Probability density based gradient projection method for inverse kinematics of a robotic human body model

Area of Science:

• Robotics
• Biomechanics
• Human-Robot Interaction

Background:

• Accurate inverse kinematics prediction is crucial for realistic human-robot interaction.
• Existing methods often struggle with the high degrees of freedom in human body models.

Purpose of the Study:

• To develop and evaluate a probability density-based gradient projection (GP) method for predicting inverse kinematics of a 25 degree of freedom robotic human body model (RHBM).
• To assess the impact of workspace discretization and subject data inclusion on the accuracy and robustness of the GP method.

Main Methods:

• Utilized a gradient projection (GP) method on the null space of the Jacobian for a 25-DOF RHBM.
• Created probability density functions based on workspace increments (1-20 divisions) and motion capture data from 10 subjects.
• Evaluated performance using root mean squared (RMS) error between predicted and recorded joint angles.

Main Results:

• Increasing workspace increments reduced RMS error, improving accuracy from 7.7° to 3.7° for included subjects.
• Higher workspace divisions led to decreased robustness of the GP method.
• The method showed varying performance based on the inclusion or exclusion of subject data in density function creation.

Conclusions:

• The probability density-based GP method offers a viable approach for RHBM inverse kinematics prediction.
• A trade-off exists between accuracy and robustness concerning workspace discretization.
• Further research is needed to optimize the balance between accuracy and robustness for real-world applications.