Distributed Simulation of Robot Tasks

Software environments which simulate the behavior of complex systems often lead to a computational expense that reaches the limit of hardware resources. The \conventional" approach partitions the global system program into a set of successively called sub-programs, where the programmer is responsible for scheduling and synchronization of all subprograms. Increasing the complexity of the global program (i.e., simulating a more complex application) reduces overall performance, because more subprograms have to be handled within the same time period. In many simulation environments | also in the area of robot simulation | some of the subprograms are independent from others, allowing them to be executed in parallel. To achieve this goal, each has to be represented as one operating system process (a task). All tasks may be distributed within a network of processors , so (1) \true" parallelism is achieved and (2) the simulation system performance is (nearly) independent of the system structure complexity. This approach could also support a heterogeneous network, where some of the tasks are instantiated on specialized nodes (e.g., transputers, graphic-workstations, etc.). In this paper, the integrated DiSCUS environment (Distributed Simulation and Control in a Unix System) is presented, which is used for distributed simulation of assembly cells.