Genetically Optimized ANFIS-based PID Controller Design for Posture-Stabilization of Self-Balancing-Robots under Depleting Battery Conditions

Abstract

It is well known that in battery-powered control applications, the continuous drop in battery’s output-power progressively degrades the dynamic performance of the system. The battery depletion phenomenon deteriorates the reliability of correctional effort if the controller gains are not adaptively adjusted as function of the available power level. Hence, this paper presents an adaptive neuro-fuzzy inference system (ANFIS) that dynamically adjusts the controller gains of a close-loop dynamic system as function of battery power-level in order to maintain desired performance while the battery is depleting. The proposed methodology is verified on an inherently unstable two-wheeled self-balancing-robot. The Proporional-Integral-Derivative (PID) controller is used for robot’s posture-stabilization. ?nitially, trivial sets of PID gains are selected via genetic algorithm to yield best control effort at various battery power-levels, using hardware-in-the-loop strategy. The acquired data is then used to train a power-level dependedent ANFIS that dynamically adjusts the PID gains in real-time. The performance of a fixed gain PID controller is compared with that of the proposed self-tuning PID controller for two different power-depletion scenarios that emulate real-world situations. The corresponding experimental results validate the robustness of the proposed control scheme to maintain the robot’s postural stability under discharging battery condition.