The precise control of manipulators with high joint-friction using base force/torque sensing

Abstract

Joint friction is a major problem in accurate robot position control, particularly during low-speed, small-amplitude tasks. This paper proposes a simple, practical, and effective method to compensate for joint friction, using a six-axis force/torque sensor mounted under the manipulator. From these measurements, joint torques are estimated and used in a torque controller, which virtually eliminates friction and gravity effects, providing high-precision motion control even for small motions at low speed. The method does not require complex analytical friction models. The method also does not require expensive and unreliable internal joint-torque sensors. Experimental results demonstrate the effectiveness and practicality of the method for an electrical PUMA and hydraulic Schilling Titan II manipulator.

Introduction

Precise position and force control is required for many important applications, including the assembly of precision-component systems, micro-manipulation, and robotic surgery (Ku & Salcudean, 1996). These applications often require very small motions at low speeds. Under such conditions, the degrading effects of joint, actuator and transmission friction can dominate system behavior, making precise position and endpoint force control of robotic manipulators difficult to achieve.

The problem is compounded when the manipulator must move heavy objects at slow speeds with high precision. The important problem of nozzle dam placement in nuclear power plant maintenance is such an application (Zezza, 1985). In this application, an hydraulic or highly geared electrical manipulator is required to precisely place a large, heavy nozzle dam. In hydraulic manipulators, joint seal friction is often very high, and likewise in highly geared electrical manipulators, transmission friction is often high. Hence, obtaining precise motion and force control in these systems is very difficult.

Substantial research has been devoted to improving friction-degraded manipulator performance. Direct-drive electrical robots have been proposed and developed that have relatively low joint friction (Asada & Youcef-Toumi, 1987). However, these systems are not appropriate for applications that require the applications of large forces. Also, direct-drive actuators are heavy compared with geared actuators, and thus are not appropriate for applications that require lightweight manipulators, such as space applications (Schenker et al., 1997).

Control methods have also been proposed to reduce the effects of joint friction. Some methods incorporate a mathematical model of friction (Canudas de Wit, 1988). An estimate of frictional forces provided by the model is used either in feedforward or feedback compensation (Armstrong, 1991; Popovic, Shimoga & Goldenberg, 1994). These methods require a precise model. Unfortunately, friction is a highly complex, nonlinear phenomena that can depend upon numerous factors, including joint position, load, temperature, and wear (Armstrong, 1991). These factors can be challenging to estimate or measure, and thus model-based friction compensation remains difficult to implement in practical applications.

To overcome modeling difficulties, adaptive methods have been proposed (Canudas de Wit, Olsson, Astrom & Lischinsky, 1996). Dither has also been utilized to improve low-speed positioning performance (Ipri & Asada, 1995). A method that utilizes short-duration torque pulses computed from fuzzy-logic reasoning has also been studied (Popovic, Gorinvesky & Goldenberg, 1995). This approach can be effective in applications where only the final end-effector position is critical. The method cannot control the trajectory that a manipulator takes to reach this final position.

Finally, measurement-based friction compensation methods have been studied (Luh, Fisher & Paul, 1983; Pfeffer, Khatib & Hake, 1989; Vischer & Khatib, 1995). In these methods, the torque applied to a manipulator's links is measured and used as the feedback signal in a torque control loop. Friction is an output disturbance to this control system. With sufficient gain and bandwidth, the torque controller can reject frictional effects. These measurement-based methods have the advantage of being model-free, and have been shown to be effective in practice (Pfeffer et al., 1989; Vischer & Khatib, 1995). However, this approach requires torque sensors mounted at the joint transmission outputs. The use of “indirect sensing” at the actuator level (such as motor current measurements or differential pressure in hydraulic systems) is not appropriate for friction compensation, because the friction disturbance is not measured by these methods. Joint-torque sensors have the drawback of added cost, increased joint flexibility, and additional cabling and electronics. Their complexity can reduce system reliability. Finally, internal sensors must be included during the design of a manipulator, as it is difficult to add them to the existing systems.

In this paper, a new measurement-based joint friction compensation method is presented. The method, called base sensor control (BSC), uses a six-axis force/torque sensor placed under the base of the manipulator (Morel & Dubowsky, 1996; Iagnemma, 1997; Iagnemma, Morel & Dubowsky, 1997). From the measured forces and torques it is possible to calculate the net dynamic torque applied to the links of the manipulator. This measurement is uncorrupted by joint friction. These calculated torques are used in joint-torque controllers. This method eliminates the need for internal joint sensors with the practical problems described above.

The BSC method in its most general form requires dynamic and gravitational models of the system, but no friction model. Nonetheless, the dynamic and gravitational model calculations can be a burden. It is shown that a nearly model-free form of the method can be applied successfully in applications that require low-speed, small-amplitude motions. Experimental results are presented for low-speed, small-amplitude tasks that show that the reduced (i.e. model-free) method can achieve very fine performance. The reduced method requires only simple kinematic coordinate transformations to implement, and hence is easy to apply and does not require substantial computational resources. Results are presented for BSC control of a highly geared PUMA electric manipulator and a hydraulically powered Schilling Titan II manipulator. For both systems, low-speed small-motion performance is greatly improved, even while transporting heavy payloads.