Active Replication for Centrally Coordinated Teams of Autonomous Vehicles

Abstract

Autonomous vehicles, drones in particular, are used to support a wide range of sensing and actuating missions. While these missions are typically coordinated by a human operator, it is attractive to automate this coordination through a computer program that retrieves information from the vehicles and issues commands to them according to the mission objectives. However, the fact that such a computer-driven system may interact with and affect the physical environment in a direct way, introduces several challenges. In particular, it is important to tolerate failures of the mission control computer as smoothly as possible, avoiding roll-backs that might lead to inconsistencies. To address this problem, we propose an active replication approach, ensuring that as long as at least one replica of the mission controller remains operational, the mission will progress in a consistent way and with full transparency for the mission program. We define the properties that should be satisfied to achieve the required consistency, and present system-level mechanisms that support both deterministic and non-deterministic mission programs. We then discuss a concrete implementation of the proposed approach for an existing programming framework targeting multi-drone applications. Finally, we give an analytical cost model for the communication overhead of the proposed approach, and report the actual execution delay incurred in our prototype implementation for indicative scenarios using a suitable simulation environment. © 2019 IEEE.