Near-optimal energy fuzzy parking of mobile robots

In this paper, the trajectory-planning problem of a mobile robot is studied with an application to near optimal energy parking. A hybrid data-driven neuro-fuzzy system composed of two steps is developed; first, we introduce a preprocessing step involving offline trajectory parking, and generating reference optimal energy trajectories, while satisfying several constraints related to robot kinematics and dynamics and parking lot limitations. The discrete augmented Lagrangean is implemented to solve the resulting non-linear and non-convex optimal control problem. The outcomes of this pre-processing step allow building a neuro-fuzzy inference system to learn and capture the robot multi-objective dynamic behavior. The second step is a sensor-based neuro-fuzzy navigation scheme. From the learnt optimal energy behavior dataset, a 6-input/2-output ANFIS network is built for online parking. This network considers the three range measurements obtained from three sonar sensors mounted at 3 directions at the front left corner of the robot. In addition, the discrepancy between the current measured distance and the previous measured one, has been implemented to generate a control output consisting of the robot motor torques. First results based on real dimensions of a typical car, demonstrate the effectiveness of the proposed controller in practical car maneuvers.