Sign in Subscribe
Machine learning

Newcastle University Researchers Develop AI-Powered Underwater Vehicle for Cetacean Conservation

Strategic Insights

— 1 min read

Researchers from Newcastle University have created an artificial intelligence-powered underwater vehicle designed to monitor marine mammals with minimal intrusion. The system, called SWiMM, uses a Unity simulation model and a Deep Reinforcement Learning backend to navigate and track targets underwater. This innovative approach aims to replace traditional tagging methods, which can harm the animals. The study's authors believe that their method can provide more accurate and non-intrusive data for cetacean conservation.

Key Takeaways:

  • The researchers from Newcastle University have developed an AI-powered underwater vehicle called SWiMM, which uses a Unity simulation model and a Deep Reinforcement Learning backend.
  • The vehicle is designed to monitor marine mammals with minimal intrusion, reducing the risk of infection and death associated with traditional tagging methods.
  • The study's authors propose a pre-processing step that exploits a state-of-the-art CMVAE to reduce dimensionality while minimizing data loss.
  • The SWiMM system provides custom behavior metrics that are unbiased and unprecedented in current ROV simulators, enabling successful ROV behavior while tracking targets.
  • The experiments showed that the SWiMM system can maximize the ROV behavior using image data alone, achieving near-perfect behavior.
  • This innovative approach aims to replace traditional tagging methods, which can harm marine mammals.
  • The study's authors believe that their method can provide more accurate and non-intrusive data for cetacean conservation.

Statistics:

  • The researchers used a Deep Reinforcement Learning backend to navigate and track targets underwater.
  • The study's authors exploited a state-of-the-art CMVAE to reduce dimensionality while minimizing data loss in the pre-processing step.
  • The SWiMM system provides custom behavior metrics that are unbiased and unprecedented in current ROV simulators.
  • The experiments showed that the SWiMM system achieved near-perfect behavior using image data alone.
  • The study's authors conducted experiments that validated the system's ability to learn and adapt to new environments.

Sources:

  • NewsRx. Newcastle University Researchers Focus on Artificial Intelligence (From Camera Image to Active Target Tracking: Modelling, Encoding and Metrical Analysis for Unmanned Underwater Vehicles). Robotics & Machine Learning. May 12, 2025; p 532.
  • AI, 2025,6(4):71. From Camera Image to Active Target Tracking: Modelling, Encoding and Metrical Analysis for Unmanned Underwater Vehicles. MDPI AG. https://doi-org.sdpl.idm.oclc.org/10.3390/ai6040071.

Read more