Sensor Reduction of Variable Stiffness Actuated Robots Using Moving Horizon Estimation

Abstract

Variable stiffness actuated (VSA) robots are expected to play an important role in physical human–robot interaction, thanks to their inherent safety features. These systems can control the position and stiffness concurrently by incorporating two or more actuators for each joint. Unfortunately, the need for extra sensors to measure the state of these actuators decreases the reliability of these systems. In this paper, we present a sensor reduction scheme for VSA robots. Specifically, we utilize moving horizon estimation (MHE) to estimate the unmeasured states of the system. Due to its ability to handle constraints, MHE is chosen as the estimation algorithm. The estimated states are then used by a nonlinear model predictive controller to implement a closed-loop control system. In order to show the efficacy of our framework, we conducted extensive simulation and real-world experiments with a reaction wheel augmented VSA system. The objective of these experiments was to compare the control performance of the sensor reduced system (from four encoders to two encoders) with the system using the full set of states for control. The results of these experiments show the feasibility of the MHE-based sensor reduction. Sensor reduction might increase the reliability of VSA robots and might facilitate their earlier introduction to the industrial environments.