Online Kinematic and Dynamic Parameter Estimation for Autonomous Surface and Underwater Vehicles

One of the main challenges in underwater robot localization is the scarcity of external positioning references. Therefore, accurate inertial localization in between external position updates is crucial for applications such as underwater environmental sampling. In this paper, we present a framework for estimating kinematic and dynamic model parameters used for inertial navigation. Accurate values of these parameters result in better trajectory estimation. Our approach can run online as well as offline, with either choice providing different advantages. Further, our framework can correct errors in the past trajectory at each estimation step. By doing so, we are able to provide improved geo-references for past as well as future spatial measurements made by the robots. This has an impact on adaptive sampling methods, which use geo-tagged measurements for building local spatial distributions and choose future sampling points. We present results from field experiments and demonstrate improvement in trajectory estimation accuracy. We also experimentally show that with optimal parameter estimates, robots can tolerate longer intervals in external positioning updates for a specified acceptable level of estimation error.