Distributed Embedded Control Systems in Robotics - Ten years of development Lars-Berno Fredriksson ABSTRACT Around 1980 the development of micro controllers had reached such a level that they were efficient and powerful enough to integrate digital servo controllers into each actuator of an industrial robot and to connect them to a serial Controller Area Network. The development of such a robot started in 1982 by the company Rovac in Sweden and was successfully demonstrated two years later. For various reasons this design philosophy was not commonly accepted to begin with but during the latest years the interest in the same has been growing. One reason for that is the CAN protocol, initially developed for multiplexed systems in cars. CAN is a standard high speed serial protocol and there are several low cost chips available with this protocol implemented, suitable for distributed embedded control systems in robots. Ten years after the first demonstration of a robot controlled by a distributed embedded control system the development of such systems finally steps up the pace. One example of this is a Swedish robot for micro mechanic production with 14 axis where all control loops are closed via the CAN network. The control system for this is designed by KVASER. INTRODUCTION A distributed embedded control system for a robot offers several advantages. A robot can be constructed of standard actuators, each with an integrated servo controller, connected to each other by distance elements. Power and control signals are distributed to the actuators via a power bus and a serial bus. The size of such a robot can then easily be customized. The servo controllers can be physically identical and tuned to actual dynamic needs by setup parameters, downloaded via the serial communication. This facilitates the service of the robot. As all communication takes place digitally over the signal bus, wiring is reduced to a minimum. A decade ago the main obstacle for such a design was the lack of small, low cost and low power consuming micro controllers, powerful enough for the task of closing a servo loop. In the beginning of the eightieth the Hitachi 6301 micro controller, a CMOS version of the Motorola 6809, offered a suitable building block for distributed embedded control systems for robotics and this was the starting point for the development of this architecture. THE FIRST PROJECT During the late seventieth a new method of producing large glass fiber products automatically by a spray-up technique was invented. A company, Rovac, was formed to exploit and further develop the idea. A pilot plant was built and the first products left the production line in February 1980. The heart of the plant was a big vacuum chamber with a standard hydraulic spry-up painting robot inside. The robot sprayed polyester and glassfiber on molds that were passed into and out the chamber through gates. All parts of the system was standard products, available on the market. The control system was composed of standard sub-systems, eg. robot control, resin temperature control, vacuum, conveyors, etc. In all there were seven sub-systems that had to be electronically interfaced to each other into a main control system plus some more systems that were coordinated by the operator. As the sub-systems was produced by independent producers, the interfacing was problematic and in most cases reduced to on/off signals. Although it was possible under favorable conditions to produce high quality products, like boat hulls and antenna disks, it was soon realized that the production method needed a fully integrated control system. Just one example: If the movement of the robot could be coordinated with the material pumps, there was a potential to reduce material cost by 50%. As the material cost for a glass fiber boat hull represents 50% of its value, you immediately recognize the big potential of our system. There was at that time not any system on the market capable of such a broad range of controlling tasks as ours. It was therefore decided that we would develop a suitable system on our own. The basic idea was to make a system with a network architecture. Any control task should be carried out by a local controller as near the source as possible by a local micro controller. Any set point value or measured value should be digitally transmitted via the network. We started the work by designing a new robot. The old one had some severe mechanical drawbacks, the most severe one being a limited working space. A design that could be easily customized in size for different products was preferred. Wires and hoses were also an headache fore the service people. The new concept was based on an overhead crane type where actuators were connected to each other by distance parts. Whenever possible the hydraulic oil and the different plastic liquids were distributed by tubes and swiveling joints. Each actuator had a built in controller that communicated with a central computer via an RS-422 bus. The controller unit built in at each actuator measured 10 cm x 10 cm and received every required setup data via the network. The design turned out to be a not only a technical success, leading to patent [1], but also an administrational one as the concept lends itself to project organized development work. During the development three groups worked in parallel: One group with the overhead X-Y table, one group with the robot arm and one group with the central processor. The groups could to a great extent work independently of each other as the network protocol was used as an essential part of the specification. The nodes should get their set-point values 20 times every second and deliver their actual values at the same pace. Mechanically and dynamically the actuators were thoroughly specified. They could then be individually tested with dummy loads and communicating with a computer simulating the rest of system. The protocol was designed to have a very small overhead and used a bit rate of 9600 bit/s. The central processor took care of all coordinate transformations needed. The programming of the robot was of a power teach type and the programmer taught the robot by force sensing handles. A program created in this way could be loaded into an ordinary computer and be edited at the desk-top at a later time. The development of the prototype was finalized in 1984 and was presented at the 14th International Robot Conference in Gothenburg in October of the same year. Later in 1984 the Rovac company went into financial problems. Three engineers of the company staff decided to buy the rights to the robot system and further develop the ideas in the company KVASER. We were fully convinced that the architecture of the Rovac system was the right thing for the future and that the benefits would be obvious for everyone. During 1985 we made the controllers more general and started to market the system. The marketing turned out to be difficult. Although we could demonstrate a working robot, most customers were reluctant. They made three alleged main objections: Firstly that systems should not be physically distributed as electronics belongs to cabinets. Secondly that an asynchronous serial communication cannot meet realtime requirements. Thirdly that no one else than KVASER used the proprietary protocol "Trainet". Nevertheless one company, VIBAB [2] in Sweden, accepted some of the ideas and designed a control system with one micro controller per axis but still placed in a cabinet. The serial bus was reduced to some 30 cm. This was in 1986 and the very same design is still in production. Some 70 systems are working in pick and place robots for blanks and stamps in sheet forming lines, mainly in the automotive industry. CAN, A SUITABLE STANDARD PROTOCOL In 1982 a project was started at BOSCH in Germany on a serial bus and protocol suitable for multiplexed control systems in cars. The protocol was designated CAN [3]( Controller Area Network ). Some years later Intel joined this project and in 1988 the Intel 82527 chip was introduced to the market. The chip was a stand alone CAN Controller with the full protocol implemented in silicon. Since then several other chip manufacturers have included CAN components in their product line, e.g. National, NEC, Motorola, Philips and Siemens. The CAN Controller takes care of ordinary communication tasks such as collision detection and resolving, error checking and, in case of a corrupted message, automatic retransmission. The maximum data length in one transmission is 64 bits and seven such messages can be sent within one millisecond at a bit rate of 1 Mb/s. CAN is used in the popular Bosch "rho 3" robot control system [4]. The architecture of this is the same as of the VIBAB robot earlier described. The servo controllers are placed in a cabinet and the CAN network is reduced to a back plane bus. This architecture is used also for other control systems than for robots. The German company Dornier introduced a similar control system for weaving machines in 1990. In my opinion the main reason for this architecture is the organization of the development department. This is the natural choice for a system producing company with a traditional development department organized by separate design groups for mechanics, electronics, electric, etc. but lacking or having a weak systems group. CAN KINGDOM The structure of the robotics market with a few big and strong companies producing industrial robots and many end users integrating them into production systems has created a low incentive to find more efficient architectures. The end user accepts the robot as a complete sub-system and the robot manufacturer dictates the design of all components in his product. Therefore we have to look into other lines of business to find the new trend in advanced realtime control systems. Such a line is weaving machines. One crucial machine element to make an efficient weaving machine is the yarn feeder. Yarn feeders are produced by a few independent companies but the feeders have to be carefully integrated into the weaving machine control system. As this integration has rapidly been more and more complex, it was clear that a concept had to be developed that separated the responsibility between the node designers and the system designer. This concept was designated "CAN Kingdom" [5] and founded on the CAN protocol. It is beyond the scope of this article to describe CAN Kingdom but in a few words it can be characterized as a set of higher layer protocol primitives that are supported by the nodes. The system designer can then create an optimized communication protocol for his system by using these primitives when the system is initiated. A positive outcome of this concept is that the nodes and the system will be documented in a systematic and conform manner. STATE OF THE ART A current project is a 14 axis CAN robot for production of micro mechanical parts inside a Scanning Electron Microscope. As the vacuum chamber volume is very expensive a hardware architecture was chosen that minimizes the need to place non-moving parts, as electronics and cables, inside the chamber. Thus servo controllers are placed outside and only CAN interfaced position sensors and motors inside the chamber. The sensors and servo controllers are connected to a common CAN bus allowing each closed loop to be updated at a rate of 400 Hz. A second CAN bus connects the servo controllers to a PC-node and to a Man-Machine-Interface node with three joy sticks, some buttons and a display. There are two modes to program the robot: Either on-line via the joy sticks and buttons in an ordinary power teach mode where the PC records the set-point values or off-line by an artificial PC program. The resolution of the robot positions is 10 nanometers. The sensor and servo control electronics used are of a standard type, all supporting CAN Kingdom. KVASER participates in the electronic part of this project. DAMEK mechatronics research group are carrying on extensive research in the field of distributed architectures and their work is reported during this conference by themselves. CONCLUSION After ten years of development we are still in the dawn of a new era in robotics, characterized by fully distributed, embedded realtime control systems where robots are parts of integrated production systems. REFERENCES "Arrangement comprising a system providing movement, processing and/or production", PCT International Publication Number WO 85/01007, date: 14 March 1985. VIBAB, P.O. Box 2012, S-520 24 Blidsberg, Sweden. "Road vehicles - Interchange of digital information - Controller area network (CAN) for high-speed communication", ISO 11898. "Faster, smaller, more cost-effective. Rho 3, the new robot control from Bosch.", Robert Bosch GmbH, Industrial Equipment Div., P.O. Box 1162, D 64701 Erbach, Germany. L-B Fredriksson, "A CAN Kingdom", 921231, KVASER AB.