Jukka Riekki & Yasuo Kuniyoshi: Architecture for Vision-based Purposive Behaviors 6 Conclusion 3.3 Sensors 4 Experiments

IROS 95 8/8 parallelism is simulated, the scheduler is the key to these properties. The experiments showed that markers improve a behavior-based system. The most important improvement is the set of new coordination methods. In addition to coordination capability, markers simplify the rest of the control system. First, markers simplify behaviors, as behaviors need not to update task related data themselves. Markers also focus the attention of the behaviors and thus constrain the situations in which the behaviors operate. Secondly, markers decrease the number of connections between the control system modules. A behavior needs not to be connected to all modules updating the data it uses. Instead, it suffices to connect the behavior to the corresponding marker. Further, an abstract behavior needs not to be connected to all modules it sends commands to, only to the corresponding marker, which distributes the command. Markers also increase the modularity of the control system. The behaviors are grouped into sets performing tasks. A further advantage is that conditions for tasks are easy to express using markers and their spatial relations. Finally, all these improvements were achieved without sacrificing the real time performance of a behavior-based system. Even though the initial experiments were encouraging, the PURE control system has to be improved. We have to integrate the components based on the optical flow in the system. We will also implement a better arbitration method. It is clear, that predefined priorities are not enough for arbitrating behaviors. A more powerful, dynamic method is needed. We are currently improving the PURE system based on the ideas presented in Chapter 3. The new scheduler will take different response times into account and guarantee delays. In the new version the mobile space markers are interpreted as directions to perceive the task related environment features. In the future we plan to build more complex systems. At the next stage we will add in the system behaviors initializing transfer tasks when there are no other robots to be helped. These behaviors will recognize a global goal where all small objects should be transferred. A single autonomous agent is not our final goal. We want to study multi robot coordination with several such agents. We will soon have two robots controlled by the PURE system. This setup allows some interesting experiments. We will test among other tasks cooperative transfer tasks and marching in formation to new task areas. In this paper …