Learning About Software Errors Via Systematic Experimentation

Classical planners assume that their internal model is both correct and complete. The dynamic nature of realworld domains (e.g., multi-user software environments) makes these assumptions untenable. Several new planners (e.g.,XII [2]) have been designed to work with incomplete information, and strides have been made in planning with potentially incorrect information. But, efficient operation in the presence of incorrect information is highly dependent on a planner’s ability to detect errors. Failing to recognize errors can result in unexpected and potentially destructive effects, as well as further corruption of the world model. This abstract describes ED (the Error Detective) which automatically generates error detection functions for a software robot (softbot). In addition to error detection, the functions generated by ED accurately diugnose the cause of the errors. The automatic generation of these functions is important due to the large number of conditions that can affect the success of command execution. In addition, the ability to diagnose the cause of an error can greatly reduce the number of preconditions which need to be checked,/resatisficd prior to a successful execution of the operator. In tackling this problem we utilize three key insights: