Modeling Intelligent Embedded Real-Time Systems using High-Level Petri Nets

Transcription

1 Modeling Intelligent Embedded Real-Time Systems using High-Level Petri Nets Carsten Rust and Bernd Kleinjohann C-LAB 1,Fürstenallee 11, Paderborn, Germany WWW: {car, Abstract In this paper, an approach for the design of intelligent embedded real-time systems is presented. The approach is based on a design methodology for conventional embedded systems, which was developed at our institute during the last years. The paper first gives an overview over this methodology and afterwards describes in particular, which possibilities for the modeling of intelligent systems are provided by our approach. Furthermore an application scenario with the C-LAB Pathfinder, a small robot with partly autonomous behavior, is described. 1 Introduction Intelligent embedded systems gain increasing importance. A wide range of applications emerges in the domain of elecronic toys and entertainment robots. Examples are Furby, My Real Baby ( and AIBO ( Even in traditional application domains, for instance in automotive systems, novel applications like driver assistance systems and autonomous driving will have to actively sense their environment and feature autonomous behavior. For the future, it may be expected that integrated design environments will be supplied for the design of such intelligent systems. Similar to existing frameworks for hardware design and for conventional embedded real-time system design, these environments should provide tool support for the entire design process from the specification to the implementation of intelligent embedded system. With respect to specification, these environments have to support some typical features of intelligent systems, namely complex algorithms for communication, for processing of perceptions, for orientation, for planning, and for the programming of behavior and learning components apart from the usual paradigms for control and regulation. Our approach, which was presented for the first time in [1], is to extend and modify an existing environment for conventional embedded real-time systems in a way that it is open for the handling of intelligent systems. The 1 C-LAB is a cooperation of University Paderborn and Siemens 1

2 methodology divides the design into the three stages specification and modeling, analysis and partitioning, and synthesis. As underlying formal model we use extended Pr/T Nets, a high-level form of Petri Nets (cf. section 3). The design is supported by the SEA environment ( which provides appropriate tools for each step in the design flow. For a more detailed description of the design flow and the accompanying tool support we refer to [2]. This paper is focussed on modeling and specification. We describe, how our modeling approach, which is well-proven for the design of conventional embedded real-time systems, may be applied to the behaviorbased specification of intelligent embedded real-time systems. 2 Application Example In order to evaluate our concepts for the design of intelligent embedded realtime systems, we developed an experimental application platform, the C-LAB Pathfinder. The pathfinder is a small robot (Figure 1 a) which may either be controlled by humans via the internet or by control algorithms. The latter run on microcontrollers, which are installed on board of the pathfinder. The microcontrollers as well as the actors and sensors installed on the pathfinder are depicted in Figure 1 b). In part c) of the figure, a small application scenario is depicted. In the scenario, several pathfinders are to drive a predefined Figure 1: The C-LAB Pathfinder course of an intersection (thick lines in the figure). The course is indicated by landmarks. Each pathfinder should follow these landmarks using the reflex light barriers installed at its underbody. The landmarks are divided into different segments. At the end of a segment a pathfinder has to find the beginning of the next line segment with the aid of its odometry. Suchlike problems are typical for applications of the pathfinder, e.g. for driving to a loading station or for crossing a prohibited area. 2

3 3 Extended Pr/T Nets Our underlying formal model is an extension of the standard Pr/T Net model introduced by Genrich and Lautenbach in [3]. Pr/T Nets are a high-level form of Petri Nets. Like Petri Nets, Pr/T Nets consist of transitions and places, that are connected by edges. A small example is depicted in Figure 2. The example net contains three places (b in, v in, v out ) and two transitions b in [false] v in [-3,5] on (flag=true) off (flag=false) nx=0; ny=0; [nx,ny] v out b in v in on (flag=true) off (flag=false) nx=0; ny=0; v out [0,0] [nx,ny] Figure 2: A Pr/T Net (on, off). Dependent on a flag, which is stored in the place b in, the net either hands over a vector from the place v in to the place v out by means of transition on or transition off deletes this vector and stores a 0-Vektor in the place v out. As can be seen in the example, transitions are the active elements of Pr/T Nets. They have concession to fire, when appropriate tokens are available on their input places. Furthermore, in case a condition is defined for the transition, the condition must be fulfilled by the values assigned to the involved tokens. During the firing process transitions may evaluate assignments to variables occuring at their outgoing edges. When modeling complex systems hierarchical specifications are needed. An example for a hierarchical Pr/T Net is (partly) depicted in Figure 3. The structured node BScale instanciates the net depicted in Figure 2. Before instanciation, the places b in, v in and v out have been mapped to ports and a graphical representation for the subnet was defined, which is visible in the instanciating node. During simulation of the model, the graphical Figure 3: An extended Pr/T Net representation is animated in order to visualize the state of the associated subnet. 3

4 4 Modeling of Robot Behaviors The approach presented in the following is based on the concepts developed by Arkin [4] and Balch [5]. We adapted their concepts for the specification of robot behavior to our modeling paradigm. Following our approach, the specification of a robot s control is based on the definition of several basic behaviors. The basic behaviors each assign a set of sensor values directly to corresponding values for the actuators. The description of a basic behavior is based on vector fields. For its specification a library is available. Some elements of the libary are visible in Figure 3 which depicts a part of the basic behavior FollowLine. The two elements on the left (OnEvent, OffEvent) yield information about changes in status of one of the reflex light barriers. The element StateLB combines this information into a flag, which represents the status of the light barrier. The output of StateLB is used as input for the element BScale, which was described above. Thus, in case the respective reflex light barrier detects a label, BScale contributes a vector moving the robot along this label to the total behavior. The overall Figure 4: Behavior-based modeling of Pathfinder Control purpose of FollowLine is to move the robot along a landmark. Further basic behaviors are for example AvoidObstacle for the avoidance of obstacles, or MoveToGoal, which leads the robot to a target given in absolute coordinates. The vector fields for two basic behaviors are depicted in Figure 4. For each point in the 2-dimensional space, they define how the robot should move according to the respective behavior. A single vector indicates the direction of the movement and (via the vector s length) the speed. Having defined a set of basic behaviors, a robot control may be specified by combining the basic behaviors to a complex one. In the simplest case, the results of the single behaviors are accumulated using a standard element from our library, for instance a weighted sum of behaviors or an element selecting the currently strongest behavior. This simple architecture is depicted in the lower part of Figure 4 a). Just as for the accumulation of basic behaviors, 4

5 library elements also exist for the conversion of a vector resulting from a complex behavior into actuator values. The simple architecture is well-suited for the specification of simple behaviors like Follow the landmark and thereby avoid obstacles. However, in most control applications, an additional discrete control is necessary which switches between several behaviors. This extension to the simple architecture is depicted in the upper part of Figure 4 a). The discrete control is realized by an automaton. It is notified about important changes in the robot s environment by so called trigger behaviors. A trigger behavior may for instance raise an event, when the robot has reached a certain position, that it has moved towards. Having been triggered, the automaton performs one step leading to a new state. According to the state, the parameterization of the robot s behavior is changed. In the above mentioned example, when the robot has reached a certain position, it may for instance be reasonable to set a new target position. However, it is also possible to let the robot continue with a completely different behavior after the state change. 5 Conclusion In this paper we have presented our approach for behavior-based modeling of intelligent embedded real-time systems using high-level Petri Nets. The approach is part of our ongoing work to open up an existing methodology for the design of embedded real-time systems for intelligent embedded systems. In the future we want to integrate further features of intelligent systems into our modeling approach, e.g. learning algorithms. References [1] B. Kleinjohann, C. Rust, and J. Tacken. Entwurf von autonomen Systemen mit High Level Petrinetzen. In 16. Fachgespräch Autonome Mobile Systeme (AMS 2000), Karlsruhe, Germany, September [2] C. Rust, J. Tacken, and C. Böke. Pr/T Net based Seamless Design of Embedded Real-Time Systems. To appear in Proceedings of ICATPN 2001, [3] H. J. Genrich. Predicate/Transition Nets. In Advances in Petri Nets 1986, volume 254. Springer Verlag, Part I. [4] R. Arkin. Behaviour-Based Robotics. MIT Press, [5] T. Balch. Behavioral Diversity in Learning Robot Teams. PhD thesis, Georgia Institute of Technologie,