Hybrid Analytical and Data-Driven Modeling for Feed-Forward Robot Control

Area of Science:

• Robotics
• Control Systems
• Machine Learning

Background:

• Model-based control requires accurate plant models, but analytical models have limitations.
• Unmodeled dynamics (friction, material properties) reduce accuracy in traditional models.
• Data-driven models can be inaccurate or inefficient if they relearn known dynamics.

Purpose of the Study:

• To propose and validate a hybrid modeling approach for feed-forward control.
• To combine the strengths of analytical and data-driven models for enhanced accuracy.
• To improve control performance in robotics by addressing model uncertainties.

Main Methods:

• Developed a hybrid modeling methodology integrating analytical and learned error models.
• Applied the hybrid approach to inverse kinematics control of a soft robot.
• Implemented the hybrid approach for computed torque control of an industrial robot.

Main Results:

• Demonstrated significant improvements in modeling accuracy using hybrid models.
• Successfully controlled a redundant soft robot using inverse kinematics.
• Achieved effective computed torque control for a rigid industrial robot with unmodeled dynamics.

Conclusions:

• Hybrid models offer a superior approach to feed-forward control by leveraging both analytical insights and data-driven learning.
• The methodology is effective for complex robotic platforms where complete analytical models are unavailable.
• This approach enhances the accuracy and applicability of model-based control in robotics.