Localization and mapping in dynamic underwater environments
Autonomous localization and environment mapping are considered key capabilities to achieve truly autonomous robotic systems. Localization is the process of estimating the robot’s global position and orientation, while mapping consists on building an accurate internal representation of the environment. Both tasks are associated with perception through onboard carried sensors. So, information about the environment usually comes from noisy measurements taken in a noisy, dynamic and unstructured environment. Sensor capabilities are also environmental dependent, take the example of underwater SONAR, that evidently it is not suitable for in land operation. Another example is the GNSS, that provides handy global references in open skies, but suffers from deteriorated performance inside canyons, becoming unreachable indoors. For those facts, no general solution, for localization or mapping, is known to work universally in every scenario.
Usually, a single sensor does not provide enough information to properly estimate localization or mapping. Hence, long-term reliable operation, with high sampling rate, results from the combination of complementary information given by different sensors. Moreover, robot perception is affected by uncertainty, therefore, instead of computing a single solution, probability distributions are used instead to represent the intrinsic uncertainty of robot’s localization and mapping states.
Underwater environments encompass extra challenges that derive from the particular properties of the aquatic medium. Due to the reduced visibility underwater, the effectiveness of optical sensing devices is limited no near range operations. Moreover, in the middle of the water column, no consistent visual clues are expected, therefore visual based techniques are useless.