Optical and acoustic range sensing for underwater robotics

The paper presents a sensor fusion architecture developed for underwater robotics, using a probabilistic octree mechanism to represent the volume of the workspace. In order for sensor data to be entered into this system probabilistic error models for each sensor must be developed. Two error models have been implemented for a pencil beam sonar-a rigorous but computationally expensive parametric probabilistic model, and a simple heuristic device. The former gives better quality results, but the latter is ten times faster in execution. Surfaces extracted from the octree at varying probability thresholds are presented, showing the utility of the approach, as data of a specified reliability may be extracted to facilitate a particular task. A laser triangulation device has also been developed, but has not yet been integrated into the octree mechanism. Preliminary results from the laser are presented which confirm that it provides a complementary source of volumetric data. The laser is more accurate and the data capture rate is several orders of magnitude faster than the sonar, but the performance of optical devices is degraded by backscatter, and marine growth. Thus the use of the probabilistic approach to combine these two complementary sensors should considerably improve the reliability of the resulting sensor system