A Hybrid Framework for Understanding and Predicting Human Reaching Motions

Area of Science:

• Robotics
• Human-Robot Interaction
• Motor Control
• Biomechanics

Background:

• Human-robot collaboration in confined spaces necessitates robots understanding and predicting human motion.
• Model-based approaches are crucial for interpreting human motor control, which relies on biomechanical properties.
• Reaching motions are often modeled as optimization problems, but the inverse problem of identifying optimality criteria is non-unique.

Purpose of the Study:

• To implement an inverse optimal control (IOC) approach to identify cost functions governing human motion execution.
• To develop an efficient and accurate method for online prediction of human reaching motions for human-robot interaction (HRI).
• To create a framework for descriptive and generative modeling of human reaching motions.

Main Methods:

• Implemented an inverse optimal control (IOC) approach to determine the combination of cost functions for motion execution.
• Combined IOC with probabilistic movement primitives to create a hybrid model for prediction.
• Evaluated the model for online capability, accuracy, and consideration of motor variability and interpersonal differences.

Main Results:

• Human reaching motions are governed by a trade-off between kinematics and dynamics cost functions.
• The initial IOC approach lacked sufficient computational efficiency for online HRI prediction.
• The hybrid model combining IOC and probabilistic movement primitives achieved online-capable prediction with high efficiency and accuracy.

Conclusions:

• The proposed hybrid framework provides a descriptive and generative model of human reaching motions.
• This model can be effectively utilized online for human-in-the-loop robot control and task execution.
• The framework accounts for motor variability and interpersonal differences, enhancing prediction robustness.