Physical rehabilitation is an area where robotics could contribute significantly to improved motor return for individuals following a stroke. This paper presents the results of a preliminary randomized controlled trial (RCT) of a robot system used in the rehabilitation of the paretic arm following a stroke.

A new self-learning visual servoing system for the robot manipulators is proposed. This system includes two main properties: on-line self training and lifelong learning that are implemented by the Q-Learning algorithm and Explanation-based Fuzzy Neural Networks (EBFNN) respectively. We demonstrate that the number of training samples and the training time for a specific robot positioning accuracy can be reduced using explanation-based fuzzy neural networks and the Q-Learning algorithm. The system uses Q-learning to find the optimal policy in conjunction with the reinforcement learning. This policy is used by a robot to reach an object that has been randomly placed in a static workspace. Background knowledge about the robot and its environment is transferred to the robot agent during the learning process using a set of previously trained neural networks. This system learns the optimal policy in order to select the best action that maximizes the cumulative reward received at each time step. This learning approach does not use either a robot or camera model, or require calibration. Simulation results prove the effectiveness of this methodology to improve the learning process and the performance of the self-learning visual servoing system.

This article describes the design, validation, and application of a dynamic biomechanical model that assesses and monitors trajectory, position, orientation, force, and torque generated by upper-limb (UL) movement during robot-assisted therapy. The model consists of two links that represent the upper arm and forearm, with 5 degrees of freedom (DOF) for the shoulder and elbow joints. The model is a useful tool for enhancing the functionality of poststroke robot-assisted UL therapy. The individualized inertial segment parameters were based on anthropometric measurements. The model performed inverse dynamic analysis of UL movements to calculate reaction forces and moments acting about the 3-DOF shoulder and 2-DOF elbow joints. Real-time fused biofeedback of a 6-DOF force sensor and three-dimensional (3-D) pose sensors supported the model validation and application. The force sensor was mounted between the robot manipulator and the subject's wrist, while the 3-D pose sensors were fixed at specific positions on the subject's UL segments. The model input and output parameters were stored in the subject's database, which is part of the rehabilitation information system. We assigned 20 nondisabled subjects three different therapy exercises to test and validate the biomechanical model. We found that when the biomechanical model is taught an exercise, it can accurately predict a subject's actual UL joint angles and torques and confirm that the exercise is isolating the desired movement.

Spasticity is a motor disorder that causes significant disability and impairs function. There are no definitive parameters that assess spasticity and there is no universally accepted definition. Spasticity evaluation is important in determining stages of recovery. It can determine treatment effectiveness as well as how treatment should proceed. This paper presents a novel cross sectional robotic pilot study for the primary purpose of assessment. The system collects force and position data to quantify spasticity through similar motions of the Modified Ashworth Scale (MAS) assessment in the Sagittal plane. Validity of the system is determined based on its ability to measure velocity dependent resistance. Forty individuals with Acquired Brain Injury (ABI) and 45 healthy individuals participated in a robotic pilot study. A linear regression model was applied to determine the effect an ABI has on force data obtained through the robotic system in an effort to validate it. Parameters from the model were compared for both groups. Two techniques were performed in an attempt to classify between healthy and patients. Dynamic Time Warping (DTW) with k-nearest neighbour (KNN) classification is compared to a time-series algorithm using position and force data in a linear discriminant analysis (LDA). The system is capable of detecting a velocity dependent resistance (p<0.05). Differences were found between healthy individuals and those with MAS 0 who are considered to be healthy. DTW with KNN is shown to improve classification between healthy and patients by approximately 20 % compared to that of an LDA. Quantitative methods of spasticity evaluation demonstrate that differences can be observed between healthy individuals and those with MAS of 0 who are often clinically considered to be healthy. Exploiting the time-series nature of the collected data demonstrates that position and force together are an accurate predictor of patient health.