Optimally Persistent Formation of AUVs With Model Uncertainty and Unknown Interaction Topology

Formation control of autonomous underwater vehicles (AUVs) has been regarded as the basis of many sophisticated marine missions. However, the complex marine environment and the weak acoustic communication on AUVs make it hard to achieve the formation task. This paper attempts to overcome the above challenge from graph theory and intelligent learning perspectives. A local topology estimator is first designed by observing the coupled state evolution of AUVs, such that the unknown interaction relationship of AUVs can be inferred on the basic of local sensing. Based on this, we adopt the graph direction and contraction to generate an optimally persistent topology for AUVs, whose aim is to reduce the communication redundancy and guarantee the topology connectivity. With the optimized network topology, a model-free inverse reinforcement learning (IRL) formation controller is developed for AUVs to keep the desired formation shape. The innovations can be summarized as follows: 1) the local topology estimator can reveal the interaction topology relationship of AUVs with multiple degrees of freedom (DOF); 2) the optimally persistent topology can balance energy efficiency and topology connectivity as compared to the neighboring rule-based solutions; 3) the IRL-based formation controller has better adaptability to the underwater unknown environment as compared to the traditional reinforcement learning solutions. Finally, simulation and experimental results are both conducted to verify the effectiveness of our solution.