Learning Closed Loop Kinematic Controllers for Continuum Manipulators in Unstructured Environments.

Area of Science:

• Robotics
• Machine Learning
• Control Systems

Background:

• Continuum manipulators offer unique advantages due to their inherent compliance and dexterity.
• Traditional kinematic control methods face challenges with the high dimensionality and nonlinearities of these robots.
• Integrating end-effector feedback is crucial for precise task-space control.

Purpose of the Study:

• To develop a model-free, machine learning-based approach for closed-loop kinematic control of continuum manipulators.
• To address the redundancy problem in continuum robots while ensuring scalability, speed, and stochasticity tolerance.
• To enable adaptive control in unstructured environments and under external forces.

Main Methods:

• A unique formulation for learning inverse kinematics using end-effector feedback.
• A model-free machine learning approach suitable for nonlinear, stochastic continuum robots.
• Experimental validation on a six-degree-of-freedom tendon-driven manipulator.

Main Results:

• The proposed controller demonstrates accurate and reliable pose control of the end effector.
• The system exhibits adaptive behavior in the presence of external forces and in unstructured environments.
• The approach requires minimal sensors and few tuning parameters, proving its practical viability.

Conclusions:

• The developed machine learning-based controller is well-suited for practical realization in continuum service robots.
• The model-free, adaptive approach overcomes key challenges in controlling complex robotic systems.
• This work advances the field of continuum robotics through innovative control strategies.