# Development of a Neural-Fuzzy-Based Variable Admittance Control Strategy for an Upper Limb Rehabilitation Exoskeleton

## TL;DR

This paper introduces a new control strategy for an upper limb rehabilitation exoskeleton that improves adaptability and accuracy during therapy.

## Contribution

The study proposes a neuro-fuzzy adaptive admittance control system with dual inputs for real-time damping adjustment in rehabilitation exoskeletons.

## Key findings

- The system achieves a maximum trajectory tracking error of less than 1.2° and an RMS error of ≤0.13°.
- Trajectory tracing experiments show an RMS error of 2.99 mm for a circular trajectory at Bd = 2.
- The control strategy effectively balances tracking accuracy and human-machine compliance.

## Abstract

Upper limb motor dysfunction resulting from stroke requires effective rehabilitation solutions; however, current exoskeletons are limited by single-input control, inadequate adaptation to various rehabilitation stages, and restriction to one limb. This study presents the development of a three-degree-of-freedom upper limb rehabilitation exoskeleton with three core innovations: (1) a neuro-fuzzy adaptive admittance control architecture that integrates human–robot interaction force and joint angular velocity as dual inputs for real-time damping adjustment, enabling accurate capture of dynamic movement intentions; (2) a Brunnstrom stage-specific fuzzy rule base that directly links clinical rehabilitation needs to adaptive control parameters; (3) a bilateral adaptable mechanical structure, allowing dual-upper limb training to enhance practical application. By combining radial basis function (RBF) neural network-based adaptive proportional–integral–derivative (PID) control with fuzzy variable-parameter admittance control, the system achieves a maximum trajectory tracking error of less than 1.2° and a root mean square (RMS) error of ≤0.13°. Trajectory tracing experiments confirm an RMS error of 2.99 mm for a circular trajectory at Bd = 2. The proposed strategy, validated through position tracking, admittance interaction, and trajectory tracing experiments, effectively balances tracking accuracy and human–machine compliance, providing valuable technical support for robot-assisted upper limb rehabilitation.