Domain-aware Control-oriented Neural Models for Autonomous Underwater Vehicles

Conventional physics-based modeling is a time-consuming bottleneck in control design for complex nonlinear systems like autonomous underwater vehicles (AUVs). In contrast, purely data-driven models require large number of observations and lack operational guarantees safety-critical systems. Data-driven leveraging available partially characterized dynamics have potential to provide reliable typical data-limited scenario high value systems, thereby avoiding months expensive expert time. this work we explore middle-ground between expert-modeled pure modeling. We present control-oriented parametric with varying levels domain-awareness that exploit known system structure prior physics knowledge create constrained deep neural dynamical models. employ universal differential equations construct blackbox graybox representations the AUV dynamics. addition, hybrid formulation explicitly residual error related imperfect compare prediction performance learned different distributions initial conditions inputs assess their suitability control.