This groundbreaking research was recently published in Science Robotics, a leading journal in the field of robotics.
Underwater robotics is becoming increasingly significant in expanding our understanding of oceans. These robotic devices can reach depths up to 4,000 meters and offer valuable in-situ data that complement other sources, such as satellite data. Through this technology, it is possible to investigate intricate phenomena, such as the CO2 absorption of marine organisms, a critical aspect of climate change mitigation.
The recent study uses reinforcement learning, a concept prevalently used in control and robotics as well as natural language processing tools like ChatGPT. This AI methodology teaches underwater robots how to optimize their actions in real-time to achieve specific objectives. The success of these action policies even outperforms traditional methods based on analytical development in certain situations.
Ivan Masmitja, the lead author of the study, who split his time between ICM-CSIC and MBARI, commented, "This type of learning allows us to train a neural network to optimize a specific task, which would be very difficult to achieve otherwise. We have demonstrated that it's possible to optimize the trajectory of a vehicle to locate and track objects moving underwater."
This research could pave the way for a more detailed understanding of marine ecological phenomena, such as the migration or movement patterns of various marine species. Joan Navarro, a researcher at ICM-CSIC who participated in the study, highlighted the potential for real-time monitoring of other oceanographic instruments via a network of autonomous robots.
For this project, the researchers employed range acoustic techniques to estimate an object's location, considering distance measurements taken from various points. The utilization of AI, specifically reinforcement learning, proves instrumental in identifying the best data collection points, thereby enabling the robot to follow the optimal trajectory.
The Barcelona Supercomputing Center (BSC-CNS), home to Spain's most powerful supercomputer, was instrumental in training the neural networks. According to Professor Mario Martin from the UPC's Computer Science Department, the supercomputer significantly expedited the parameter adjustment process of various algorithms.
Following the training phase, the algorithms were tested on several autonomous vehicles, such as the AUV Sparus II developed by VICOROB, during experimental missions in Sant Feliu de Guixols and Monterey Bay. The testing phase was facilitated by the Bioinspiration Lab at MBARI, led by Kakani Katija.
Narcis Palomeras from UdG noted that the simulation environment incorporated the control architecture of actual vehicles, allowing for efficient implementation of the algorithms before deployment at sea.
Looking ahead, the team will explore the applicability of these algorithms in solving more complex missions, like deploying multiple vehicles to detect temperature changes in water (thermoclines), locate objects, or detect algae upwelling using multi-platform reinforcement learning techniques.
The study was made possible due to the European Marie Curie Individual Fellowship awarded to researcher Ivan Masmitja in 2020 and the BITER project, supported by the Ministry of Science and Innovation of the Government of Spain, currently in progress.
Research Report:Dynamic robotic tracking of underwater targets using reinforcement learning
Related Links
Institut de Ciencies del Mar
Water News - Science, Technology and Politics
| Subscribe Free To Our Daily Newsletters |
| Subscribe Free To Our Daily Newsletters |
