I would also like to thank all the laboratory members of the Robotics and Autonomous Systems Laboratory (RASL) for their assistance and unending kindness. I would especially like to thank Phil Davis for taking the time to teach an inexperienced machinist the ropes.
Background
Approximately two million people worldwide live with a spinal cord injury (SCI), of whom nearly 250,000 live in the United States [6]. Due, in part, to the physical barriers to basic mobility experienced by those with SPIs, the global unemployment rate of adults with spinal cord injury is over 60% [8].
Robotic Assistive Devices
Robot-aided Rehabilitation
However, none of the systems described attempt to bridge the gap between robotic rehabilitation and robotic assistance in the outside world. However, each of the designs presented so far has limitations in its ability to be adapted to real assistive robotic systems.
Human Intention Detection
Eye-Gaze Implementation
In their review of eye tracking technology, Morimoto and Mimica discuss the appeal of remote eye gaze trackers (REGTs), noting that REGTs offer ease of use and easier and faster setup [65]. Equipping an assistive robotic device with remote eye tracking has the potential to effectively enable the system to track human eye gaze and thus detect human intent.
Project Overview
Ideally, this prototype will be adapted to attach to a support above the user and assist in human movement from above, allowing the table to be fully cluttered without interfering with the use of the robotic aid. Potentiometers were used to measure the position of the end effector with sufficient accuracy, eliminating the need for expensive encoders.
Thesis Outline
The second responsibility of the wrist interface is to sense the net force between the system and the user. When the controller is in position mode, the desired position is calculated by the system from eye-tracker data.
Mechanical Design
Mechanical Design Overview
The linear portion of the system rotates around the fixed rotating portion of the system, which is located in the frame on the left of the photo. The red dot indicates the axis around which the linear subsystem rotates, and the green ring represents the resulting workspace of the device.
Design Requirements
It was also desired that the system would be able to move the end effector to its side. With a desired acceleration of 2 m/s2 and an estimated mass of 5 kg, the system should be able to apply 10 N (about 2.5 lbs) of force in all directions.
Linear Actuation
- Motor Selection
- Spring Selection
- Potentiometer Integration
The spring selected for the system also had to meet the requirements listed earlier in the chapter. One end of the spring is attached to part C and the other end to the pivoting part B.
Rotational Actuation
- Motor/Gearbox Selection
- Torque Transmission
The engine is used with a 64:1 P80 planetary gearbox, which brings maximum torque to 115 N-m (due to the limits in the gearbox) and a maximum angular velocity of just over 8 rad/s. To check the security of the connection between the bushings and the aluminum plate, both the screws and the aluminum plate should be checked.
Wrist Interface Design
- Transmitting Force
- Recording Forces
- Design and Validation
The upper ANSYS figures show the stresses due to a horizontal force acting on the center of the design towards the upper right side. Testing of the final design with real loads confirmed the simulations, with the FSRs only registering forces in the direction of the applied force.
Eye-tracker Mount
Mechanical Safety Considerations
Bill of Materials
This chapter then discusses the specifications of the electronic components of the system in more detail. The graph below shows the trajectory of the end effector during the entire system step response.
Electronics and Hardware Architecture
Overall Schematic
Brushed DC Motor for Linear Subsystem
Brushed DC Motor/Gearbox for Rotational Subsystem
The photo above and to the right is of the BaneBots P80 planetary gearbox with a 64:1 ratio that the FIRST CIM motor is mated to. Even with an efficiency of 50%, this is able to meet the requirements set out in Chapter 2.
RoboClaw Brushed DC Motor Controller
The RoboClaw can also receive control signals such as analog, serial or RC signals, making it a very versatile controller and easily controllable with an Arduino, which was the microcontroller used for the system.
Arduino Uno Microcontroller
XBee Wireless Module
Supply voltage and standard baud rate compatibility, in addition to a 45mA operating current, make the XBee S2 compatible with Arduino Uno, while a data transfer rate of 250,000 bps and a range of 90m show that the XBee S2 meets all system requirements.
Potentiometer for Linear Subsystem
Potentiometer for Rotational Subsystem
Force-Sensing Resistors (FSRs)
Batteries
Eye-Tracker
Bill of Materials
To evaluate the performance of the position-based controller, the system responses to various position inputs were recorded. Again, the system has a smaller position error for the front coordinates than the rear coordinates, both in eye-gaze accuracy and in the accuracy of the end-effector's motion. Additionally, the eye tracker must be able to track eye gaze data when the user looks beyond the width of the eye tracker.
Software Architecture
Overall Schematic
The diagram below shows the overall structure of the system software, according to the color scheme of the diagram in Chapter 3.
Xbee to Arduino Wireless Communication
This way there is no real communication between the Arduino and the stationary XBee module, but the XBee is wired to be a wireless extension to the Arduino's virtual serial port. During Arduino startup, the Arduino issues a command to set the baud rate of its virtual serial port to 57,600 bits per second, the fastest speed allowed for XBee communications. When such a request is received, the Arduino sends a 19-byte request command through its virtual port, which is broadcast wirelessly through the stationary XBee.
Arduino to MATLAB Communication
At startup, MATLAB is set to open a serial port with the Arduino at a baud rate of . The Arduino code receives these sets of messages, determines the message type based on the first byte input, and responds accordingly. The figure on the left shows the results when the Arduino is programmed to respond to a message of [1,0,0] with a single analog value.
MATLAB to C# Application Communication
One to initialize the Arduino, one to request analog data from the Arduino, and one to command the Arduino to send the desired pulses to the RoboClaw motor controller. When MATLAB sends a request, it will wait until it receives a response or until 20 milliseconds pass without a response, in which case it will resend the request. The final setting allows MATLAB to accept potentiometer data at more than 50 Hz, receive FSR data at more than 10 Hz, and send pulse commands at more than 50 Hz.
Eye-Tracker to C# Application Communication
When entering position mode, the desired position of the end effector is updated to the current position, which prevents the system from ignoring adjustments made under force control. For the entire system, the accuracy of the desired position calculation based on eye gaze data needs to be verified. The graph below shows an example of the x-y position graph of the user's gaze, the desired position displayed by the controller, and the actual position of the end effector for one trial.
Controller Design
Overall Controller Structure
The entire controller can be split into two separate sub-controllers, one controlling the system via force inputs and the other controlling the system via eye tracker and position data inputs. Only one subcontroller is active at any given time, meaning the system is either in forced or positioned mode. It will be shown that the system is still able to follow the desired velocity trajectories.
Force-Based Control
- Desired Characteristics
- Force-Feedback Structure
A force-based controller must cycle through all these steps in a smooth manner, making the system kinetically invisible to the user. When the system is at rest and the user applies a force, even though the force is averaged with previous negligible forces, the average is still non-zero. To make the system more responsive, the end effector is set to start decelerating when the input force is below a second force threshold.
Position-Based Control
- Desired Characteristics
- Position-Feedback Structure
- Eye-Tracker to Position Calibration
- Potentiometer to Position Calibration
Similar to a force-based controller, the requirements for a position sub-controller can be divided into two main components: calculating the desired position from the eye tracker data and driving the end effector to the desired position. The eye tracker was initially evaluated separately to determine the overall characteristics and accuracy of the eye gaze data it can provide. The graph on the next page shows the eye gaze locations as measured by the eye tracker for 50 consecutive fixations between two points 100 mm apart.
Conversion of Sub-controller Outputs to Arduino Pulse Outputs
- Desired Velocity to Pulse-Width Calibration
As the graph shows, the system quickly drives the end effector to a constant speed (in this case -1080 mm/s) and maintains this speed until the end effector travels the length of the linear rail. The graph below shows the angular velocity data when the rotary motor was driven with an analogReference value of 171, the lowest value used by the system. In the future, this value may be calibrated while the system is in operation, as it will likely vary with user fatigue or energy.
Control-based Safety Considerations
This allowed each subsystem to be evaluated separately and as part of the entire system. While the previous section examined the accuracy of the eye tracker, this section discusses the accuracy of the end effector when controlled by the eye tracker. The positioning of the eye tracker is the area that needs the most attention during future development.
System Evaluation
System Mechanical Responses
- Step Response
- Ramp Response
- Frequency Response
The following graphs show the responses of the system when the (x,y) components of the desired position were changed from (900,0) to (400,0) and vice versa. The graphs below show the responses of the system when the (x,y) components of the desired position were changed from (700,300) to (700,0) and vice versa. The following graphs show the responses of the system when the (x,y) components of the desired position were changed from (900,0) to (600,250) and vice versa.
System Response to Force Inputs
The gain level can be safely estimated to be 10 dB, which supports the explanation of system stability presented in Chapter 5. The graph below shows the extracted start and end points for the motion shown in the graph above, which better shows the overall accuracy of the force - based controller. One of the 16 movements triggered the emergency stop described in Chapter 5 (due to an out-of-range potentiometer reading).
Eye-Tracker Evaluation
The third point was the data collected just before the eye tracker successfully tracked the next coordinate. The overall average distance between the eye tracker output position and the actual desired position was 15.98 mm. The eye tracker was slightly less accurate for looks further away than looks closer.
System Response to Combined Force and Eye-Tracker Inputs
In an attempt to best evaluate the system, the position error of the end effector was calculated in three different ways. Method 2 measured the distance between the actual desired location and the end-effector location when a stable position was reached or an intentional force was recorded. Method 3 measured the distance between the actual desired location and the end effector position immediately before the eye tracker output a new desired position.
System Cost
Proceedings of the 9th IEEE International Conference on Rehabilitation Robotics; 2005 June 28 - July 1; Chicago, USA: IEEE Press; 2005. Kobayashi, “An Upper Extremity Mobility Assistive Robot,” in Proceedings of the 6th IEEE International Conference on Rehabilitation Robotics ICORR1999, Stanford, CA, 1999. Beer, “Development of MACARM—A New Cable Robot for Upper Extremity Neurorehabilitation,” 9th International conference on rehabilitation robotics, pp.
Contributions and Future Work
Future Work
- Mechanical Development
- Electronic/Software Development
- Experimentation with Intention-Detection
One such improvement would be to automatically adjust, or calibrate, the estimated moment of inertia of the system. Better estimation of the moment of inertia can elicit much improved responses from the rotary motor. Error handling if a complete response is not received if(a.BytesAvailable<2&&a.BytesAvailable>0) fread(a,a.BytesAvailable);.
Systematic review of the effect of robot-assisted therapy on the recovery of the hemiparetic arm after stroke. Masiero, “Nerebot: a wire-based robot for neurorehabilitation,” in Proceedings of the 8th IEEE International Conference on Rehabilitation Robotics ICORR2003, Daejeon, Republic of Korea, April 2003.
MATLAB Code
