Simultanteous Localization And Mapping

     

The idea of Simultanteous Localization And Mapping (SLAM) was introduced by an early 1980s ASCI-based computer game called Rogue. In this game, you controlled the motion of a character through a randomly generated "dungeon", searching for treasure while avoiding monsters. A map of everything your character had seen thus far was generated as the game evolved, so that you could more completely explore its passageways and, once you had retreived the principle treasure contained in the dungeon, easily retrace your steps back out. Shortly after the game of Rogue was introduced, an Artificial Intelligence (AI) program called Rogomatic was developed to play the game of Rogue autonomously; Rogomatic was thus the genesis of modern attempts at SLAM, which is essentially the same problem, with the character replaced by a real robot, exploring a real dungeon, in search of real treasure, while avoiding real monsters.

The current state of the art in SLAM is illustrated in the center of the figure above. Besides being implemented on an actual vehicle, exploring a physical environment, and the result being depicted in a crude 3D representation instead of a 2D map, the quality of the information returned has not evolved all that much in the 25 years since Rogomatic was developed. Based on the emergent robotic, vision, and communication technologies now becoming available, the time is ripe to take the next major step.

The 3D virtual environment visualized in a modern first-person-shooter video game is illustrated at right in the figure above. For those who haven't played such games, the information presented in each frame is very natural, as if you were exploring the environment in person: you can see what is on the walls, peek around corners, and decide quickly what is interesting to explore further, and what is not.

We believe that the next major advance in mobile robotics will be the effective use small agile vehicles, such as those developed in our lab, leveraging advanced imagers and laser rangefinders to create an effective robotic exploration system that can develop a 3D virtual environment summarizing everything the vehicles have thus far encountered. Note in particular that the 2D photo stitching problem is essentially already solved, with effective commercial software readily available; the next natural step in image processing is 3D photo stitching, in the context described here, in order to "sew" together a 3D virtual model of the physical environment explored by a robotic system. Navigating such a virtual environment then allows a warfighter, firefighter, or mine rescue team to have an excellent view of what they are going to find before moving into a hazardous or inaccessible area.