An Active Perception Framework for Autonomous Underwater Vehicle Navigation Under Sensor Constraints
Dongsik Chang · Matthew Johnson-Roberson · Jing Sun
TL;DR
An underwater vehicle with only an IMU and depth sensor moves deliberately to estimate water flow, which in turn improves its own navigation.
BRIEF
Underwater vehicles that rely on inertial sensors accumulate drift, and fixing it usually means adding costly sensors that work only in limited settings. This work uses only an IMU and a depth sensor. The vehicle estimates the surrounding water flow alongside its own state, then uses that flow estimate to correct its navigation. Because how well the flow can be estimated depends on how the vehicle moves, a controller steers motion to improve an observability or constructability measure. Simulations of a vehicle descending through a time-varying flow compare four implementations against baselines without active perception.
Abstract
Inertial navigation for autonomous underwater vehicles (AUVs) is challenging because of the drift error caused by the noise and measurement errors of inertial sensors, typically packaged as an inertial measurement unit (IMU), integrated over time. To mitigate the drift error, recent AUV state estimation approaches incorporate external references or environmental information obtained from exteroceptive sensors, with increased costs and limited operational domains. For improved navigation under sensor constraints, this article proposes an active perception framework that exploits vehicle motion to estimate the flow state together with the vehicle state using IMU and depth sensors only. The proposed framework uses the estimated flow state as external information to improve vehicle state estimation. We construct a linear time-varying system for the flow state, separated from a nonlinear system for the vehicle state. This formulation allows us to analyze uniform complete observability for the flow state, which is found to depend on vehicle motion. Then, along with vehicle and flow state estimators, we design a vehicle controller to enable vehicle motion to maximize an information metric pertaining to estimation performance based on either observability or constructability Gramian for the flow state. The proposed framework is validated through simulations for a case study with a vehicle descending through the water column in a time-varying flow field. The effectiveness of the framework is demonstrated by comparing results obtained from its four implementations with those from baseline approaches without active perception.