Learning bioinspired joint geometry from motion capture data of bat flight

Area of Science:

• Robotics
• Bioinspired Engineering
• Biomechanics

Background:

• Traditional bioinspired robots use simplified kinematic models and constraints.
• These constraints often rely on designer interpretation or fabrication limitations, not direct biological data.
• Existing methods struggle to precisely replicate complex biological joint movements.

Purpose of the Study:

• To develop a new methodology for learning joint geometry directly from biological motion data.
• To restrict robotic joint motion to the observed reachable set of a biological system.
• To create more accurate and biomimetic robotic systems.

Main Methods:

• Constructing an analytical-empirical potential energy function from biomotion observations.
• Identifying the zero-potential (ZP) configuration set by thresholding the energy function.
• Learning joint geometry from the ZP contour to restrict ball joint motion.

Main Results:

• Successfully learned joint geometry that restricts motion to the observed reachable set.
• Developed a bioinspired bat wing robotic model using the learned joint geometry.
• Motion capture experiments confirmed that the learned geometry effectively constrained movements to the ZP set.

Conclusions:

• The proposed method enables data-driven learning of joint geometry for bioinspired robotics.
• This approach offers a more accurate way to replicate biological joint constraints compared to traditional methods.
• The technique holds potential for designing more sophisticated and biomimetic robotic systems.