Flinders University: AI Guides Evolution of Adaptive Control Systems
Unmanned Underwater Vehicles (UUVs) play a crucial role in conducting various missions in challenging and unpredictable conditions. In a groundbreaking study led by Flinders University and French researchers, a new bio-inspired computing artificial intelligence solution has been developed to enhance the performance of UUVs and other adaptive control systems in rough seas and dynamic environments.
The study, published in the Institute of Electrical and Electronics Engineers journal IEEE Access, introduces the Biologically-Inspired Experience Replay (BIER) method. Unlike traditional approaches, BIER addresses data inefficiency and performance degradation by utilizing valuable recent experiences that are incomplete. Dr. Thomas Chaffre, the first author of the study, explains that BIER outperforms standard Experience Replay methods, achieving optimal performance twice as fast in UUV scenarios.
The method showed exceptional adaptability and efficiency, exhibiting its capability to stabilize the UUV in varied and challenging conditions, says Chaffre.
The BIER method incorporates two memory buffers—one focusing on recent state-action pairs and the other emphasizing positive rewards. To validate its effectiveness, researchers conducted simulated scenarios using a robot operating system (ROS)-based UUV simulator and gradually increased the complexity of the scenarios. These scenarios included varying target velocity values and intensity of current disturbances.
Flinders University Associate Professor in AI and Robotics, Paulo Santos, the senior author of the study, emphasizes that the success of the BIER method holds promise for improving adaptability and performance in diverse fields that rely on dynamic, adaptive control systems.
While UUV capabilities in mapping, imaging, and sensor controls continue to improve, challenges arise when they encounter unforeseen variations in real-world environments. Existing Deep Reinforcement Learning (DRL) methods face difficulty in delivering optimal performance due to limited observability of UUV maneuvering tasks and the complex dynamics of the underwater environment.
The introduction of BIER represents a significant leap forward in enhancing the effectiveness of deep reinforcement learning methods in general. Its ability to navigate uncertain and dynamic environments efficiently signifies a promising advancement in the field of adaptive control systems.
The study titled Learning Adaptive Control of a UUV using A Bio-Inspired Experience Replay Mechanism, authored by Thomas Chaffre, Paulo E Santos, Gilles Le Chenadec, Estelle Chauveau, Karl Sammut, and Benoit Clement, has been published in IEEE Access. The work was made possible by funding from Flinders University, ENSTA Bretagne, as well as support from the Government of South Australia, Région Bretagne, and Naval Group.
With the development of the BIER method, Flinders University and the collaborating researchers have opened new doors for enhancing the efficiency and adaptability of UUVs and other adaptive control systems. As technology continues to progress, these breakthroughs will continue to shape the future of underwater exploration and defense missions.