Learning an Autonomous Dynamic System to Encode Periodic Human Motion Skills.

Area of Science:

• Robotics
• Control Theory
• Biomechanics

Background:

• Human motion skills transfer is crucial for robotics.
• Existing methods primarily focus on goal-directed movements, neglecting periodic motions.
• Periodic Dynamic Movement Primitives (DMP) are susceptible to spatial disturbances due to phase parameters.

Purpose of the Study:

• To develop a novel approach for learning an autonomous dynamic system (ADS) for periodic motion skills transfer.
• To address the limitations of existing methods, particularly the sensitivity to spatial disturbances in periodic Dynamic Movement Primitives (DMP).
• To create an ADS with a stable limit cycle without relying on phase parameters.

Main Methods:

• A data-driven Lyapunov function (energy function) is learned to represent periodic human motion trajectories.
• The autonomous dynamic system (ADS) is learned by solving constrained optimization problems related to the energy function.
• Constraint functions are designed to ensure trajectory convergence to an energy level surface mimicking human demonstrations.

Main Results:

• The proposed approach successfully learns an ADS with a stable limit cycle for periodic motion.
• The developed system demonstrates robustness against spatial disturbances, unlike traditional periodic DMP.
• Experimental results validate the effectiveness of the novel approach for periodic human motion skills transfer.

Conclusions:

• The novel approach effectively enables periodic motion skills transfer by learning an autonomous dynamic system (ADS) with a stable limit cycle.
• Eliminating phase parameters enhances the system's stability and robustness against spatial disturbances.
• This method offers a promising direction for more versatile human motion imitation in robotics.