Learning-Based Estimation of Forward Kinematics for an Orthotic Parallel Robotic Mechanism

Area of Science:

• Robotics
• Biomechanics
• Artificial Intelligence

Background:

• Cervical spondylosis requires assistive devices for patient rehabilitation.
• Parallel robots offer complex motion capabilities but face kinematic challenges.
• Analytical solutions for forward kinematics in such systems are often intractable.

Purpose of the Study:

• Introduce a 3D parallel robot for cervical spondylosis patients.
• Investigate learning-based methods (Koopman operator, neural networks) for solving forward kinematics.
• Validate the predictive accuracy of these methods through simulation and hardware experiments.

Main Methods:

• Developed a 3D parallel robot with three 5-DOF chains and a circular end-effector.
• Solved inverse kinematics analytically.
• Employed Koopman operator and neural network approaches for forward kinematics prediction.
• Generated training data from analytical inverse kinematics solutions.

Main Results:

• Learning-based methods accurately predicted end-effector position and orientation trajectories.
• Both Koopman operator and neural network approaches demonstrated high performance.
• Simulations and physical hardware experiments confirmed the effectiveness of the proposed methods.

Conclusions:

• Learning-based methods are suitable for solving forward kinematics of complex parallel mechanisms.
• This approach advances the development of intelligent assistive devices for rehabilitation.
• The study validates the potential of AI in robotic mechanism analysis.