Agre compares two different methodologies to the planning component for autonomous mobile robotics in “What are plans for?”. The first methodology for creating a planning component is the “plan-as-program” view. This approach takes a plan and uses it similar to a program execution. In this context, executing a plan means walking over it in a mechanical fashion, carrying out the primitive action and monitoring conditions specified by the planner and performing little or no new reasoning about the activity in which the agent is engaged. This has been the typical approach used in most of the architectures for the planner component. The main issue with this methodology is that it is not very flexible and this poses a problem because an execution of a plan in the real world environment brings risks because unpredictable external processes can change the world and affect the robot. The plan-as-program view understands activity as a matter of problem solving and control. The world presents an agent with a series of formally defined problems that require solutions. A planner produces solutions to these problems. The plan plays a central roles in determining activity. The executive implements these solutions by trying to make the world conform to them. Agre tries to state the case that the alternative to these issues is the plan-as-communication view.

The alternative view proposed is the plan-as-communication view. The basic idea is that instead of manually executing instructions without any sense of what the actions would entail, this approach would allow the planner to improvise. Plans do not directly determine their user’s activity as with the plan-as-program view. The agent uses the plan as a resource not as a set of instructions for interacting with the environment. This leads the robot to participate in the world instead of trying to control it. In reviewing the article, I felt that the technical explanation for how the plan-as-communication was too abstract without delving into specific details. The plan-as-communication view is an interesting concept but it would nice to have had some elaboration on how one would implement such a methodology.

Reference:

  1. Agre, Philip E. “What Are Plans For?” <http://www.textfiles.com/bitsavers/pdf/mit/ai/aim/AIM-1050A.pdf>