A comparison of methods for transforming sensor conditional maps in occupancy grid mapping

Mobile Robotics has seen growth in leaps and bounds in the past couple of decades. There have been numerous problems specific to robotics that have either been completely solved or ingeniously circumvented. Yet, there are two fundamental challenges to Robotics that make this field a very interesting research arena - (i). Mapping and (ii). Localization.

Limitations on a robot's on board computational abilities and inaccuracies in sensor hardware and motor effectors further make the problem hard to solve. Such deficiencies with the hardware often have been compensated for through the development of probabilistic methods. These approaches have enabled us to perform satisfactory mapping with relatively inexpensive sensors, such as SONAR.

This research is a comparison of various techniques used in transforming sensor conditional maps in occupancy grid mapping - the most widely used map representation in recent times. The role of raster rotation while applying conditional probability and consequent error is analyzed and discussed. Alternate algorithms, which bypass raster rotation, are explored. With the help of different algorithms and techniques, the research also highlights the significance of rotation errors and the possible reduction in space and time requirement when the error is minimized or totally avoided. The research explores other sensor data evaluation heuristics and map transformations and compares them to existing methods.