Bayesian state estimation and behavior selection for autonomous robotic exploration in dynamic environments

Abstract

In order to be truly autonomous, robots that operate in natural, populated environments must have the ability to create a model of these unpredictable dynamic environments and make use of this self-acquired uncertain knowledge to decide about their actions. A formal Bayesian framework is introduced, which enables recursive estimation of a dynamic environment model and action selection based on this estimate. Existing methods are combined to produce a working implementation of the proposed framework. A RaoBlackwellized particle filter (RBPF) is deployed to address the Simultaneous Localization And Mapping (SLAM) problem and combined with recursive conditional particle filters in order to track people in the vicinity of the robot. In this way, a complete model is provided, which is utilized for selecting the actions of the robot so that its uncertainty is kept under control and the likelihood of achieving its goals is increased. All developed algorithms have been applied to the domain of the Autonomous City Explorer robot and results from the implementation on the robotic platform are presented.

Learning search polices from humans in a partially observable context

“…However, very few have studied the effect of uncertainty in the decision process and do not consider it during the learning or assume that it is implicitly handled. A noticeable exception is [18], in which a human expert guides the exploration of a robot in an indoor environment. The high-level actions (explore, loop closure, reach goal) taken by the human are recorded along with three different features related to the uncertainty in the map.…”

The Autonomous City Explorer: Towards Natural Human-Robot Interaction in Urban Environments

“…In cluttered outdoor environments the grid-based approach provides more robust and accurate mapping. The chosen SLAM approach was previously presented in [18].…”

Histogram of Oriented Cameras - A New Descriptor for Visual SLAM in Dynamic Environments

“…Wang et al [30] proposed to maintain a stationary and a dynamic occupancy grid map, constructed out of laser data. Lidoris et al [20] combined a Rao-Blackwellized particle filter for robot pose estimation with a person tracker for moving object detection. Other authors try to filter out false measurements from 2D laser data before incorporating them into an occupancy grid map [6], or use map matching between the current scan and the already generated map for moving object detection [29].…”