Planlite: Adaptive planning using weightless systems

Abstract

This paper proposes an adaptive planning system design based on the weightless net paradigm[1]. The system learns local state transitions through ’exploration’ of the application environment and then generalizes on the learnt data — enabling the system to produce optimal plans based on action cost. The generalization also allows data to be trained into the planner during the generalization process, allowing the planner to adapt relevant plans according to the new data. A simulation of the proposed system has been implemented and successfully applied to several problem domains. The amount of computation during operation of this system is less than that of a conventional rule-based planner.