Autonomous adaptive acoustic relay positioning

Abstract

We consider the problem of maximizing underwater acoustic data transmission by adaptively positioning an autonomous mobile relay so as to learn and exploit spatial variations in channel performance. The acoustic channel is the main practical method of underwater wireless communication and improving channel throughput and reliability is key to improving the capabilities of underwater vehicles. Predicting the performance of the acoustic channel in the shallow-water environment is challenging and usually requires extensive modeling of the environment. However, a mobile relay can learn about the unknown channel as it transmits. The relay must balance searching unknown sites to gain more information, which may pay off in the future, and exploiting already-visited sites for immediate reward. This is a classic exploration vs. exploitation problem that is well-described by a multi-armed bandit formulation with an elegant solution in the form of Gittins indices. For an autonomous ocean vehicle traveling between distant waypoints, however, switching costs are significant. The multi-armed bandit with switching costs has no optimal index policy, so we have developed an adaptation of the Gittins index rule with limited policy enumeration and asymptotic performance bounds. We describe extensive shallow-water field experiments conducted in the Charles River (Boston, MA) with autonomous surface vehicles and acoustic modems, and use the field data to assess performance of the MAB decision policies and comparable heuristics. We find the switching-costs-aware algorithm offers superior real-time performance in decision-making and efficient learning of the unknown field.