Bioinspired design of a tissue-engineered ray with machine learning.

Area of Science:

• Biomimetic design
• Robotics
• Tissue engineering

Background:

• Biomimetic design aims to replicate biological structures for functional devices.
• Biohybrid robotic swimmers face performance issues due to scale differences between engineered devices and natural counterparts.
• Existing optimization techniques may not be sufficient for designing high-performance biohybrid systems.

Purpose of the Study:

• To introduce and validate machine learning-directed optimization (ML-DO) for biohybrid robot design.
• To demonstrate ML-DO's superiority over other nonlinear optimization methods in selecting high-performance geometries.
• To maximize thrust generation in a tissue-engineered mobuliform miniray.

Main Methods:

• Utilized machine learning-directed optimization (ML-DO) to guide the design process.
• Compared ML-DO with Bayesian optimization and other nonlinear techniques for geometry selection.
• Engineered a tissue-based robotic swimmer (miniray) for millimeter-scale locomotion.

Main Results:

• ML-DO outperformed traditional optimization techniques in identifying effective geometries.
• Achieved enhanced thrust generation in the tissue-engineered miniray.
• Developed millimeter-scale biohybrid swimmers that better adhere to natural locomotive scaling laws.

Conclusions:

• ML-DO offers a quantitatively rigorous and automated approach for designing biohybrid robots.
• This method enables the engineering of muscular structure-function relationships for improved robotic performance.
• The study advances the field of biohybrid robotics by providing a novel design optimization strategy.