【正文】 s is even more unfruitful: On applications level there is no mon standard. Existing standards only regularize specific parts of a vision system, . the munication with frame grabbers [5] or digital cameras [6]. The prenorm DIN V 663111 [7] deals with the interaction between robots and their sensors in general. It defines a frame for munication messages (. message header and order of messages). Here, the robot always acts as master while the sensors act asslaves. The actual mands for controlling different sensors are not part of this prenorm. Furthermore, it did not result in a final norm yet. 3. Visual Inspection Requirements Two main application cases can be distinguished referring to robot movement and munication: (1) Images are acquired from single positions. The trajectory on which the sensor head is moved to these positions does not matter (except for collision avoidance and time optimization). The vision system has to be triggered to start its analysis when a position is reached. (2) A series of images is acquired while the sensor head is moved on a predefined continuous path. Possibly the vision system has to be informed periodically about the position of the sensor head. To gain experience with both application cases two demonstrators are set up: A motor partment inspection (case 1) and a weld seam inspection (case 2). . Motor Compartment Inspection The sensor head, consisting of camera and lighting, is moved to the desired position. The vision system is triggered to acquire an image and analyze it. The result of the analysis (part ok/not ok) is send to the host. Features to be examined are for example the mounting of hoses and clips. The distance of neighboring hoses must not fall below a defined threshold, otherwise they could excoriate. Clips clamping hoses are checked for existence and position. . Weld Seam InspectionHere, the sensor head is moved by the robot on a continuous path which is given by the run of the weld seam. The trajectory is determined in advance. For our demonstrator currently no seam tracking is done. The sensor head contains a camera and a laser source which are used to obtain a profile of the weld seam. For proper results the vision system needs information about the positions from where senor data is acquired. At an adjustable time interval of about 15 ms this information is sent directly from the robot system to the vision system. After tracing and analyzing the weld seam the result is transmitted from the vision system to the host. Main instances of the architecture are a host, a robot and a vision system . Tasks of the robot are receiving mands which define movements, performing movements and announcing its status. The vision system acquires images, analyzes them and extracts features on a high abstraction level. For example in case of an inspection task the vision system can even decide: ‘part ok’ or ‘part not ok’. It may even propose the host better positions of the sensor head. The host is responsible for every task robot or vision system are not responsible for: Management of tasks and results, collision avoidance and path planning, munication with higher level instances. The host can be considered as master. Normally he dispatches mands and waits for the results, but he can be condemned to quietness, when robot and vision system exchange data in realtime mode. Due to the later defined TCP/IPsocket interface the host can run on a separate puter or on either of the puters of the robot or vision system without changes.The modular division in host, robot and vision system and their interfaces (defined in Section 5 and 6) allows an easy exchange of these main ponents. To demonstrate this flexibility the weld seam inspection vision system and the motor partment inspection vision system are both operated with two different robot systems: a REISRV6L robot and a KUKAKR6 robot. Software on host, robot and vision system has not to be changed.5. Communcication ProtocolThe protocol stack which defines munication between system ponents is shown in Figure 1. The presentation is following the ISO/OSI reference model [8].ApplicationXMLTCPIPEthernet 10Mbit/s10BaseTApplicationXMLTCPIPEthernet 10Mbit/s10BaseTLayer 7Layer 6 Layer4/5Layer 3Layer 2 Layer 1 Figure1：Protocol stack. Ethernet with TCP/IP and SocketsAs physical medium a 10BaseT cable is used for 10 MBit/s Ethernet. On layers 2 to 5 the IP and TCP protocols are used. TCP/IP provides connection oriented, acknowledged, reliable exchange of data. Sockets are an encapsulation of the TCP/IP functionality. If munication is handled by TCPsockets, it does not matter whether the municating processes run on the same puter or on different puters connected via Ethernet. Software has not to be modified. Only transfer times will be affected.. XMLAs representation of data XMLdocuments are used [9][10]. XML is a text based format. The structure of every XMLdocument has to follow some general rules (wellformedness). Additionally, it is possible to define own text structures by schemas or documenttypedefinitions (DTDs). In section 6 some examples for XMLmessages can be found. Advantages of the XMLmessage format over a binary format are:? Readable by humans and editable with every simple text editor? Wellstructured by schemas or DTDs? Meta rules for schemas and DTDs are better documented than selfdefined rules for binary data.? Easy to change and to extend? For syntactic parsing existing standard modules can be used. For example the ApacheXerces parser [11].? Layer 6 (presentation layer) services are already included. Conversions, . ‘big endian’, ‘little endian’, are done automatically. Disadvantages are:? Increased amount of data on the network? The parser generally is larger.? Parsing may need more time.? Binary data, . pictures, can not be included directly..