Skip to main content

Locating chemical hot-spots in the water using underwater robots

Final Activity Report Summary - AQUA-FINDER (Locating chemical hot-spots in the water using underwater robots)

Exploring and monitoring the Earth's waters is important for many scientific, economic, conservation, security and health related human activities; example applications include identifying aquatic environmental pollutants and locating their sources. Autonomous underwater vehicles (AUVs) have the potential to make continuous monitoring and real-time data widely available through low-cost, large-scale, long-term robotic missions, even in deep, toxic or otherwise inaccessible areas. Currently, the greatest obstacle to realising this vision is the lack of AUV independence: the robots' path is always programmed in advance, meaning that they do not intelligently modify their course according to their online analysis of the environmental conditions in the water. Adaptive sampling refers to onboard artificial intelligence that autonomously and continuously adjusts the vehicle's path in response to measurements of the sampled environment. Adaptive sampling can greatly increase the efficiency of robotic tracking of aquatic pollutants to their source. This process is composed of two phases: finding the chemical plume, often in a very large area ('plume finding'), and then tracking it to its source ('plume tracking').

For plume finding, AQUA-FINDER's first objective created a novel, bio-inspired search strategy named Lévy-taxis, where search paths are determined online by the robot using adaptive sampling of the ambient flow and target chemicals. In computerised simulations, Lévy-taxis located odour plumes significantly faster and with a better success rate than other search strategies such as Brownian walks, Lévy walks, correlated random walks and systematic zig-zag. These results are important because the simulations contained, for the first time ever, a digitalised real-world water flow and chemical plume instead of their theoretical model approximations.

AQUA-FINDER's second objective tackled the plume tracking problem, and created a complete architecture of autonomous robotic control for chemical hot-spot detection. This omnibus chemical localisation system extends the use of Lévy-taxis into the realm of plume tracking, and also enhances it with robust obstacle avoidance and novel source confirmation capabilities. The algorithms utilise the environmental information - flow direction, chemical concentration and chemical intermittency - to autonomously and continuously tweak the parameter values of the stochastic equations controlling the motion of the vehicle. Thus, the vehicle automatically optimises its search behaviour according to the information it is receiving from its sensors.

The third objective of the AQUA-FINDER project entailed a field experiment of an adaptive sampling plume-tracing mission in an aquatic environment, the Mondego River in Portugal. These field trials were groundbreaking in four ways: first, instead of an artificial dye, an ambient plume (effluent of a sewage treatment plant) was tracked; second, an ultra-low cost vehicle was built for this experiment, the LAUV ('light AUV') which cost under EUR 30 000; third, the mission included obstacle avoidance; and fourth, neither GPS nor acoustic beacons were used, relying instead on local topographical features for navigation using onboard imaging sonar.

The successful trials proved, for the first time, that small, ultra-low-cost autonomous underwater vehicles can reliably navigate and perform complex plume-tracing missions while being completely autonomous, i.e. without the need for GPS or the deployment of acoustic beacons in the water. This 'hardware' advancement, taken together with the 'software' advancements described in the former objectives of the project, represents a significant improvement in our ability for detecting marine pollution and locating marine resources. Thus, all the parts of this interdisciplinary project create a strong synergy for improving real-world search processes by robotic agents.