When the Cyber Backdrives the Physical: Information Flow vs. Power Flow
Abstract
A course entitled “Embedded Control Systems” that has been taught at the Univeristy of Michigan since 2000 and recently at the Swiss Federal Institute of Technology (ETH) has generated a rich supply of not only teaching materials but also research questions. Students are challenged to implement virtual (cyber) environments on the Freescale MPC 5553 microprocessor that can be manipulated through a custom built single-axis haptic interface. Students work in teams of two to program various virtual environments starting with a virtual wall all the way to a networked multi-vehicle driving simulator during the 14-week course. Coupling physical environments (including the student’s own arms and hands) to cyber environments gives rise to many interesting phenomena. For example, dissipativity appears in the coupled dynamics that include the human, even when the cyber system is conservative and the damping in the haptic device is accounted for. In one laboratory exercise, students implement a virtual spring-mass system and apply a “step input” by rapidly turning the haptic wheel, holding it in its new position, and feeling the reaction torque. While the expected step response of a simple spring-mass system is sustained oscillation, students instead feel oscillations that quickly die out. Evidently dissipativity arises from the inability of users to hold the haptic wheel immovably and from damping in the physical system (including the human biomechanics). Understanding such phenomena can provide insight into the design of cooperative multi- agent systems when one of the agents might be a human operator.
Empirical models of the driving point impedance of the hand and arm exist, but these are insufficient when volitional control is also involved (as in the step input experiment). We seek to augment models of the passive characteristics of human body with features that capture volitional control, muscle dynamics, and neuromuscular reflexes. The step input is relatively simple in that it includes two phases that are somewhat separated in time: the active turning and the passive holding of the haptic wheel. The turning part is under volitional control, while holding would seem to involve only open-loop control. However, the duration of the experiment is also long, so reflexes cannot be precluded. Significant gaps remain in the determination of a coupled model that captures the user’s neuromotor intent, the biomechanics of the hand/arm turning and holding the haptic wheel, and the cyber spring-mass system.
In this third year of the project, we aim to develop a simple competent human user model that describes the observed phenomena. We have found that a second order spring-mass-damper model describes the source impedance with which a human user is able to impose a position input on the haptic device, and that this impedance model can be incorporated into a model of the user's neuromotor intent by placing the spring as a series elastic element with a motion source. The physical spring-mass-damper model is identified using frequency domain system identification techniques and then is coupled to the cyber spring-mass system. We use additional experiments to identify the smooth inputs generated by the user's neuromotor system. Simulation studies based on the proposed coupled model have shown similar phenomena to the experimental data and it is the damping component in human hand that causes the energy to decay. There also exist some differences between the experimental data and simulation results. We will investigate the reasons for the differences and further improve our coupled model. Specifically, we will model the neuromuscular systems of the human user, adopt a more complex human muscle model, and investigate the variation in the values of the impedance of the human user under volitional control. It is also interesting to investigate how the limitation in our model will affect the performance of a multi-agent system. We are also currently developing teaching materials so that we can deliver our research results to students, and help them to understand this phenomena and related research topics.
Award ID: 1035271
Submitted by James Freudenberg
on
Abstract
A course entitled “Embedded Control Systems” that has been taught at the Univeristy of Michigan since 2000 and recently at the Swiss Federal Institute of Technology (ETH) has generated a rich supply of not only teaching materials but also research questions. Students are challenged to implement virtual (cyber) environments on the Freescale MPC 5553 microprocessor that can be manipulated through a custom built single-axis haptic interface. Students work in teams of two to program various virtual environments starting with a virtual wall all the way to a networked multi-vehicle driving simulator during the 14-week course. Coupling physical environments (including the student’s own arms and hands) to cyber environments gives rise to many interesting phenomena. For example, dissipativity appears in the coupled dynamics that include the human, even when the cyber system is conservative and the damping in the haptic device is accounted for. In one laboratory exercise, students implement a virtual spring-mass system and apply a “step input” by rapidly turning the haptic wheel, holding it in its new position, and feeling the reaction torque. While the expected step response of a simple spring-mass system is sustained oscillation, students instead feel oscillations that quickly die out. Evidently dissipativity arises from the inability of users to hold the haptic wheel immovably and from damping in the physical system (including the human biomechanics). Understanding such phenomena can provide insight into the design of cooperative multi- agent systems when one of the agents might be a human operator.
Empirical models of the driving point impedance of the hand and arm exist, but these are insufficient when volitional control is also involved (as in the step input experiment). We seek to augment models of the passive characteristics of human body with features that capture volitional control, muscle dynamics, and neuromuscular reflexes. The step input is relatively simple in that it includes two phases that are somewhat separated in time: the active turning and the passive holding of the haptic wheel. The turning part is under volitional control, while holding would seem to involve only open-loop control. However, the duration of the experiment is also long, so reflexes cannot be precluded. Significant gaps remain in the determination of a coupled model that captures the user’s neuromotor intent, the biomechanics of the hand/arm turning and holding the haptic wheel, and the cyber spring-mass system.
In this third year of the project, we aim to develop a simple competent human user model that describes the observed phenomena. We have found that a second order spring-mass-damper model describes the source impedance with which a human user is able to impose a position input on the haptic device, and that this impedance model can be incorporated into a model of the user's neuromotor intent by placing the spring as a series elastic element with a motion source. The physical spring-mass-damper model is identified using frequency domain system identification techniques and then is coupled to the cyber spring-mass system. We use additional experiments to identify the smooth inputs generated by the user's neuromotor system. Simulation studies based on the proposed coupled model have shown similar phenomena to the experimental data and it is the damping component in human hand that causes the energy to decay. There also exist some differences between the experimental data and simulation results. We will investigate the reasons for the differences and further improve our coupled model. Specifically, we will model the neuromuscular systems of the human user, adopt a more complex human muscle model, and investigate the variation in the values of the impedance of the human user under volitional control. It is also interesting to investigate how the limitation in our model will affect the performance of a multi-agent system. We are also currently developing teaching materials so that we can deliver our research results to students, and help them to understand this phenomena and related research topics.
Award ID: 1035271