Lorenzo Masia

Machine Learning for Online Inverse Dynamic Model Estimation applied in a Soft Wearable Exoskeleton

Machine Learning for Online Inverse Dynamic Model Estimation applied in a Soft Wearable Exoskeleton

The dynamics of a robotic system characterizes the relationship between the kinematics (i.e. positions, velocities, and accelerations) and the applied forces/torques. The closed-form equation describing this relationship can be formalized as follow:

 eq1

where are positions, velocities and accelerations respectively; denotes the applied forces/torques. represents the inertia matrix, represents the Coriolis and centripetal forces, denotes the gravitational forces, and denotes the unmodelled dynamics of the system.

 pic1

 Figure 1(A): Control scheme for the computed torque controller.

pic2

Figure 1(B): Control scheme for the computed torque controller with inverse dynamic model learning.

Figure 1(A) shows the control scheme for the computed torque controller. The control law determines the forces/torques for following the desired trajectory . It consists of two terms: feedback term which is a linear PD controller stabilizes the system, whilst feedforward term is required to perform the desired trajectory, and it is a function of . Thus it is important to estimate the relationship between  and , this relationship is referred as inverse dynamic model.

eq2

In this work, we chose Extreme Learning Machine and modified it in the localization fashion (Localized ELM) to estimate the inverse dynamic model of an actuation stage driving a soft exosuit for human arm assistance (see References for details).

 pic3

 Figure 2: The exosuit and actuation stage worn by a healthy subject.

 

pic4

Figure 3: Desired and measured elbow joint trajectory driven by the exosuit and the embedded actuation stage. (a) with no feedforward term, a significant time delay and inaccuracy in position tracking show that the feedback term alone is not sufficient to compensate for the nonlinearities of the system. (b) when the feedforward term is activated, the amplitude mismatch and time delay are considerably reduced.

Figure 3(a) shows three repetitions of elbow flexion/extension movements tested on a healthy subject wearing the exosuit. With only simple PD controller, the nonlinear behaviours of the system, such as friction and backlash, introduce a significant time delay and a large amplitude mismatch between the desired and measured trajectory. The Root Mean Square Error (RMSE) is over 33 [deg]. Figure 3(b) shows the same trial obtained when the inverse dynamic model estimation for computing the feedforward torque is activated and continuously updated by Localized ELM. The feedforward term clearly improves the tracking performance, reducing the RMSE value to 3.67 [deg].

References:

  1. Binh Khanh Dinh, Cappello L.,Masia L. "Localized Extreme Learning Machines for Online Inverse Dynamic Estimation in Soft Wearable Exoskeleton", International Conference on Biomedical Robotics and Biomechatronics (BIOROB), Singapore, June 2016.
  2. Binh Khanh Dinh, Cappello L, Xiloyannis M, and Masia L. 'Position Control using Adaptive Backlash Compensation for Bowden Cable Transmission in Soft Wearable Exoskeleton'. IEEE International Conference on Intelligent Robots and Systems IROS 2016, Daejeon, South Korea 9th-14th October 2016
  3. Binh Khanh Dinh, Xiloyannis M, Cappello C, Antuvan CW, Yen SC, Masia L. "Adaptive Backlash Compensation in Upper Limb Soft Wearable Exoskeletons ", Robotics and Autonomous Systems (Accepted for Publication).
  4. Dinh BK, Xiloyannis M, Antuvan CW, Cappello L, Masia L. 'A Soft Wearable Arm Exoskeleton based on Hierarchical Cascade Controller for Assistance Modulation'IEEE Robotics and Automation Letters and ICRA 2017, Singapore.
  5. Xiloyannis M., Cappello L., Khanh D. B., Antuvan C. W., Masia L. "Preliminary design and control of a soft exosuit for assisting elbow movements and hand grasping in activities of daily living", Journal of Rehabilitation and Assistive Technologies Engineering (RATE), October 2016.
  6. Leonardo Cappello, Khanh Dinh Binh and . 'Design of SARCOMEX: a Soft ARm COMpliant Exoskeleton'In 2016 6th IEEE International Conference on Biomedical Robotics and Biomechatronics (BioRob) 26-26 June, Singapore.

 

 

 

Conference

IEEE BIOROB 2016 (clickhere)

 

Editors

  • L. Masia, Nanyang Tech. University
  • S. Kukreja, National University of Singapore

 

 


New Journal on Rehabilitation Technology

Journal of Rehabilitation and Assistive Technologies Engineering (click here)

Lorenzo MASIA (Associate Editor)

 

 

Journal Topics

Research Topic (click here)

 

 

Red

IEEE ICORR 2015

International Conference on Rehabilitation Robotics

Singapore 11th-14th August 2015 Hosted by NTU

General Chair: Wei Tech ANG

Program Chair: Lorenzo MASIA

 

Conference

Lorenzo MASIA Editor of Biorob 2014

 

 

Hot the NEW WrisBot

 

Top

 

visit the robot report website for a comprehensive view of robotic business worldwide and other great info!

http://exoskeletonreport.com/

http://www.therobotreport.com/

http://www.robobusiness.eu/

http://www.italiadelfuturo.cnr.it/en