OPTIMAL VISION-BASED POSITION ESTIMATION OF AN UNDERWATER SPACE SIMULATION ROBOT

Abstract

This thesis describes the development of the Vision Positioning System (VPS), a real-time 3-D inertial translational state estimation system for free-flying neutral buoyancy space simulation robots. Key contributions include a technique for the accurate calibration of long-baseline underwater vision systems, and a three degree of freedom Extended Kalman Filter (EKF) that merges camera measurements with robot telemetry to create an optimal estimate of 3-D translational position and velocity. Results from static and dynamic underwater positioning tests are presented that characterize the system accuracy. Static tests indicate VPS is capable of locating the robot with 3 to 4 cm accuracy, while dynamic test results show similar accuracy given ideal lighting conditions and flight in a region of complete camera coverage. The results indicate that with further development to correct for lighting and better reject erroneous camera measurements, VPS has the potential for accuracy comparable to that achieved by GPS navigation systems