Bipedal Walking of Underwater Soft Robot Based on Data-Driven Model Inspired by Octopus

Area of Science:

• Robotics
• Biomimetics
• Artificial Intelligence

Background:

• Nature, particularly soft organisms like octopuses, inspires novel robotic designs.
• Developing robots with flexible movement in three-dimensional space is a key challenge in soft robotics.

Purpose of the Study:

• To design and fabricate a cable-driven soft arm inspired by octopus tentacles.
• To develop a data-driven model for posture control of the soft arm using deep reinforcement learning.
• To assemble two soft arms into an octopus-inspired biped walking robot.

Main Methods:

• Fabrication of a cable-driven soft arm mimicking octopus tentacle characteristics.
• Establishment and training of a data-driven model using the TensorFlow framework.
• Application of deep reinforcement learning for precise posture control.
• Assembly of two soft arms into a biped walking robot for locomotion experiments.

Main Results:

• The soft arm demonstrated flexible motion in three-dimensional space.
• The data-driven model successfully controlled the posture of the soft arm.
• The biped walking robot achieved an average speed of 7.78 cm/s and a maximum instantaneous speed of 12.8 cm/s.
• The robot exhibited forward and turning locomotion capabilities.

Conclusions:

• The developed soft arm and biped walking robot effectively replicate octopus-like movement.
• Deep reinforcement learning is a viable strategy for controlling complex soft robotic systems.
• The research showcases a promising approach for creating agile, biomimetic robots for versatile applications.