Tactics Planning

Requirements Summary

The Tactics Planner (TP) rapidly became the alter ego of the Intent Inference portion of the PVI.  Where the Intent Inference subsystem was posting plan and goal instances based on inferred pilot intent, the TP was creating proposed courses of action based upon knowledge of the current situation and how the standard tactical knowledge (including the tactics-of-the-day briefed on the MST before take-off) suggested that the pilot react to circumstances detected by SA and SS.

 It was responsible for:

• Post monitors with  SA to detect any new situation requiring a tactical response,
• develop a tactical response in the form of a proposal and present it to the PVI,
• when a plan is approved by the pilot via the PVI, establish monitors for starting execution, and for ensuring its continued feasibility,
• when a plan begins execution, add the monitors to determine its completion or failure, and
• when failures or completions are detected, clean up the nodes and their children.  

Phase 1 Implementation

The TP was implemented in Phase 1 as a Skeletal Planner using a proprietary tool from Teknowledge called KADET.  The underlying approach that enabled the TP was the concept of small collections of local knowledge around each PGG node expressed as rule sets which, because of their small size, could be run by brute force when necessary, rather than via some enormous inference engine such as would be needed with large, monolithic rule bases.

Phase 2 Implementation

In Phase 2, the TP was recoded in C++.  One major change for the purpose of efficiency was a redesign of the PGG that in Phase 1 had been very straggly in character with some arbitrarily long branches.  In Phase 2, the PGG was limited in depth to 8 layers, 4 goal layers and 4 plan layers, the lowest of which were actually action scripts.

last updated 10/5/2002 by David Smith