Positioning of Autonomous Underwater Vehicles using Machine Learning Techniques

Autonomous underwater vehicles have many applications in surveillance, search and rescue, and cargo delivery. However, these vehicles lack an effective positioning mechanism such as GPS in outdoor environments. Although several positioning alternatives have been proposed previously, the use of acoustic signals in conjunction with trilateration techniques is one of the simplest and cheapest options, but there are still problems related to multi-path propagation, signal fading, and reverberant environments. On the other hand, machine learning is revolutionizing the field of data-driven models for making predictions and/or pattern recognition. In fact, machine learning regression is an efficient technique for establishing the relationship between a set of independent features and a dependent variable that takes continuous values. Therefore, this work proposes the use of machine learning as an alternative for the positioning of underwater drones depending on the sound received by an array of hydrophones. An experimental testbed was set up to obtain the training and validation datasets that include the sound generated by an underwater drone while moving in a pool with 4 hydrophones. The results show that the precision of this proposal is better than those previously obtained in other works, reaching an average accuracy of 27 cm in a 10 × 5 meter pool.