Localization between a swarm of AUVs can be entirely estimated through the use of range measurements between neighboring AUVs via a class of techniques commonly referred to as sensor network localization. However, the localization quality depends on network topology, with degenerate topologies, referred to as low-rigidity configurations, leading to ambiguous or highly uncertain localization results. This paper presents tools for rigidity-based analysis, planning, and control of a multi-AUV network which account for sensor noise and limited sensing range. We evaluate our long-term planning framework in several two-dimensional simulated environments and show we are able to generate paths in feasible time and guarantee a minimum network rigidity over the full course of the paths.
In many environments, such as underwater, it can be difficult to get reliable localization using visual features. This could be due to the environment being highly repetitive or there being a general sparsity of features. In these situations the performance of traditional localization techniques tends to degrade and can give highly incorrect errors.
A key direction in my research is leveraging the use of inter-agent ranging, a measurement which does not depend on environmental observation, to aid the multi-robot localization task. However, there is a caveat, the quality of range-based localization varies based on the connectivity and spatial arrangement of the network of agents.
In this work I developed a planner which allowed for networks of agents to efficiently plan trajectories that allowed them to stay in configurations which were favorable for range-based localization. Read more about it here!