For the 400,000 people affected by stroke, rehabilitation is an important part of recovery. The goal is to improve function so individuals become as independent as possible.
Companies focusing on medical technology are looking at mechanical devices that can do the work of therapists specializing in limb rehabilitation for stroke patients, thereby freeing therapists to serve more patients. One of these firms is Canadian-based Quanser Inc., and its work in developing haptic robotic technology to create interactive, intelligent devices and software. Working in partnership with the University of Toronto and the Toronto Rehabilitation Institute, Quanser is developing a commercial rehabilitation product for stroke victims - the Autonomous Upper-Limb Stroke Rehabilitation Device, or rehab robot.
Replicating Hands-On Therapy
Quanser's device accurately replicates traditional rehab exercises by allowing patients to experience realistic forces and pressures traditionally applied through manual manipulative push-pull therapy that typically occurs in clinics. The rehab robot allows patients to conduct these exercises at home by pushing on the robotic arm, feeling resistance and watching results on a video monitor.
The primary challenge in developing the rehab robot was to provide realistic sensory feedback for the patient, while also enabling the robot to sense how to accurately respond to each patient.
The solution was found in seamlessly integrating hardware with an artificial intelligence (A.I.) component. As described by Dr. Alex Mihailidis at the University of Toronto, the system is guided using input from the various encoders located directly on the robot. Each degree of freedom uses a high-quality MicroMo motor that has great haptic specifications. The encoders are mounted on the motor which allows for high-resolution joint position sensing. These positions are then used to monitor the location of the patient's hand and react accordingly within the virtual exercise.
Input from these encoders is used by the A.I. controller called a partially observable Markov decision process (POMDP), essentially the decision-maker of the system. It is a typical "sense, think, act" sequence:
Sense: The position of the patient's hand is measured by utilizing the high-resolution encoders on each of the two axes. The encoder information is then used to calculate the kinematic position of the robot.
Think: The current state of the robot and the previous state of the training allow the artificial intelligence to quickly calculate its next move.
Act: Once corrective or assistive forces are calculated by the A.I., the information gets translated into the required current for each servomotor to effectively display the force back to the patient.
Data from the robotic device is passed to a "state estimator" that shows the state of the patient, such as level of fatigue or ability to reach a target. This determines the progress of the user as a belief state, which is a representation of what the A.I. controller thinks is going on with the patient and the completion of the exercise. A policy then maps the belief state to an action for the system to execute. This can be either setting a new target position and resistance level or stopping the exercise.
Typical Rehab Robot Sequence
The system makes a decision on the starting point for the patient's rehab session, including the starting level of resistive force to apply to the patient and the target distance the patient needs to hit.
The system provides a prompt to the patient with respect to the target that needs to be reached.
The patient performs the necessary movement using the robotic platform.
The system collects data from the motion just performed and compares it with the target values.
The system decides the next resistive force and target to be reached.
This process is repeated until the targets are met, or the system decides if the patient has become fatigued and then stops the exercise.
Quanser's team created the physical (robotic) and virtual (rehab simulation) interface between the user and the computer. The robotics stage provides realistic force feedback technology achieved through advanced mechanical design and careful selection of the motors, encoders and other components that are integral to the overall system. For example, motors and transmission ratios were calculated to achieve the required force specs. As with all haptic interface design, special care was taken to ensure the mechanism has low friction and minimized inertia, and that it is easy to use.
The rehab robot's real-time control system algorithm was developed and tested using Quanser's real-time control software and incorporated using the company's Q4 high-speed I/O data acquisition board that responds to the patient and changes the experiment 1,000 times a second.
The third component is the Java interface running a variety of virtual training sessions and communicating with the real-time robotic process via shared memory space. Patients are able to view their progress on a video monitor. An added feature is a software-driven gaming aspect, also displayed, which patients control through motions required during reaching exercises. One example is catching a bunny. Another is a labyrinth where patients move a ball through a maze by manipulating the robotic arm in the required directions.
According to Paul Gilbert, Quanser's CEO, the company is working with rehab institutes to ensure efficacy and affordability of a production-grade system that is portable and easy to use. Commercialization of the rehab robot is expected in 2011.