Learning human inverse kinematics solutions for redundant robotic upper-limb rehabilitation

This paper proposes a real-time human inverse kinematics approach based on the Gaussian Process (GP) to address the open problem of finding appropriate inverse kinematic solutions for upper-limb rehabilitation using redundant robotic exoskeletons. The proposed approach is validated using recorded data from a Motion Capture (MoCap) system and is shown to achieve accurate and unique upper-limb natural solutions for different rehabilitation exercises. This approach is a computationally tractable alternative to existing solutions, which are not always feasible and unsuitable for real-time rehabilitation tests, as they are based on optimizing discomfort indices online or computationally demanding learning methods. A comparison with existing inverse kinematics solutions confirms the superior performance of the proposed approach. Additionally, a robust Model Predictive Control (MPC) with an Integral Sliding Mode (ISM) combination is proposed for safe trajectory tracking during rehabilitation exercises. Several real-time experiments were conducted on a 7-DoF exoskeleton robot with and without the presence of a healthy subject to verify the effectiveness of the proposed methods.