An EMG Enhanced Impedance and Force Control Framework

An EMG Enhanced Impedance and Force Control Framework for Telerobot Operation in Space Ning Wang¹, Chenguang Yang², Michael R. Lyu¹, and Zhijun Li³ ¹Dept. of Computer Science & Engineering, The Chinese University of Hong Kong, Hong Kong ²School of Computing and Mathematics, Plymouth University, United Kingdom ³Key Lab of Autonomous System and Network Control, College of Automation Science and Engineering, South China University of Technology, Guangzhou, China

Outline Introduction Tele-robotics in space Tele-impedance control EMG signal characteristics Working framework Simulation & demonstration Conclusion & future work 2

What’s telerobot? Robotics Deals with design, construction, operation, and application of robots. Interdisciplinarity: control, mechanics, artificial intelligence, etc. Tele-operation Employs automated machines to take the place of humans. Remotely operation from a distance by a human operator, rather than following a predetermined sequence of movements. Telerobot 3 Tele-operated robot.

Telerobot operation challenge Local human operator and remote autonomous robot Exchange of force and position signals, i. e. , haptic feedback. Long-range communications suffer from time delay. Big challenge Control instability! Delayed transmission of haptic signals lead to instability in robot control. Possible solutions? 4 Wave scattering, passivity, small gain theorem, etc. Remains a difficulty.

Telerobot operation status quo In space Requiring stability. Handling unpredictable environments. Neural path of human being also subject to time delay. In presence of time delay, Human neural control can easily maintain stability. Humans show even superior manipulation skills in unstable interactions. Transfer skills from human operator to robot! Tele-impedance 5 Operation stability of humans comes from adjusting mechanical impedance. Transferring a human operator’s muscle impedance to a telerobot.

Principle of tele-impedance Tele-impedance using electromyogram (EMG) (Ajoudani et al. , 2011). 6 Estimating stiffness and force from EMG signal. Transferring impedance from human operator to robot.

Control strategy Reference task trajectory: qr(t), t∈[0, T]. Impedance and feed-forward torque: with minimal feedback 7

Research focus A framework of EMG enhanced impedance and force control for telerobot operation in space Real-time extraction and processing of EMG. On-line estimation of human muscle impedance and force. Performance demonstration in simulated unstable scenario. 8

EMG signal Physiological signal generated by muscle cells. Reflects human muscle activations and tensions. 9 Long been utilized for human motor control. Suitable for extracting force and impedance of human muscles.

How to acquire EMG data? Data recording Noninvasive electrodes. Bi-dimensional electrical field on the skin surface. Generated by summation of motor unit action potentials (MUAP). Surface EMG 10

Amplitude and frequency properties in EMG An EMG signal is typically a train of MUAP. A band-limited signal that describes the kth EMG wave is characterized by two sequences: -- amplitude; -- phase. AM-FM Signal modeling 11 Signal decomposition. Primary component identification: amplitude A(n) and frequency Ω(n).

Observations: EMG signal decomposition EMG & decomposed waves in 5 frequency bands: 12 Band 1: 10 -100 Hz Band 2: 100 -200 Hz Band 3: 200 -300 Hz Band 4: 300 -400 Hz Band 5: 400 -500 Hz

Observations: primary EMG components Instantaneous amplitude estimate A(n) and frequency estimate Ω(n) in the decomposed EMG waves 13

Working Framework EMG enhanced impedance and force control based teleoperation system in a typical aerospace operation scenario. 14

How to estimate stiffness from EMG? Human muscles and tendons act as a spring-damper system during movement. Changing stiffness via co-activation of antagonistic muscle pairs. Tele-operation by adjusting co-activations and corresponding endpoint stiffness profile (Ajoudani et al. , 2011). Discarding up to 99% of EMG signal power before estimation (Potvin et al. , 2003). involving only 400 -500 Hz (Band 5)! 15

Stiffness estimation formulation Assuming linear mapping between muscle tensions and surface EMG Endpoint forces in Cartesian coordinates: Processed EMG amplitudes in 400 -500 Hz band At ith agonist muscle: At jth antagonist muscle: Parameter set: 16 , and

Stiffness estimation method Iterative least squares (LS) approach to achieve online estimation of parameter set. Online endpoint force and stiffness estimation. 17 Based on proportional muscle stiffness-torque relationship. Expressions under Cartesian coordinates

Force estimation The key idea: Filter most of the low frequency power of the EMG signal, i. e. , use only Band 5 EMG signal. Nonlinearly normalized With is obtained by linearly normalized to 100% of the maximum. Involved muscles: FCR (flexor carpi radialis), ECR (extensor carpi radialis) FCR ECR 18 Force Estimation & Torque Calculation Wrist Torque

Simulation Experimental set-up: Two-joint simulated robot arm with the first joint motionless. Right wrist of human operator in charge of simulated robot arm. Motion reference trajectory at initial position. Implemented using Matlab Robotics Toolbox in Simulink. 19

Demonstration 20

Observations on result Stiffness K and damping rate D: 21 Stiffness K and damping rate D enlarged dramatically after impedance increase.

Observations on result Angle shifting of simulated robot arm from reference trajectory (initial position at 0 radian). 22 Shifting angle reduced greatly after impedance increase.

Conclusions Transferring muscle impedance from human to robot introduced for reducing instability and enhancing control performance of tele-operation. Real time processing of EMG signal proposed for impedance and force estimation. Integrated framework built for the telerobot in aerospace applications to fully capture operator’s control skills. Promising demonstration results shown for impedance control in simulated scenario. 23

What’s the next step? Complete experimental studies on physical robot arm is planned to carry out to test and validate the framework proposed in this paper. 24

Thank you very much! Q&A 25