haptic physical human assistance
Physical assistance of humans implies the artificial increase of human force output and mechanical power output by means of assistive devices. Such assistance is needed if the human body is unable to perform, or repeat, a task due to physical inability, loss of limb function or fatigue. Exoskeletons are passive or active robotic devices that are worn around the upper-extremities, the lower-extremities, trunk or neck. These devices deliver physical assistance and can increase the strength and endurance of the human wearer. The use of active exoskeletons gives the possibility of applying sophisticated methods of control that make an exoskeleton useful for a wide range problems where physical assistance is required. Robotic intelligence can help the user by intentionally shaping the interaction dynamics, i.e. the haptics, of the exoskeleton itself and the environment it interacts with. By masking friction forces, gravity forces and mechanical inertia of large devices it becomes possible to reduce the user’s physical effort to perform physically demanding tasks. In this dissertation we present several improvements for active exoskeletons for physical human-machine interaction in an attempt to make especially upper-extremity exoskeletons more versatile and applicable.
To tackle several problems faced in exoskeleton design we investigated three research questions:
Kinematics & motion: How to support the full range of motion of the human shoulder?
We present a novel visualization method to display and communicate the rotational range of motion (ROM) of both the human shoulder girdle and devices that possibly support such motions. The human shoulder ROM is commonly presented along five medical directions. This neglects any coupling between these directions. Our method is based on the fact that the rotation of the upper arm can be decomposed into three angles that are displayed on a discrete grid in a Mollweide map projection. This makes it possible to evaluate and compare human and device ROM. Also quantitative device properties, such as joint conditioning or dynamical parameters, can be readily displayed and interpreted. The visualization provided by this method enables one to improve the shoulder exoskeleton design that would fit the human’s ROM, kinematic properties and dynamical properties. Additionally, we present the differential inverse kinematics solution for a novel four degree of freedom exoskeleton to assist the full range of motion of the human shoulder. Compared to conventional exoskeletons, an additional, fourth, actuated degree of freedom is used to achieve one degree of freedom of movement redundancy around the shoulder. This redundancy is exploited to avoid kinematic singularity, and body and limb collision. Redundancy resolution is performed through nullspace motions, without influencing the movement of the arm. Instead of gradient projection of some secondary objective, feedback along the one-dimensional self-motion-manifold steers the exoskeleton towards preferred configurations. The proposed inverse kinematics method, together with such a four degree of freedom exoskeleton, allow for collision-free and singularity-free motions of the arm that would otherwise not have been possible with conventional three degree of freedom designs.
Haptics & Control: How to get devices such robots or exoskeletons to behave as some defined impedance in a stable manner when interacting with human users; how to implement stable admittance control with inertia reduction? We present an analysis of admittance control as an interaction control method and give a set of seven guidelines how to implement admittance control to achieve apparent inertia reduction for the user. The design guidelines are derived from a desire for device passivity; dynamical behavior in which the device does not deliver an excessive amount of energy back to the human operator. In this way, safety and stability can be guaranteed during physical human-robot interaction.
Furthermore, we present the main origin of the admittance controller’s inherent active behavior. From this analysis we derive a passivity inspired variable virtual damping controller for admittance control that stabilizes interaction environments characterized mainly by stiffness. By keeping track of the stored energy in the virtual model dynamics and the exchanged energy between the device and the human operator, a variable damper is switched on or off in the virtual model. Although passivity cannot be achieved with this controller, nearly passive and stable behavior can be guaranteed.
Human Factors: How do humans respond to dissipative shared control forces? We present the evaluation of a simple haptic shared controller based on position dependent damping forces. These damping forces act on the user’s hands or arms during fast reaching movements. Such reaching movements represent, for example, transportation movements performed by workers in real-world industrial production processes or transportation of heavy objects. Simple sensors can gather limited environment information and the shared controller applies these damping forces to assist the user in accurate placement and faster movements towards task related targets. By assisting goal directed movements with damping around reaching targets, it is shown that humans increase their reaching accuracy as well as decrease their movement times, when compared to free-air conditions. The relation between increasing end-point accuracy and distance and increasing movement time, known as Fitts’ Law, is shown to hold for these dynamical conditions as well. It is hypothesized that damping forces mitigate the mechanical effects of activation dependent, or multiplicative, motor noise. The attenuation of the effects of noise on end-point accuracy allows for higher muscle forces and accelerations, while still guaranteeing the requested accuracy. Nonlinear stochastic optimal control models show agreement with measured data.
This supports the hypothesis that humans preform their reaching movements while optimizing for energy efficiency and minimizing variance to an amount that the task requires. This research showed that haptic shared control can be passive and as simple as damping around an estimated reaching target. This is an addition to the more common methods of assistive and repulsive potential forces used in haptic guidance, or mechanical constraints imposed in passive Cobot systems. Such damping forces possibly allow for haptic guidance like behavior on completely passive assistive devices.
