Maneuverable gait selection for a novel fish-inspired robot using a CMA-ES-assisted workflow

Area of Science:

• Robotics
• Bio-inspired Engineering
• Machine Learning

Background:

• Fish-inspired robots offer efficient locomotion but struggle with maneuverability in confined spaces.
• Existing designs face challenges in optimizing complex, high degree-of-freedom robots for unpredictable environments.

Purpose of the Study:

• To present a new fish-inspired robot design with enhanced maneuverability.
• To develop a machine learning workflow for efficient gait training and transfer from lab to real-world environments.
• To address the limitations of current machine learning techniques in real-world robotic experiments.

Main Methods:

• Designed a fish-inspired robot with two-degree-of-freedom pectoral fins and a single-degree-of-freedom caudal fin.
• Employed a machine learning workflow for training goal-specific swimming gaits using automated lab trials.
• Utilized online learning methods to identify and transfer high-performing gaits with consistent performance across environments.

Main Results:

• The robot achieved a forward swimming speed of 0.385 m/s (0.71 body lengths per second).
• The robot demonstrated the ability to achieve a near-zero turning radius.
• The developed workflow successfully identified robust gaits transferable from simulation to the real world.

Conclusions:

• The proposed fish-inspired robot design enhances maneuverability in challenging aquatic environments.
• The machine learning workflow effectively reduces the cost of real-world experimentation for robotic gait optimization.
• This approach enables the development of adaptable and efficient bio-inspired underwater vehicles.