Omnidirectional Assistive Wheelchair: Design and ... - ScienceDirect

Available online at www.sciencedirect.com

ScienceDirect Procedia Computer Science 105 (2017) 68 – 74

2016 IEEE International Symposium on Robotics and Intelligent Sensors, IRIS 2016, 17-20 December 2016, Tokyo, Japan

Omnidirectional Assistive Wheelchair: Design and Control with Isometric Myoelectric Based Intention Classification Ananda Sankar Kundua,∗, Oishee Mazumdera , Prasanna K. Lenkab , Subhasis Bhaumikc a School

of Mechatronics and Robotics, IIEST, Shibpur, Howrah-711103, India Institute of Orthopaedically Handicapped, Bonhooghly, Kolkata-700091, India c Aerospace Engineering and Applied Mechanics, IIEST, Shibpur, Howrah-711103, India

b National

Abstract Smart electric wheelchairs are becoming a natural substitute of the conventional wheelchairs as an assitive device for geriatric population and patients suffering from mobility disorders. There is a demand for developing powered wheelchairs with intelligent control to suit wide range of application in the field of assistive technology. This paper deals with the development of a 4 wheeled omnidirectional wheelchair and its control using a myoelectric user intention interface. Developed system is driven by holonomic drive system, exploring greater maneuverability compared to conventional powered wheelchairs. Myoelectric signals from forearm muscles are processed to extract some features for seven different wheelchair motion namely forward, backward, left, right, clockwise and anticlockwise turn and stop. A neural network classifier classifies the user intention and maps the intention to wheelchair motion. The developed system finds its direct application in transporting people with locomotor disability, geriatric population as well as an indoor navigation vehicle. c 2017 Elsevier B.V. 2016The TheAuthors. Authors.Published Published © byby Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of organizing committee of the 2016 IEEE International Symposium on Robotics and Intelligent Peer-review under responsibility of organizing committee of the 2016 IEEE International Symposium on Robotics and Intelligent Sensors (IRIS 2016). Sensors(IRIS 2016).

Keywords: Omnidirectional whelchair; Myoelectric; Neural Network; Holonomic Control;

1. Introduction Powered wheelchairs have been developed over the years for locomotion disabilities and for geriatric assistance. Smart electric wheelchairs are special class of powered wheelchairs, which are becoming a natural substitute of the conventional wheelchairs as an assitive device. Moreover, due to the ease of control, application specific human machine interface and smooth mobility, electric wheelchairs are becoming a popular indoor navigation vehicle. One of the first prototypes of smart wheelchair was proposed by Madarasz et.al 1 in 1986 which presented a wheelchair designed to transport a person to a desired room within an office building given only the destination room number. Since then, many such smart wheelchairs have been developed and few have been commercialized 2 , 3 . Most of the developed smart wheelchairs are modification over existing commercially available powered wheelchairs with add on facility to enhance maneuverability, navigational intelligence and multi-modal control interfaces. To name a few, ∗

Corresponding author. Tel.: +091-9830317143; Email: ananda [email protected]

1877-0509 © 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of organizing committee of the 2016 IEEE International Symposium on Robotics and Intelligent Sensors(IRIS 2016). doi:10.1016/j.procs.2017.01.200

Ananda Sankar Kundu et al. / Procedia Computer Science 105 (2017) 68 – 74

NavChair 4 , Office wheelchair with high Maneuverability and Navigational Intelligence (OMNI) 5 , Mobility Aid for elderly and disabled people (MAid) 6 , Smart Power Assistance Module (SPAM) 7 , TinMan 8 , etc. provides controlled indoor navigation. Among the wheelchairs developed with omnidrive or omnidirectional mobility, the OMNI (Office Wheelchair for High Manoeuvrability and Navigational Intelligence for People with Severe Handicap) is a mecanum wheeled wheelchair developed for individuals with severe mental and physical disabilities. Another example of an omnidirectional wheelchair is iRW 9 , which provides a telehealth system with easy-to-wear, non-invasive devices for real time vital sign monitoring and long-term health care management for the senior users, their family and caregivers. Research on smart wheelchair control now a days mainly focuses on different types of user interfaces to control the wheelchair. Conventional control of smart electric wheelchairs are mainly based on joystick 10 , but joystick control possess limitation for elderly and disabled people, lacking full dextrous control of their upper limb. The keyboard and mouse are often used as the HCI devices. However it needs much training for the disabled and the elderly who are not familiar with computer. Many advanced alternative controllers and human-machine interfaces (HMIs) have been proposed, which includes vision based techniques, voice recognition, hand gestures, head or chin control, biosignal based control like EMG, EOG or EEG control 11 - 15 , etc. Inspite of different control modes being researched and developed, each alternative controller has its own disadvantages due to application specific restrictions like voice interface is affected by noise, vision based techniques are computation costly and slow. There are still no clear standard control mode which can be universally accepted and easy to implement like joystick control. Among bio-signals, EMG signals are considered better for control purpose and are used to control a variety of assistive devices, e.g. robot arms 16 , hand prostheses 17 , and electric wheelchairs 18 , 19 . EMG-based powered wheelchair control is a much researched topic 20 - 24 and has been shown to be effective as an alternative control mode, specially for rehabilitation aid. In this paper, we present development of an omnidirectional wheelchair and formulate an intention based control to operate the wheelchair. All the wheelchairs or indoor transporters with holonomic drive are developed with mecunum wheels or are a three wheeled omni platform. Mecunum wheels are inherently suitable for handling high load but its turn rate is slow compared to omni wheels. 4 wheel platform with omni wheels are difficult to design, mainly because of its unequal ground reaction force. If designed properly, 4 wheeled omni platform provides better performance than platform developed with mecunum wheels. We propose a unique wheelchair design with omni wheels and proper suspension mechanism to provide enhanced mobility in indoor environment. Developed system has been controlled an user intention based control using users myoelectric signal extracted from four forearm muscles. User’s intention to perform certain hand gesture like forward, backward, etc. are mapped with motion commands of the wheelchair. EMG signals are processed to extract some features for seven different wheelchair motion namely forward, backward, left, right, clockwise and anticlockwise turn and stop. A neural network classifier classifies the user intention and maps the intention to wheelchair motion. Developed wheelchair has been tested by five users to validate the performance of the developed system. Effectiveness and accuracy of the control interfaces have been compared with standard joystick based control paradigm. 2. Methodology 2.1. Omnidirectional wheelchair Platform Development Omni directional wheelchair posses special maneuverability due to the omni wheels which allows translational as well as lateral mobility. Unlike differential or steering drive, omni drive systems does not possess holonomic constraints, allowing motion in both the body axis possible. Moreover, translational movement along any desired path can be combined with a rotation, so that the robot arrives to its destination at the correct angle 25,26,27 . In order to achieve this, the wheel is built using smaller wheels attached along the periphery of the main wheel. Each wheel provides traction in the direction normal to the motor axis and parallel to the floor. The forces add up and provide a translational and a rotational motion for the robot. Design of a 4 wheel driven Omni Wheel based platform needs special attention. Regardless the surface type, all four wheels should receive equal ground reaction force (GRF) or else there are chances of wheel slippage. A Omni wheelchair is designed to support 120 Kg including the platform’s own weight and payload with proper suspension mechanism to provide equal GRF in all wheels.

69

70


(a)

(b)

Fig. 1: Omni directional Wheelchair: a) CAD model of developed wheelchair. b) Developed omni directional wheelchair.

Wheelchair design comprises designing of the motor wheel assembly, a suspension mechanism and a chasis with sufficient load bearing capacity. Fig.1 shows different parts of the developed omni wheelchair. Along with the front motor-wheel assembly mounted with suspension mechanism, the rear and the side motor-wheel assemblies are directly fixed with the chassis. The motors are positioned in the motor-wheel assembly horizontally with offset. In the front side of the chassis a pair of footrest is connected. The footrest position can be adjusted vertically with the help of slots, and a pair of load cell is attached with the footrest for using them as brake pedal optionally. Electronic components of the wheelchair consists of four Buhler DC motors (77W) and their drivers (40V, 20A) from ‘Rhino Motion Controls’ to drive the motors. These motor drivers accept PWM/DIR input from the controller. The output velocity is proportional to the input PWM duty cycle. The driver measures the speed of the motor from current harmonics and regulates it by close loop PD control. ACS712-20A current sensor modules are connected with individual motors to monitor the current consumption. These sensors analog voltage outputs are connected to the analog input pins of STM32 microcontroller. Finally the STM32 microcontroller communicates with the master controller (NUC) via serial port. 2.2. Myoelectric Control There has been substantial amount of work related to EMG based powered wheelchair control but there is still no standard technique that is regarded better over the other. We have used a pattern recognition based classifier to map intention associated with forearm muscle activity to motion command of the wheelchair. EMG signals were acquired from four forearm muscles (Fig.2a) of the user namely: Flexor Carpi Radialis (Channel 1), Flexor Carpi Ulnaris (Channel 2) , Extensor Carpi Radiallis Longus (Channel 3), Brachioradialllis (Channel 4) during different isotonic contraction for seven different wheelchair motion states. Acquired EMG signals are processed to calculate the RMS envelope for each channel. Six ratiometric features are calculated from four channel RMS EMG to construct the feature vector based on which a neural network classifier classifies the motion command with the user’s intention of the desired motion. 2.2.1. Myoelectric unit Instrumentation Myoelectric unit includes bio-potential electrode (Ag/AgCl), pre-amplifier circuit, ATMEGA-8 microcontroller based acquisition electronics and a 9 Volt alkaline battery as shown in Fig.2b. EMG signals are acquired simultaneously from four forearm muscles and amplified by pre-amplifier circuit. Pre-amplifier circuit is designed with AD-623 instrumentation amplifier, which is a specially designed for low power biomedical application 28,29 . Pre amplifier circuit design is divided into two stages. First stage provides high input impedance and gain factor of 11. Second stage with a gain factor of 455 is cascaded with first stage via AC coupling. Amplified EMG signals are acquired by 10 bit ADC of Atmega-8 microcontroller with 1000Hz sampling frequency. Acquired sensor data are send to computing

71


(a)

(b)

Fig. 2: a) Selected muscles for myoelectric control of wheelchair b) Myoelectric unit

(a)

(b)

(c)

Fig. 3: a) Raw EMG extracted from selected muscles. b) RMS envelope calculated on 50ms window length. c) RMS envelope calculated on 200ms window length.

unit using serial to USB converter FT-232. Raw EMG signal extracted from the four selected muscles are shown in Fig.3a. From the raw EMG, signal envelope is calculated using moving window Root Mean Square (RMS) detection technique, taking 100ms overlapping window. Generation of a smooth envelope as well as real time processing of the EMG signal is vital, so consideration of the number of overlapping points is critical. Fig.3b and 3c shows the RMS envelope nature of the four acquired channels on varying the window length from 50 ms to 200 ms. A smooth filtered envelope requires higher window length and is desirable for proper feature extraction but a longer window also makes the system sluggish and computationally heavy. Optimum window length is chosen to be 100 ms for proper feature set identification. 2.2.2. Motion intention classification For motion intention classification to operate the wheelchair, EMG signals were extracted from five healthy male subjects. Subjects were asked to perform the desired intention for seven motion states of forward, backward, left, right, clockwise turn, anticlock turn and stop. Fig.4a shows the RMS EMG value of the four muscle channels for each of the desired motion commands. From the figure, it is evident that all the channels contribute during the seven different motion commands but their range of activation differs according to the motion. Feature vector for classification is constructed using different combination of ratio-metric RMS value from different channels. From the four channels, six unique features are calculated, which are Channel 1/channel 2; channel 1/channel 3; channel 1/channel 4; channel 2/channel 3; channel 2/chnnel 4; and channel 3/channel 4. Ratio-metric feature removes the

72


Table 1: Feature values for classification: Mean feature set for subject 1

Forward Backward Left Right Clock CClock Stop

C12

C13

C14

C23

C24

C34

0.187 1.546 0.586 1.327 0.732 0.850 0.167

0.118 0.791 0.459 0.211 0.313 0.245 0.028

0.087 1.630 0.475 0.649 0.774 0.358 0.032

0.659 0.512 0.764 0.1613 0.650 0.287 0.162

0.471 1.057 0.818 0.4871 1.460 0.419 0.183

0.738 2.079 1.111 3.103 2.505 1.463 1.135

effect of fatigue, difference in activity for different users and provides a robust feature for long term use. Feature vector thus consists of 6 features which is used for classifying the motion intention using Neural network classifier. A typical set of values for each activity, averaged over all user is shown in Table 1. Neural network toolbox in MATLAB has been used for motion classification. The network has six input channels, ten hidden layers and seven output channels. For training the classifier, subjects were asked to wear the myoelectric unit and perform the seven predefined gestures. Experiment session were of 10 minute duration for each user. During this ten minutes the user performed all the seven motion commands as per their ease and wish. Around 6000 data packets of mixed data set were obtained from each user. From this data pool, the classifier has been separately trained for each user using 70% of data and validated with the remaining 30%. Fig.4b shows the classifier performance for subject 1 in terms of confusion matrix. Classifier achieved an overall classification accuracy of 98.9%.

3. Result and Discussion For validating the performance of the developed system, 5 healthy male subjects were asked to control the wheelchair using conventional joystick control and myoelectric intention based control to traverse a predefined path. During navigation, parameters like total traverse time, energy spent and path length were measured to serve as performance comparison metric. System energy was estimated as sum of square of current values of individual motors during traversing. Traversed path length and time were calculated using odometer data of the wheelchair. Table 2. tabulates the performance index during experiments with three control modes termed as ‘test set A’ for joystick control, ‘test set B’ for myoelectric control.

(a)

(b)

Fig. 4: a) Muscle activity in terms of RMS for seven set of motion commands. b) Classifier performance for subject 1.

73


For voice and EMG based control, number of commands issued varied as the user was free to chose the trajectory according to their ease of habituation to the omni directional movement. Fig.5 shows the path traced by different user during navigation. Average performance metric for all three test cases traveled length to be similar for both cases. Due to higher speed of traversal, energy spent during joystick is higher than the other two modes. High rate of comparability of myoelectric control algorithm with respect to conventional joystick control suggests that the proposed myoelectric control method can be used as an alternative approach to control wheelchairs or mobile robotic platforms. Table 2: Wheelchair control: Comparison between control modes Subject

Commands issued

Path Length (m)

Travel Time(sec)

Energy Spent (W.sec)

A

B

A

B

A

B

A

B

1 2

-

98 49

16.10 16.43

16.60 13.53

76.14 56.24

96.68 67.32

16295 25753

18323 13415

3 4 5

-

44 43 40

17.10 14.39 15.18

15.41 14.525 14.30

84.02 66.76 61.34

76.42 71.01 72.14

19351 21455 19855

15256 15311 13949

15.84

14.58

68.9

76.78

21191.2

15250.8

Average

Fig. 5: Path traced by different user and wheelchair navigation snap shots

4. Conclusions This paper presents the development of a 4 wheeled omnidirectional wheelchair and its control using intention based myoelectric control. 4 wheeled wheelchair design is innovative, incorporating hydraulic suspension system for equal load distribution in the wheels. A classification algorithm has been designed for myoelectric control mode, where, user’s intention of motion are mapped to seven different motion command of the wheelchair. The neural network based classifier achieved an overall classification accuracy of 98.9% which is highly satisfactory for wheelchair control. Performance of the control logic is compared with joystick based control in terms of path length, traversal time and energy for a predefined path with fixed start and end points. Results indicate a high similarity index between results obtained from joystick and myoelectric control modes. Initial test results reported in this paper are based on 5 healthy users traversing a predefined path with static obstacle in controlled environment. In near future, the system will be

74


tested for navigating in constrained environment with dynamic obstacles and its application as a rehabilitation aid or assistive system will be investigated. References 1. R.L.Madarasz, L.C.Heiny, R.F.Cromp and N.M.Mazur,1986, “The design of an autonomous vehicle for the disabled”, IEEE Journal on Robotics and Automation, vol.2(3), pp: 117-126. 2. R.C.Simpson,2005, “Smart wheelchair: A literature review ”, Journal of Rehabilitation Research & Development, vol.42(4), pp: 423-436. 3. R.H.Krishnan and S. Pugazhenthi,2014, “Mobility assistive devices and self-transfer robotic systems for elderly, a review ”, Intelligent service Robotics, vol.7, pp: 37-49. 4. S.P.Levine, D.A.Bell, L.A.Jaros, R.C.Simpson, Y.Koren, J.Borenstein, “The NavChair assistive wheelchair navigation system ”, IEEE Transaction on Rehabilitation Engineering, vol.7(4),pp: 443-51. 5. U.Borgolte, H.Hoyer, C.Buehler, H.Heck, R.Hoelper,1998, “Architectural concepts of a semi-autonomous wheelchair ”, Journal of Intelligent Robotic System, vol.22(3) pp: 233-53. 6. E.Prassler, J.Scholz, P.Fiorini,2001, “A robotic wheelchair for crowded public environments ”, IEEE Robotic and Autonomous Magazine, vol.8(1), pp: 38-45. 7. RC.Simpson, E.F.LoPresti, S.Hayashi, S.Guo, D.Ding, R.A. Cooper, 2003, “Smart Power Assistance Module for manual wheelchairs. Technology and Disability: Research, Design, Practice and Policy ”, 26th International Annual Conference on Assistive Technology for People with Disabilities (RESNA), Jun 1923, 2003; Atlanta, GA. Arlington (VA): RESNA Press. 8. D.P.Miller, M.G.Slack,1995, “Design and testing of a low-cost robotic wheelchair prototype ”, Autonomous Robots, vol.2(1), pp: 77-88. 9. Po Er Hsu, Yeh Liang Hsu1, Kai Wei Chang and Claudius Geiser, 2012, “Mobility Assistance Design of the Intelligent Robotic Wheelchair”, International Journal of Advanced Robotic Systems, Vol. 9, pp:244-53. 10. L. Fehr, W.E. Langbein, and S.B. Skaar, 2000, “Adequacy of power wheelchair control interfaces for persons with severe disabilities: A clinical survey ”, Journal of Rehabilitation Research and Development, vol.37(3), pp. 353-60. 11. P. Jia, H. Hu, T. Lu, and K. Yuan,2007, “Head gesture recognition for hand-free control of an intelligent wheelchair”, Internation Journal of Industrial Robotics, vol.34(1), pp. 60-8. 12. S. Nakanishi, Y. Kuno, N. Shimada, and Y. Shirai, 1999, “Robotic wheelchair based on observations of both user and environment”, IEEE/RSJ International Conference on Intelligent Robots and Systems, Kyongju, Korea, pp. 912-7. 13. R. Barea, L. Boquete, M. Mazo, and E. Lopes, 2002, “System for assisted mobility using eye movements based on electrooculography”, IEEE Transaction on Neural System and Rehabilitation Engineering, vol.10(4), pp. 209-18. 14. K. Tanaka, K. Matsunaga, and H.O. Wang,2005, “Electroencephalogram based control of an electric wheelchair ”, IEEE Transaction on Robotics, vol. 21(4), pp. 762-6. 15. M.A. Oskoei, and H. Hu, 2007, “Myoelectric control systems - A survey, ”, Biomedical Signal Processing and Control, vol. 2(4), pp. 275-94. 16. O. Fukuda, T. Tsuji, M. Kaneko, and A. Otsuka,2003, “A human-assisting manipulator teleoperated by EMG signals and arm motions”, IEEE Trans. Rob. Autom., vol. 19(2), pp. 210-22, Apr. 2003. 17. K. Englehart, B. Hudgins, and P.A. Parker, 2001, “A wavelet-based continuous classification scheme for multifunction myoelectric control”, IEEE Trans. Biomed. Eng., vol. 48(3), pp. 302-11, Mar 2001. 18. Y. Oonishi, S. Oh, and Y. Hori, 2010, “A new control method for power assisted wheelchair based on the surface myoelectric signal, IEEE Trans. Ind. Electron., vol. 57(9), pp. 3191-6. 19. I. Moon, M. Lee, J. Ryu, and M. Mun, 2003 “Intelligent Robotic Wheelchair with EMG, Gesture, and Voice-based Interfaces ”, Proc.of IEEE/RSJ IROS 2003. 20. M. R. Ahsan, M. I. Ibrahimy, and O. O. Khalifa, 2010, “Advances in electromyogram signal classification to improve the quality of life for the disabled and aged people ”, Journal of Computer Science, vol. 7(6), pp. 706715. 21. M. Khezri and M. Jahed, 2007, “A novel approach to recognize hand movements via semg patterns ”, 29th Annual International Conference of the IEEE EMBS, pp. 49074910, Aug. 2007. 22. G. Shuman,2009. “Using forearm electromyograms to classify hand gestures”, IEEE International conference on Bioinformatics and Biomedicine, pp. 261264, 2009. 23. T. S. Saponas, D. S. Tan, D. Morris, and R. Balakrishnan, 2008, “Demonstrating the feasibility of using forearm electromyography for musclecomputer interfaces,”, Proceedings of 26th SIGCHI Conference Human Factors Computing System, Florence, Italy, 2008, pp. 515524. 24. I. Moon, S. Joung, and Y. Kum, 2002, “Safe and Reliable Intelligent Wheelchair Robot with Human Robot Interaction ”, Proc. of IEEE Int. Conf. on Robotics and Automations (ICRA02), 2002. 25. Kundu, A.S., Mazumder, O., Chattaraj, R., Bhaumik, S. “Close loop control of non-holonomic wmr with augmented reality and potential field. ”In: Engineering and Computational Sciences (RAECS), 2014 Recent Advances in. 2014, p. 15. 26. Mazumder, O., Kundu, A.S., Chattaraj, R., Bhaumik, S. “Holonomic wheelchair control using emg signal and joystick interface.”In: Engineering and Computational Sciences (RAECS), 2014 Recent Advances in. 2014, p. 16. 27. Kundu, A.S., Mazumder, O., Chattaraj, R., Bhaumik, S. “Door negotiation of a omni robot platform using depth map based navigation in dynamic environment. ”In: Contemporary Computing (IC3), 2014 Seventh International Conference on. 2014, p. 176181. 28. Biswas, K., Mazumder, O.,Kundu, A.S., “Multichannel fused EMG based biofeedback system with virtual reality for gait rehabilitation. ”Intelligent Human Computer Interaction (IHCI), 2012 4th International Conference on. 2012, p. 1-6. 29. Mazumder, O., Kundu, A.S., Bhaumik, S. “Design of wearable, low power, single supply surface EMG extractor unit for wireless monitoring. ”Proceedings of the 2nd International Conference on Nanotechnology and Biosensors, 2011, p. 69-74.