Highly Accurate SLAM for Rotary Wing UAVs

A new solution for the SLAM problem is presented which makes use of a scan matching algorithm, and does not require the platform's dynamic model. The virtual map is represented as an occupancy grid, which is scanned by means of ray casting to get a scan of the virtual world, called "virtual scan". The virtual scan therefore contains data from all previously acquired laser measurements, and hence serves as a very good representation of the surroundings. New laser scans are scan-matched against the virtual scan to get an estimate of the new position, which allows the map to be updated with the new information. The scan matching cost function is minimized by both an adaptive direct search, and a gradient based minimization method, until convergence is achieved. Experimental validation of the SLAM method is presented by mapping a typical office hallway environment, using a single rotorMAV, carrying a laser range scanner. The mapping results are highly accurate, within 0.1% of the traveled course. The resulting method is model-free and can be implemented on a variety of autonomous agents. Moreover, It is also computationally lightweight, providing real time capability on small form factor computers.