Concurrent reactive plans: anticipating and forestalling execution failures

  • Authors:

  • Michael Beetz

  • Affiliations:

  • Universität Bonn, Institut für Informatik III, Bonn, Germany

  • Venue:
  • Concurrent reactive plans: anticipating and forestalling execution failures
  • Year:
  • 2000

Quantified Score

Hi-index 0.00

Visualization

Abstract

Autonomous robots that accomplish their jobs in partly unknown and changing environments often learn important information while carrying out their jobs. To be reliable and effcient, they have to act appropriately in novel situations and respond immediately to unpredicted events. They also have to reconsider their intended course of action when it is likely to have flaws. For example, whenever a robot detects another robot, it should predict that robot's effect on its plan and -- if necessary -- revise its plan to make it more robust. To accomplish these patterns of activity we equip robots with structured reactive plans (SRPs), concurrent control programs that can not only be interpreted but also reasoned about and manipulated. These plans specify how the robot is to respond to sensory input in order to accomplish its jobs. In this book we describe a computational model of forestalling common flaws in autonomous robot behavior. To this end, we develop a representation for SRPs in which declarative statements for goals, perceptions, and beliefs make the structure and purpose of SRPs explicit and thereby simplify and speed up reasoning about SRPs and their projections. We have also developed a notation for transforming SRPs, which does not only make the physical effects of plan execution explicit, but also the process of plan interpretation, as well as temporal, causal, and teleological relationships among plan interpretation, the world, and the physical behavior of the robot. Using this notation a planning system can diagnose and forestall common flaws in robot plans that cannot be dealt with in other planning representations. Finally, we have extended the language for writing SRPs with constructs that allow for a flexible integration of planning and execution and thereby turned it into a single high-level language that can handle both planning and execution actions. Experiments in a simulated world show that by simultaneously forestalling flaws and executing SRPs, the robot can perform its jobs more reliably than, and almost as effciently as, it could using fixed control programs.