Deep Evolutionary Reinforcement Learning for Integrated Sensor Design, Dynamics, and Acoustic Target Recognition

Abstract

Small, low-cost underwater vehicles that can accomplish a mission autonomously are of great importance to future Navy operations. These vehicles should be able to operate in shallow-water environments that are difficult because underwater navigation relies on sonar and the proximity to the sea floor or other underwater objects produces large amounts of sonar echoes as well as acoustic reverberation. In addition, the environmental conditions can vary a lot around the world and hence a method that can automatically design sonar systems for different uses and areas of operation would be highly desirable. The proposed research seeks to design a miniature sonar that is inspired on the biological sonar systems found in certain species of bats that are able to fly through dense vegetation. These bats diffract their outgoing biosonar pulses with little megaphones ("noseleaves") and the returning echoes with their outer ears. Bat biosonar has a unique dynamics where the noseleaves and ears are deformed by a large number of muscles associated with each structure. To replicate the capabilities of bat biosonar in a technical design, a process will be used that mimics evolution and individual learning will be implemented. In the process, the layout of the actuators that deform the baffles will be optimized using an evolutionary algorithm and each actuator layout will be trained by a deep-learning process similar to what Google has used to beat human Go champions. The outcome of this effort will be a miniature biomimetic sonar that can determine its location in a natural environment from the analysis of echoes and hence is no longer dependent on GPS. This will enable small underwater vehicles that do not need to surface for operating a GPS receiver. Furthermore, the evolutionary AI design process can be repeated to create capable sonars that are optimized for specific missions.

Document Details

Document Type
DoD Grant Award
Publication Date
Mar 18, 2025
Source ID
N001742210007

Entities

People

  • Rolf Meuller

Organizations

  • United States Navy
  • Virginia Tech

Tags

Readers

  • Marine Mammal Biology
  • Neural Network Machine Learning.
  • Robotics and Automation.

Technology Areas

  • AI & ML
  • AI & ML - Autonomous Systems
  • Biotechnology
  • Space
  • Space - Spacecraft Maneuvers