The researchers are motivated by the potential for UUVs to tackle dangerous tasks such as removing biofilms from ship hulls, which can introduce invasive species to the environment and increase shipping costs. The team plans to test the new training algorithm on physical UUVs in the ocean in the future.
Key takeaways:
- Researchers are using machine learning techniques to improve the navigation of uncrewed underwater vehicles (UUVs), which currently struggle with poor communication and navigation control due to water's distorting effect.
- The study, published in IEEE Access, used deep reinforcement learning to teach UUVs to navigate more accurately under difficult conditions, including strong ocean currents.
- The researchers altered the UUV's training to sample from its memory buffer in a way more akin to how human brains learn, giving more weight to actions that resulted in large positive gains and those that happened more recently.
- Using this adapted-memory-buffer technique, UUV models could train more quickly and consume less power, offering significant advantages when a UUV is deployed. The team plans to test the new training algorithm on physical UUVs in the ocean.