| This thesis addresses the problem of parallel parking a car equipped with multiple sensors. The car is an autonomous robotic vehicle subject to nonholonomic constraints. A kinematic simulation of the car is used to investigate control and sensor regimes in multiple parking maneuvers. An observer contains a car model with 25% steering-system and drive-train acceleration errors. The car model is available to the observer at every simulator time interval.; A short parking space requires multiple reversals of direction. Several s-shaped curves are compared using the rate at which the car approaches the curb as the objective function. A fifth-order parking polynomial is found to dominate the other curves in curvature-constrained docking maneuvers.; An odometer, a steering-wheel angle meter, an orientation sensor and an absolute position sensor enable correction of car-model error. The sensors' reliability is inversely related to their availability. The 20 sample-per-second internal sensors--the odometer and the steering-wheel angle meter--give the observer the ability to perform dead-reckoning with 10% measurement error. The external sensors--the orientation meter and the absolute position sensor--are error-free but can be sampled only 5 times per second.; Both ad-hoc logic and nonholonomic constraints are used to perform sensor fusion. The combination of external and internal sensors outperforms either. Each by itself is, however, superior to open-loop control using only the observer's model.; Bang-bang control is used to change the steering-wheel angle and the drive-train. Control of these variables allows the car to attain any feasible configuration. Accelerations are piecewise constant, and velocity is unconstrained. Proportional plus derivative control is used to formulate the bang-bang control. The coefficients of the control equations are found by experiment. Although the trajectory of the rear wheel is prescribed by the parking polynomial, only the front wheel is subject to direct control.; All simulations are performed with STELLA{dollar}sp{lcub}rm TM{rcub}{dollar} (Structured Thinking Experimental Learning Laboratory with Animation). STELLA allows interactive set up and simulation of nonlinear, coupled, initial boundary value problems. Some symbolic manipulations are performed with MAPLE{dollar}sp{lcub}rm TM{rcub}{dollar}. |
