Run-time autotuning of a robot controller using a genetics based machine learning control scheme

A genetics based machine learning (GBML) method is proposed and analyzed for learning and enhancing the control of a microrobot with stepping motor drives. This approach tries to combine several advantages of fuzzy logic and genetics based machine learning using slightly modified classifier systems. The paper discusses the learning capabilities of the proposed control system. The PID gains of a conventional controller were tuned at run-time in order to minimize the effect of the nonlinear disturbances (nonlinear variable load torque applied to the controlled plant). The tuning is based on a predictive estimation method of the controller's gains, performed by a GA driven fuzzy classifier system, which has to evolve an adequate rule set to tune properly the controller's gains.