Automatic linear robot control synthesis using genetic programming

Abstract

An automatic controller synthesis method for a single axe linear robot is considered. The robot motions are modeled by a Delay Time Petri Net (DTPN). The search refers to automatically finding a controller modeled by a Time Petri Net (TPN) that fulfills some specified requirements. The controller model is synthesized using a Genetic Programming (GP) method. The mapping between TPN model and the tree representation of individual genotypes is performed using a formal language named here TPNL (Time Petri Net based Language). This language is suited for formal description of the controller behavior traits like sequential, concurrent, selection, loop or input/output. The use of control traits guaranties the construction of individuals that are capable and useful to control the robot moves. To diminish the search durations, besides the usual genetic operators like mutation, permutation and crossover, a new atrophy operator was introduced.