Enabling Adaptive Intelligence in Cloud-Augmented Multiple Robots Systems

Multiple robot systems (MRSs) have been widely used in both military and civil applications. In recent years, machine learning approaches, especially deep learning, have been adopted to enhance the ability of MRSs in many scenarios, such as path planning, cooperative inference, and situation awareness. However, the robot individuals are usually severely limited in resources and they cannot afford the high computation requirements of deep learning approaches. Cloud-augmented multiple robot systems try to solve this problem by performing heavily computational tasks on the central cloud and performing only simple tasks on the robots. In this strategy, the robots need to transmit their collected data to the cloud, which might cause very long transmission latency and thus are not be suitable for many critical applications that have stringent requirements on response delay. The recently emerging cooperative deep inference can be a potential solution to enabling adaptive intelligence in cloud-augmented MRSs. Rather than transmitting all the data to the cloud and executing all computational tasks on the cloud, cooperative deep inference adaptively determines whether to perform some tasks on the robot before sending the data to the cloud, aiming to reduce the volume of data to be transmitted and reduce the latency. Moreover, the robot can adaptively determine which part of the task should be executed on the robot and which part should be executed on the cloud to optimize the time or energy efficiency subject to some task specific requirements. In this paper, we overview the recent progresses in cooperative deep inference and propose a framework that enables adaptive intelligence in cloud-augmented MRSs. We then demonstrate with a target detection scenario that the proposed framework can effectively reduce response latency.