The master is a unilateral aluminum manipulator with four revolute joints. It tracks the shoulder-elbow motions of the right arm of the human operator (see Fig.1, 17Kb). The manipulator is fixed at one end to the chair that PHI is mounted on. The other end of the manipulator is a handle that is grasped by the operator. The master is strapped to the right arm of the human operator by using a set of inflatable cushions and various length adjustments to accommodate different operators. The spherical joint of the shoulder is implemented by three cylindrical joints J1, J2, and J3, with orthogonal axes intersecting at the center of the shoulder. The elbow joint is implemented by the cylindrical joint J4.

The joint displacements are acquired by using linear position transducers mounted parallel to each cylinder. The signals from the transducers are sent to a custom built electronic board for analog differentiation. Then, the analog signals that represent the kinematic data for each piston are transmitted to A/D converters housed in a Power Macintosh 7100. The digital position and velocity data, are then transmitted and SGI Indigo2 workstation that implements the virtual slave and the environment.

The force feedback is relized by actuating the joints of the master manipulator by the pneumatic cylinders (P1, P2, P3, and P4). The stroke of the pistons and the connection points on the structure are chosen to provide the operator with a convenient range of motion. A photograph of the system is presented in Fig.2 (124Kb). Each double-acting cylinder is connected to a pneumatic proportional valve. We have used electrically actuated 4-way valves with voice coils (Numatics PositionX Valve). Pressure transducers are used as sensors in the control loop that is responsible for applying the desired forces at the actuators. The pressure signals are amplified and communicated to the A/D converters on the Mac that houses the force actuation controller.

Two control systems are developed and implemented on-line to perform the force actuation and gravity compensation that are needed by the master manipulator. The algorithms for both controllers reside in the Mac 7100 computer. The pneumatic force actuation controller is a modified PD (Proportional plus derivative) algorithm that includes specialized schemes to compensate for the aerodynamic effects in the pneumatic circuitry and the cylinders. The gravity compensation controller is a neural network based algorithm that can be trained on-line to compensate for the gravity vector of the master manipulator and provide transparent interface.