Disturbance Observer Based Closed Loop Force Control for Haptic Feedback

Most commonly used impedance-type haptic interfaces employ open-loop force control under the assumption of pseudostatic interactions. Advanced force control in such interfaces can increase simulation fidelity through improvement of the transparency of the device, and can further improve robustness. However, closed loop force-feedback is limited both due to the bandwidth limitations of force sensing and the associated cost of force sensors required for its implementation. In this paper, we propose the use of a nonlinear disturbance observer for estimation of contact forces during haptic interactions. This approach circumvents the traditional drawbacks of force sensing while exhibiting the advantages of closed-loop force control in haptic devices. The feedback of contact force information further enables implementation of advanced robot force control techniques such as robust hybrid impedance and admittance control. Simulation and experimental results, utilizing a PHANToM Premium 1.0A haptic interface, are presented to demonstrate the efficacy of the proposed approach. INTRODUCTION Force or haptic feedback can enhance a user’s feel of realism in a virtual environment simulation by conveying touch-related sensory information to the user. The user can be conveyed information about physical attributes of the simulated objects, like hardness, texture or inertia through haptic feedback. It is also ∗Address all correspondence to this author. possible to simulate additional physical forces and fields, which may or may not be part of a natural environment, in order to convey information to the user. This makes a haptic display suitable for a variety of applications like remote operation in hazardous environments and simulators for surgical training [1–3]. According to Massie and Salisbury [4] the following three criteria are employed in the design of haptic mechanisms: • Free space must feel free; • Solid virtual objects must feel stiff; and • Virtual constraints must not be easily saturated. The first criterion implies that apparent inertia and viscosity of the device should be as low as possible. The second criterion implies that the haptic interface should be able to generate sufficiently high stiffness to emulate contact with a rigid object. A stiff mechanism and high bandwidth controller are required to simulate high stiffnesses. The last criterion implies that the force output of the device should be high enough to simulate solid objects. Fidelity of a haptic interface is characterized by the level of impedance discrimination that can be detected at the interface [5]. The primary hindrance to achieving high fidelity are the dynamics of the haptic device, as they appear to the user as a part of the simulated environment. Low force output devices can be designed to have low dynamic properties with use of efficient drive trains (cable, harmonic drives) and high strength-to-weight materials. Counterbalancing can also be used to to remove gravitational effects, although at the cost of increased inertia. 1 Copyright c © 2007 by ASME However, when larger force output is desired, it becomes increasingly difficult to passively reduce dynamic effects of manipulator dynamics. High force output devices require use of larger actuators, drive mechanisms, and linkages leading to increased inertia and friction in the device. Even if the force output is adequate, the dynamics of the device can hinder high fidelity required for display of some details in the environment, which can degrade performance in dexterous tasks. Active control is needed for further reductions in haptic device dynamics. This can be achieved either through model feedforward or force feedback from a sensor mounted at the humandevice interface. While model feedforward can improve performance, it is very susceptible to modeling errors as the impedance displayed by the device will be incorrect in presence of such errors. Carignan and Cleary [5] note that while closed loop force feedback controllers offer a possible solution for reducing device dynamics, the use of force/torque sensors in haptics is limited due to added mass and cost considerations. In this paper the authors present a nonlinear disturbance observer for sensorless closed loop control of haptic interfaces. Katsura et. al note that use of force sensors for closed loop force control is limited due to their limited bandwidth and high cost [6]. As a force sensor employs a strain gauge, it introduces some compliance into the structure of the robot. In order to alleviate the instability associated with force control, large viscous gains are required that slow the robot response. Instead they propose the use of a disturbance observer as a force sensor for contact force control, and demonstrate the efficacy of the same in improving force control performance. Their proposed approach employs a linear plant model for the observer design, thereby requiring the nonlinear components to be canceled separately. In comparison, the method presented in this paper employs a nonlinear disturbance observer and requires no additional modeling or computation. Exponential convergence of the nonlinear disturbance observer for constant disturbances is also shown. The rest of the paper is organized as follows. The next section presents an overview of the haptic controller architectures. Subsequently, the nonlinear disturbance observer design is presented. To conclude, simulation and experimental results, demonstrating the applicability of disturbance observer base force estimation to closed loop haptic control, are presented. HAPTIC CONTROL DESIGN Figure 1 shows a typical haptic system setup in the impedance mode architecture. The interface comprises of the human operator, the haptic device and the virtual environment simulation. Visual interface to the environment may also be present. The human operator imposes motion on the haptic device, which exerts forces back to the operator. Sensors on the device measure position and these position signals are then used to compute changes in the virtual environments. The model of the virtual environment computes desired forces at the human-robot interface based upon the position of the robot and environmental properties. The haptic controller determines motor torque based upon sensed position and/or force and the desired interaction forces as computed by the virtual environment model. Figure 1. Haptic Operator-Interface Loop Control architectures for haptic interfaces can be broadly classified into two classes: impedance control and admittance control. Impedance controlled systems measure the motion commanded by the user and control the force reflected back by the device. Conversely, admittance controlled interfaces measure the force applied by the user and control the velocity or position of the device. In some cases, force or displacements can be used as additional inputs to the impedance or admittance controllers respectively. Open Loop Impedance Control Conventionally, control of most haptic interfaces is achieved through an open loop impedance controller. Figure 2 depicts the block diagram for an open loop impedance controller, where the linearized device dynamics are represented by Zm. Sensors mounted on the haptic device measure the position of the tool tip, x. The controller takes this position of the device, and multiplies it with the environment impedance, Ze, to obtain the desired force output Fd . The desired end-effector force, Fd is mapped to corresponding motor torques through the Jacobian, J. Note that F is an external force applied by the human operator to move the device and Zh is the human impedance. The following relationships describe an open loop haptic interaction 2 Copyright c © 2007 by ASME