Learning From Human Demonstrations for Wheel Mobile Manipulator: An Unscented Model Predictive Control Approach.

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Industry 4.0 necessitates adaptable robots for flexible and efficient production.
• Learning from Demonstration (LfD) is a key method for robots to acquire human-like skills.
• Existing control methods struggle with uncertain parameters and disturbances in mobile manipulators.

Purpose of the Study:

• To develop a framework for a wheel mobile manipulator to learn and execute tasks from human demonstrations.
• To enable robots to adapt to unstructured environments through learned skills.
• To address trajectory tracking control challenges in mobile manipulators.

Main Methods:

• Utilized a modified dynamic movement primitives (DMPs) model for trajectory learning from human hand and body movements.
• Developed a novel model to address limitations of nonlinear feedback control for mobile manipulators.
• Implemented an unscented model predictive control (UMPC) strategy for robust trajectory tracking under system constraints.

Main Results:

• Successfully learned human movement trajectories for mobile manipulator control.
• Achieved trajectory tracking control without violating system constraints.
• Derived a condition for input to state practical stability (ISpS) and defined an error bound.

Conclusions:

• The proposed framework effectively enables mobile manipulators to learn skills from human demonstrations.
• The UMPC strategy provides robust control for mobile manipulators in complex environments.
• The developed method enhances robot adaptability and efficiency for Industry 4.0 applications.