Sensor Reduction, Estimation, and Control of an Upper-Limb Exoskeleton

A multi-degree-of-freedom (multi-DOF) exoskeleton relies on an array of sensors to communicate its state (e.g., positions/orientations) and operator-exoskeleton contact interactions (e.g., forces/torques) to its control system. Although sensor redundancy is common in biological systems to cope with uncertainty and partial failure of sensors, in man-made systems, sensor redundancy increases the overall system's cost and control complexity. This study presents a sensor reduction technique for force/torque (F/T) sensors utilizing a Kalman filter-based sensor fusion system in the context of admittance control. The methodology is applied to the EXO-UL8 exoskeleton, which is a powered, redundant, dual-arm, upper-limb robotic system with (7 arm + 1 hand) DOFs incorporating three 6-axis F/T sensors in each arm. Motivated by improving wearability through minimizing human-exoskeleton contact interfaces, which reduces spurious contact forces due to joint misalignment; and reducing cost, the proposed strategy emulates the admittance controller's virtual dynamics with only a subset of sensors, resulting in the physical human-robot interaction feeling the same from the operator's perspective. Experimental results indicate that human-exoskeleton power exchange and actuation stresses of the operator's joints, with the proposed strategy on a subset of two sensors, are comparable to those in the full three-sensor case (p $< $ 0.01). The experiments verify the proposed methodology for the EXO-UL8, and support the feasibility of operating other Kalman filter-based sensor fusion systems with fewer sensors without sacrificing transparency in physical human-robot interaction.