiguration which allows for a fully autonomous ground vehicle to detect and react to an obstacle in its environment. We will also discuss how these methods can be used to produce more robust systems from existing implementations.