Nonparametric Online Learning Control for Soft Continuum Robot: An Enabling Technique for Effective Endoscopic...

Area of Science:

• Robotics
• Control Systems
• Bio-inspired Engineering

Background:

• Soft robots offer safe interaction but face challenges in precise manipulation due to complex dynamics and external forces.
• Traditional model-based control methods often rely on simplified assumptions that limit accuracy in real-world scenarios, especially during endoscopic procedures.

Purpose of the Study:

• To develop a generic, model-free control framework for soft robots that learns the inverse dynamics online.
• To enable precise trajectory tracking for soft robots in dynamically constrained environments without prior knowledge of structural parameters.

Main Methods:

• A nonparametric, online, local training approach was used to learn the robot's inverse model directly.
• Finite element analysis (FEA) was employed to initialize the control policy, avoiding random exploration.
• Experimental evaluation was performed on a redundant, fluid-driven continuum soft robot prototype.

Main Results:

• The proposed framework successfully enabled the soft robot to achieve precise 3D trajectory tracking.
• The control system demonstrated high accuracy and adaptability even under dynamic external disturbances.
• The method eliminated the need for random exploration, streamlining the control policy initialization.

Conclusions:

• The developed control framework offers a robust solution for precise manipulation with soft robots.
• This approach enhances control accuracy and adaptability, paving the way for effective endoscopic navigation in complex environments.
• The model-free, online learning strategy overcomes limitations of traditional methods in handling unmodeled forces and complex geometries.