Data-driven geometric system identification for shape-underactuated dissipative systems

Area of Science:

• Robotics
• Geometric Mechanics
• System Dynamics

Background:

• Modeling system dynamics is challenging when component properties are unmeasurable.
• Dissipative systems offer unique opportunities for easier model identification and motion optimization.
• Existing geometric mechanics tools are extended to shape-underactuated dissipative systems (SUDS).

Purpose of the Study:

• To extend geometric system identification tools to shape-underactuated dissipative systems (SUDS).
• To demonstrate the applicability of SUDS models to various animal and soft robot motions.
• To develop a method for efficient system identification and motion optimization in dissipative systems.

Main Methods:

• Geometric mechanics for dynamics reduction.
• System identification tailored for shape-underactuated dissipative systems (SUDS).
• Conversion of shape velocity actuation inputs to torque inputs for modeling.

Main Results:

• SUDS models effectively predict motion for simulated viscous swimming platforms.
• Shape velocity actuation can be directly converted to torque inputs for modeling.
• Model complexity scales linearly with passive shape coordinates, reducing identification trials and overfitting.

Conclusions:

• The developed SUDS modeling approach simplifies identification and optimization for dissipative systems.
• This method is applicable to diverse systems, including micro-swimmers, granular locomotors, and soft robots.
• The sample efficiency of the method is valuable for robotics, control, optimization, and studying organismal motion.