April 2017: Constructing informative Bayesian map priors: A multi-objective optimisation approach applied to indoor occupancy grid mapping
Summary: Simultaneous localisation and mapping (SLAM) is a key challenge for a number of robotics applications including search and rescue and driverless cars. The aim of SLAM is to generate a map of an unknown environment that a robot needs to navigate around whilst simultaneously estimating the location of the robot in the environment. However, there is a tendency for the estimates of the map and location to grow in uncertainty. The main research effort in SLAM is to reduce this uncertainty to ensure that both the map and location estimates are as accurate as possible.
Despite significant research in SLAM and theoretically rigorous approaches that demonstrate excellent performance in simulations, experimental results typically demonstrate poor performance in real applications. In this paper a new approach to SLAM that uses prior information about the environment is used in order to improve performance in real applications. In particular this paper focuses on robots navigating in indoor environments, for example in a search and rescue mission. The prior information used is an architectural diagram of the building. The paper describes how to convert the architects diagrams into a useable form for SLAM and then demonstrates significant performance improvements for both simulations and real robot experiments.