We propose the energy scalability and power-delay product scalability as new metrics for calculating and estimating boundaries on the energy overhead of parallel processing systems. Because both properties consider network topology and the algorithm to be implemented, they are shown to be effective in predicting the energy scalability of a multiprocessor in the most general case. Power-delay scalability is then shown to provide upper bound and lower bound estimates when analyzing the energy scalability of a multiprocessor while significantly reducing the computational cost of directly calculating the energy scalability. Energy scalability is an effective metric for predicting the energy overhead cost when using a parallel processor versus a single processor. Power-delay scalability is a means for estimating energy scalability on a system when exact calculation may be difficult due to variation in power levels which have become common with dynamic voltage frequency scaling.

This paper discusses the developing market for medical robotics. It first describes some of the dynamics and market drivers in health care, and then provides an outline of the areas of consideration when developing a commercial medical robot. The paper also offers three case studies of robotic systems that have been commercialized. Finally, it summarizes some of the key ingredients to be considered for the commercialization process.

This paper discusses the technology and developments behind transformations made to a commercial robotic surgical system, Computer Motion, Inc.'s Zeus/spl trade/, in order to make it possible to fully and safely support minimally-invasive human telesurgery performed over very large distances. Because human life is at stake, issues relating to safety, detection of errors, and fail-safe operation are principal in importance. Therefore, it was paramount that all of the safety features of the commercial product Zeus/spl trade/ remained intact during this transformation. This paper discusses the commercial robot and its safety features as well as the real-time communications system added to it as part of Operation Lindbergh, the first transatlantic human telesurgery. Particular attention is paid to the limiting effects of latency. Key techniques developed during this project are discussed, including the need to send the full robot state in each transmitted packet rather than incremental or modal data, thereby treating all telecom problems uniformly as dropped packets. The use of hierarchical design (incorporating Zeus as a drop-in component rather than modifying its internals) allowed the project to focus on the new issues arising due to teleoperation while gaining the robust, error checking, and fail-safe aspects of the design from the use of the unmodified commercial robot. The concept of local mode was created and used during initialization and during communications abnormalities and outages in order to keep the local and remote subsystems active and safely quiescent.

This paper presents an historic achievement. A commercial robotic surgical system was altered to enable remote surgery. This presented huge challenges because the Zeus/sup TM/ system is a tightly-coupled electromechanical system with tight sample and communication loops. We were, however, able to separate Zeus/sup TM/ into a remote surgeon-side system (located in New York City) and a patient-side system (located in Strasbourg, France). Never before have surgeons performed complete surgical tasks in the fashion presented in this paper. The remote surgical task was Laparoscopic Cholecystectomy (removal of the gall bladder) performed on animals. In this unique setup the patient (pig) was readied in the operating room in Strasbourg, France, while the operating surgeon was in Manhattan, NY. The Zeus/sup TM/ robotic system is a very safe surgical system for use in the operating room. For the remote transatlantic case, numerous safety features were built on top of those already available in Zeus/sup TM/. The ultimate goal of this development is to perform a transatlantic Lap Chole case on a human subject. As such, the animal cases were performed with the same setup, personnel, and equipment as will be used for the human case. The choices of the robotic system, networking equipment, and telecommunications lines were critical to the success of the project.

The research discussed in this paper builds on the success of the network time protocol (NTP) which has been synchronizing clocks on hosts all across the Internet for many years. With a different end goal (NTP seeks to setup and maintain agreement of UTC time in network-connected systems) we augment NTP methods with hardware to improve the accuracy of synchronization that is possible within an ethernet subnet by more than three orders of magnitude. Our approach involves a study of the sources of nondeterminism in NTP exchanges and their removal by special hardware aids. The added hardware has a real-time synchronized clock output and a programmable 'event synch' output that can be used to synchronize attached real-time equipment. The hardware aids allow us to achieve much better synchronization precision (100 nsec) while also facilitating the convergence to synchronization in a vastly shorter time-about one minute rather than hours or days as required by NTP.

Describes a real-time system supporting a tele-surgery application based on the Zeus/sup TM/ system. The application involves performing minimally invasive surgical procedures remotely - i.e. without the surgeon being present in the same room with the patient. Because there is human life at stake, the underlying real-time system must be robust, fail-safe and resilient to communications problems. This paper focuses primarily on the approach taken in communicating between the surgeon-side and patient-side subsystems. In particular, techniques for dealing with the challenges of real-time communications are discussed, e.g. bit errors, packet errors and synchronization. In addition to the control actions and feedback data, several serial data streams are multiplexed, transmitted between one side and the other, demultiplexed, and delivered by the real-time system. These streams have a fundamentally different character from the control actions and feedback values used with the robots. Both types of real-time streams are sent over the same communications link.

This paper presents the architecture of a very high performance 4-input, 4-output asynchronous transfer mode (ATM) switch that has been designed as part of the ARPA-sponsored "Thunder and Lightning" project at the University of California, Santa Barbara. This research project is focused on the design and prototype demonstration of ATM links and switches operating at or above 40 gigabits per second per TDM link, with potential scalability to 100 Gbps. Such aggressive link rates place severe requirements on switch architecture, particularly the buffering scheme. In this paper we present the ATM switch structure and justify the main design choices.

The authors discuss a concurrent programming environment and its application to a two-arm cooperative manipulation experiment on RIPS (robot instruction processing system). RIPS is a hierarchical multiprocessor architecture in which various custom and general-purpose processors are applied to a partitioning of the robot control problem. The system provides hardware support for synchronization and communication primitives, making it easier to write concurrent programs for RIPS' heterogeneous processors. The experiment demonstrated the viability of RIPS in supporting computationally intensive robot control methodologies, and as a byproduct has helped to develop a parallel programming environment for RIPS, called USE RIPS (user software environment for RIPS). By building USE RIPS on Unix and using a layered approach, it is possible to adapt or make use of various existing programs and utilities, such as the GNU and C++ compilers.>

The design and performance benchmarking of the prototype robotic instruction processing systems (RIPS) at the Center for Robotic Systems in Microelectronics at the University of California-Santa Barbara, is discussed. The architecture of RIPS is specifically geared for robot control, yet the system is general and fully programmable and does not assume any manipulator characteristics. RIPS provides sufficient computing capacity to attack important real-time sensory-based robot control problems. Compliant control, multiple manipulator coordination, and manufacturing cell-level integration are presently under study. Experimental testing in these areas is possible because of RIPS' high computational performance for robot control problems: a Stanford manipulator using the computed torque method can be controlled with an update time of 255 mu s. This time includes inverse kinematics, dynamic compensation, and proportional derivative servoing for the six-degrees-of-freedom manipulator.>

The design and simulation of RIPS, a robotic instruction processing system, are discussed. The authors are building a prototype version of RIPS which will be used as a general research tool to study problems of robot control. The architecture of RIPS is geared for robot control, yet the system is general and fully programmable and does not assume any manipulator characteristics. There is sufficient computing capacity to address any of several important automation problems such as force control, multiple manipulator coordination, or manufacturing cell-level control. Simulations show that using RIPS to control a Stanford manipulator with the computed torque method gives an update rate of 255 mu s. This simulation includes all interprocessor communication as well as the computing time necessary for equation evaluation.>

In this paper we discuss the demanding computational needs of robot control and describe a new computer architecture optimized for such problems. Inverse kinematics and dynamics for 6 degrees of freedom arms are often used as benchmarks to measure controller speed. According to current estimates, the new processor we describe can compute both of these problems, using 32 bit fixed-point operations, in less than 250 microseconds. This is an order of magnitude faster than existing state-of-the-art systems. Such benchmarks only measure a smallportion of the true robot control problem. The issues of interpolation, input/output, and multiple axis synchronization must also be addressed. As robotics becomes more advanced, the need for more compatationally intensive control algorithms (hybrid position/force control, sensory-based control, multiple robot cooperation, etc.) will grow. These now computationally-intractable problems must ultimately be computed in real-time. In this paper, we discuss current progress on a flexible (i.e. programmable) system which is fast enough to approach such problems. The system is targetted to support up to 40 synchronized axes, performing advanced control algorithms with an update rate over 1 KHz.

In this correspondence we present a statistical model which relates mean computer failure rates to level of system activity. Our analysis reveals a strong statistical dependency of both hardware and software component failure rates on several common measures of utilization (specifically CPU utilization, I/O initiation, paging, and job-step initiation rates). We establish that this effect is not dominated by a specific component type, but exists across the board in the two systems studied. Our data covers three years of normal operation (including significant upgrades and reconfigurations) for two large Stanford University computer complexes. The complexes, which are composed of IBM mainframe equipment of differing models and vintage, run similar operating systems and provide the same interface and capability to their users. The empirical data comes from identically structured and maintained failure logs at the two sites along with IBM OS/VS2 operating system performance/load records.