Chemical signals contained in hydrothermal fluid provide the clues to infer the location of hydrothermal vent by autonomous underwater vehicle (AUV). A partially observable Markov decision process (POMDP) is introduced to model the searching area and observation behavior in non-buoyant plume height. A POMDP-based belief state estimation method is introduced to maintain source-likelihood mapping. A real-time simulation platform is built to verify the source-likelihood mapping approach which provides a novel solution to effectively infer the vent location. The influences of memory length of flow sensor information on mapping is studied, and the length is not recommended to be very long.
Hydrothermal vent prospecting is important to broaden our understanding of enormously valuable vent ecosystems and deep-sea mineral resources. The hydrothermal vent along mid-ocean ridges emits hydrothermal fluid like black-smoker geysers. The hydrothermal fluid consists of hydrothermal plumes. Hydrothermal plumes discharged from the vent typically rise above sea floor to a height of 100-400m, and then spread laterally along ocean current at the scales of 1000km (Jakuba, 2005). Hydrothermal plumes which rise upward are called buoyant plumes, and hydrothermal plumes which move laterally are called non-buoyant plumes. In a time-averaged sense, buoyant plumes usually expand laterally from on the order of a few centimeters to a diameter on the order of about 100m (Jakuba, 2005). Non-buoyant plumes can be detected far away from the vent site. The hydrothermal fluid is characterized by different physical and chemical factors. The chemical signals contained in the hydrothermal fluid provide the clues to infer the vent location by autonomous underwater vehicle (AUV) in the height of non-buoyant plumes.
There are different approaches which can guide the AUV to further investigate the searching area when sensing chemical signals of hydrothermal fluid. An initial approach attempts to compute the concentration gradient and follow the gradient. However, gradient based approach is not feasible in environments with medium to high Reynolds numbers (Farrell, 2003; Elkinton, 1984; Jones, 1983). The second approach is bio-inspired (Shigaki, 2017). Animals are capable of using olfaction for foraging or reproductive activities, such as lobsters, ants and moths. A sequence of search speed and heading commands for the AUV are generated when the chemical signals are sensed or not such that the AUV trajectories is likely to locate the odor source (Li, 2001). Wei Li adopted moth-inspired behavior-based adaptive mission planner (AMP) to trace the chemical plume to its source and reliably declare the source location (Li, 2006). The third approach performs on-line mapping when the chemical signals are sensed or not (Pang, 2006), and the AUV reacts based on on-line mapping instead of the chemical sensor directly. On-line mapping integrates the necessary historical information, and constructs an intermediate layer between the chemical sensor and AUV planning. Farrell (2003) suggested a typical vehicle hardware, control, guidance, mapping and planning architecture. The advantage of computational capabilities and information storage capabilities are utilized by the third approach compared with the former two approaches. Hidden Markov methods (HMMs) are proposed to provide efficient algorithms for predicting the likelihood of odor detection versus position and the most likely path taken by the odor to a given location (Farrell, 2003). Pang (2006) proposed a source-likelihood mapping approach based on Bayesian inference methods, and the source-likelihood map is propagated in response to both detection and non-detection events. Source-likelihood mapping is to maintain the online map which represents the vent probability distribution. The AUV can further investigate the searching area by using the source-likelihood map.
