Learning hydrodynamic signatures through proprioceptive sensing by bioinspired swimmers

Area of Science:

• Robotics
• Fluid Dynamics
• Biomimetics

Background:

• Vortex wakes generated by objects in water contain information about the object and flow conditions.
• Underwater robots often have limited sensing capabilities, hindering their ability to extract this information.
• Fish use lateral lines for multimodal sensing, inspiring artificial lateral line sensors for robots.

Purpose of the Study:

• To explore an alternative embodied sensing approach for underwater robots.
• To investigate if a robot's body kinematics in response to vortex wakes can encode wake information.
• To enable robots to 'blindly' identify hydrodynamic signatures of other bodies.

Main Methods:

• Utilizing artificial neural networks trained on angular velocity data.
• Developing fish-like robotic swimmers capable of sensing hydrodynamic forcing.
• Training the robots to label vortex wakes based on their own motion responses.

Main Results:

• Robotic swimmers successfully learned to identify vortex wakes.
• Angular velocity of the robot's body was a sufficient input for wake identification.
• The system demonstrated a capability for 'blind' identification of hydrodynamic signatures.

Conclusions:

• A robot's body motion in response to vortex wakes can be used for environmental sensing.
• Embodied sensing offers a viable alternative to traditional sensor arrays for underwater robots.
• This approach allows for the identification of moving objects through their hydrodynamic signatures.