Modern MAP inference methods for accurate and fast occupancy grid mapping on higher order factor graphs

Abstract

Using the inverse sensor model has been popular in occupancy grid mapping. However, it is widely known that applying the inverse sensor model to mapping requires certain assumptions that are not necessarily true. Even the works that use forward sensor models have relied on methods like expectation maximization or Gibbs sampling which have been succeeded by more effective methods of maximum a posteriori (MAP) inference over graphical models. In this paper, we propose the use of modern MAP inference methods along with the forward sensor model. Our implementation and experimental results demonstrate that these modern inference methods deliver more accurate maps more efficiently than previously used methods. I. INTRODUCTION Mobile robot problems like navigation, path planning, localization and collision avoidance require an estimate of the robot's spatial environment; this underlying problem is called robot mapping (2). Even in environments in which maps are available, the environment may change over time necessitating a mapping ability on the mobile robot. Robot mapping hence remains an active field of research (3)- (5) as it is an important problem in application areas like indoor autonomous navigation, grasping, reconstruction and augmented reality. Although robot mapping can be performed in many ways—metric or topological; with range sensors, like sonar (6), laser scanners (6) and RGBD (7), or bearing-only sensors (8), (9)—metric mapping with range sensors is the most common. Bearing-only sensors provide estimates up to scale; topological maps still require local metric estimates for certain problems like navigation. We hence focus on metric mapping with range sensors, specifically, laser scanners. Occupancy grid mapping (OGM) is a popular and useful range-based mapping method (10), (11). It affords a simple implementation and avoids a need to explicitly seek and match landmarks in the environment (12), (13). In contrast, it discretizes the environment into cells, squares (2D) or cubes (3D), and associates a random variable with each cell