Task Offloading in Mixed IoT of AUVs and ASVs

We address a mixed IoT system that is composed of autonomous underwater and surface vehicles (AUV and ASV). They employ sonar and RF signals to communicate and cooperate together while performing underwater missions. The presence of ASVs not only helps extend the communication range. To lengthen the underwater operation time, the formed vehicle-to-vehicle network also proactively offloads heavy computation jobs from resource limited AUVs to more powerful ASVs. Task offloading nevertheless results in exchanging extra data between them, and increasing the communication latency to handled events. Finding the optimal offloading policy is extremely difficult given that there are a large number of factors in the system state to consider. This study hence proposes a deep reinforcement learning - based strategy for distributing heavy tasks among multiple ASVs. Task mapping is optimized with respect to energy consumption and responsiveness. The framework also jointly deals with routing tasks to the best ASVs, as assessing the comprehensive cost of a task distribution policy.