Inspenet, October 14, 2023.
A group of researchers from the Universitat Politècnica de Catalunya – BarcelonaTech (UPC) has shown that deep reinforcement learning has the capacity to allow autonomous underwater vehicles and robots to accurately locate and track marine objects and beings. This advance has been documented in an article published in Science Robotics .
What does reinforcement learning offer?
This study shows that reinforcement learning, a technique widely used in fields such as control and robotics, as well as in the development of current natural language processing tools such as ChatGPT, gives underwater robots the ability to learn how to carry out specific actions at particular times to achieve specific goals. In fact, these action policies equal and even surpass conventional approaches based on analytical methods.
“ This type of learning allows us to train a neural network to optimize a specific task that would otherwise be very difficult to achieve ,” explains Ivan Masmitjà, lead author of the study.
The success of the research was based on the application of range acoustic techniques, which make it possible to estimate the position of an object by measuring distances from several different points.
To train the neural networks, the computer cluster of the Barcelona National Supercomputing Center (BSC-CNS) was used in part, which houses the most powerful supercomputer in Spain and one of the most prominent in Europe.
Once the algorithms were trained, they were tested using various autonomous vehicles in a series of experimental missions that were carried out in the port of Sant Feliu de Guíxols, in the Baix Empordà region, as well as in Monterey Bay. in California.
In subsequent research, the team will explore the feasibility of using these same algorithms to tackle more complex missions.
Subscribe to our platform and follow us on social networks: