Usability of Upper Limb Electromyogram Features as Muscle Fatigue Indicators for Better Adaptation of Human-Robot Interactions

Abstract

Human-robot interaction (HRI) is the process of humans and robots working together to accomplish a goal with the objective of making the interaction beneficial to humans. Closed loop control and adaptability to individuals are some of the important acceptance criteria for human-robot interaction systems. While designing an HRI interaction scheme, it is important to understand the users of the system and evaluate the capabilities of humans and robots. An acceptable HRI solution is expected to be adaptable by detecting and responding to the changes in the environment and its users. Hence, an adaptive robotic interaction will require a better sensing of the human performance parameters. Human performance is influenced by the state of muscular and mental fatigue during active interactions.

Researchers in the field of human-robot interaction have been trying to improve the adaptability of the environment according to the physical state of the human participants. Existing human-robot interactions and robot assisted trainings are designed without sufficiently considering the implications of fatigue to the users. Given this, identifying if better outcome can be achieved during a robot-assisted training by adapting to individual muscular status, i.e. with respect to fatigue, is a novel area of research. This has potential applications in scenarios such as rehabilitation robotics. Since robots have the potential to deliver a large number of repetitions, they can be used for training stroke patients to improve their muscular disabilities through repetitive training exercises.

The objective of this research is to explore a solution for a longer and less fatiguing robot-assisted interaction, which can adapt based on the muscular state of participants using fatigue indicators derived from electromyogram (EMG) measurements. In the initial part of this research, fatigue indicators from upper limb muscles of healthy participants were identified by analysing the electromyogram signals from the muscles as well as the kinematic data collected by the robot. The tasks were defined to have point-to-point upper limb movements, which involved dynamic muscle contractions, while interacting with the HapticMaster robot. The study revealed quantitatively, which muscles were involved in the exercise and which muscles were more fatigued. The results also indicated the potential of EMG and kinematic parameters to be used as fatigue indicators. A correlation analysis between EMG features and kinematic parameters revealed that the correlation coefficient was impacted by muscle fatigue. As an extension of this study, the EMG collected at the beginning of the task was also used to predict the type of point-to-point movements using a supervised machine learning algorithm based on Support Vector Machines. The results showed that the movement intention could be detected with a reasonably good accuracy within the initial milliseconds of the task. The final part of the research implemented a fatigue-adaptive algorithm based on the identified EMG features. An experiment was conducted with thirty healthy participants to test the effectiveness of this adaptive algorithm. The participants interacted with the HapticMaster robot following a progressive muscle strength training protocol similar to a standard sports science protocol for muscle strengthening. The robotic assistance was altered according to the muscular state of participants, and, thus, offering varying difficulty levels based on the states of fatigue or relaxation, while performing the tasks. The results showed that the fatigue-based robotic adaptation has resulted in a prolonged training interaction, that involved many repetitions of the task. This study showed that using fatigue indicators, it is possible to alter the level of challenge, and thus, increase the interaction time.

In summary, the research undertaken during this PhD has successfully enhanced the adaptability of human-robot interaction. Apart from its potential use for muscle strength training in healthy individuals, the work presented in this thesis is applicable in a wide-range of humanmachine interaction research such as rehabilitation robotics. This has a potential application in robot-assisted upper limb rehabilitation training of stroke patients.