We have developed cooperative work system.

Force feedback for each user is realized by a 6 degree-of-freedom master manipulator. Visual information is displayed by visual display device. The one site of the system employs two computers: a graphics computer for real-time monocular image of virtual space, and an I/O computer that supervises sensors and actuators. The I/O computer is equipped with analog-to-digital(A/D) converters and a parallel input/output unit. The graphics and I/O computers are connected by a serial(RS-232C) communication line. The graphics computer are IRIS Indigo2 and IRIS IndigoXS; the I/O computer are NEC PC-9801. Those sites communicate each other via TCP/IP socket connection. Hand position data and a flag which indicates grasping object are transmitted to the other site. At the same time each graphic workstations updates the database of virtual world. In this system, net transmission delay via ethernet is about 0.1 sec.
A 6 degree-of-freedom manipulator was developed as a force display. The manipulator applies reaction forces to the fingers of the operator. The manipulator employs parallel mechanism. The typical design feature of parallel manipulators is an octahedron called "Stewart platform." In this mechanism, a top triangle and a base triangle are connected by six length-controllable cylinders. This compact hardware has the ability to carry a large payload. The structure, however, has some practical disadvantages in its small working volume and its lack of backdrivability (reduction of friction) of the mechanism. In our system, three sets of parallelogram linkages(pantograph) are employed instead of linear actuators. Each pantograph is driven by three DC motors. Each motor is powered by a PWM(Pulse Width Modulation) amplifier. The top end of pantograph is connected with a vertex of the top platform by a spherical joint. This mechanical configuration has the same advantages as an octahedron mechanism has. The pantograph mechanism improves the working volume and backdrivability of the parallel manipulator. The inertia of motion parts of the manipulator is so small that compensation is not needed. The working space of the center of the top platform is a spherical volume whose diameter is approximately 40 cm. Each joint angle of the manipulator is measured by potentiometers. Linearity of the potentiometers is 0.5%. The maximum payload of the manipulator is more than 700gf, which is more than a typical hand.
(2) Graphic computer
Image of the virtual space is generated by graphics work stations, IRIS Indigo2 Extreme and IRIS Indigo XS. The CPU are R4000(100MHz) and R3000(33MHz), which manage virtual space and haptic representation.
mpeg 961kbyte Cooperative
work
In usual virtual reality system using DataGlobes, the users can't grasp virtual object simultaneously. Using force display, we can grasp virtual object simultaneously.

649Kbyte
mpeg : Grasp hand
(1) For the trainee, pulling force toward delayed trainer's hand is presented. Trainee feels like his/her hand is bound to the trainer's hand and is pulled by trainer.
(2) For the trainer, We present pulling force toward delayed his/her own hand. Trainer can check trainee's movement by visual feedback. However the trainer must keep in mind that there is time delay. If the trainer forgets the time delay and he/she reacts quickly to trainee's delayed movement, they are confused and that causes instability. These force plays role like viscosity or friction force to the trainer. The trainer can't move the hand so quickly.
As a usability test of this method, we examined repositioning task of virtual objects. The interface of this test is shown above. We put three targets at similar distance from the start point. There is a virtual ball at the start point. Users can only move it when both trainer and trainee grasp it. The ball is located at the middle point between trainer's hand and trainee's hand during it is grasped. Through the experiment, predictor display[6] is used. It displays current user's hand as semitransparent and delayed hand as solid. We can recognize what will be doing by predictor display. In this experiment, we use virtual balls in stead of flat surface. The reason why we don't use the same environment as previous experiment is that subjects can easily recognize whether they grasp the object or not. The experiment was conducted to estimate this technique under two conditions: with and without force feedback, through 0-s time delay, 1-s and 3-s respectively. Each condition contains 2 trials. We took 6 volunteer subjects from the student of our university. We examined accuracy of tracing the trainer's trajectory. Mean distance between trainer's hand and trainee's is calculated at each program cycle.

The avobe graph shows mean distance of the each condition. Horizontal axis indicates time delay. The mean distances between the trainer and trainee are indicated by bar chart. The data includes error bars which indicates standard deviation. Each value of condition with force feedback is 50% smaller than without force feedback in the same time delay (t=5.9882155, t=6.5586435, t=5.7867037 respectively, and critical value= 2.20098627).
[2]Carl Loeffler. "Distributed Virtual Reality:Applications for Education, Entertainment and Industry" ICAT'93 Proceedings (1993)
[3] M.Ishii,M.Nakata and M.Sato. "Networked SPIDER" PRESENCE, Vol.3,No.4 (1994)
[4] H.Yano and H.Iwata. "Collaboration in Virtual Environment with Force Feedback" JPSJ SIG Notes 94-CG-69(1994)
[5] H.Yano and H.Iwata. "Cooperative Work in Virtual Environment with Autonomous Free-form Surface" The Transaction of The Institute of Electrical Engineers of Japan Vol.115-C,No. 2 (1995)
[6] T.B.Sheridan. "Space Teleoperation Through Time Delay" IEEE TRANSACTION ON ROBOTICS AND AUTOMATION,VOL.9,NO 5 (1993)
>