Action Selection for Single-Camera SLAM

Abstract

Abstract-A method for evaluating, at video rate, the quality of actions for a single camera while mapping unknown indoor environments is presented. The strategy maximizes mutual information between measurements and states to help the camera avoid making ill-conditioned measurements that are appropriate to lack of depth in monocular vision systems. Our system prompts a user with the appropriate motion commands during 6-DOF visual simultaneous localization and mapping with a handheld camera. Additionally, the system has been ported to a mobile robotic platform, thus closing the control-estimation loop. To show the viability of the approach, simulations and experiments are presented for the unconstrained motion of a handheld camera and for the motion of a mobile robot with nonholonomic constraints. When combined with a path planner, the technique safely drives to a marked goal while, at the same time, producing an optimal estimated map.Index Terms-Action selection, active vision, bearing-only simultaneous localization and mapping (SLAM), mutual information, path planning.

Tree-connectivity: Evaluating the graphical structure of SLAM

“…It is easy to show that the log det of CRLB is a lower bound for the log det of the covariance matrix of any unbiased estimator. The D-criterion, due to the reasons mentioned above, is a popular choice and has been frequently used as the design criterion in many design problems in robotics, including sensor selection [14] and active SLAM [15]. In this section we define a connectivity metric based on the number of spanning trees, and reveal its impact on the D-criterion.…”

Appearance-based Active, Monocular, Dense Reconstruction for Micro Aerial Vehicles

“…The goal was to select a fixation-point of a moving stereo head attached to a mobile robot in order to minimize the motion drift along a predefined trajectory. Vidal Calleja et al [28] demonstrated an active feature-based visual SLAM framework that provides realtime user-feedback to minimize both map and camera pose uncertainty. Bryson and Sukkarieh [6] demonstrated a similar visual and inertial EKF-SLAM formulation for active control of flying vehicles.…”

“…Extensive simulation results were provided of a MAV that is restricted to fly on a plane. Similar to [28,6] the exploration in our algorithm is driven by a set of states (i.e., dense depth estimates in the reference view) that are initialized with high uncertainty at the start of the exploration. Our resulting map is spatially smaller but denser and exhibits more detail, which is crucial e.g., for path planning in cluttered environments.…”

“…Our resulting map is spatially smaller but denser and exhibits more detail, which is crucial e.g., for path planning in cluttered environments. Furthermore, in [28,6] the image is only used to extract features and subsequently neglected. On the other hand, our proposed approach is direct [14]-the intensity values in the image are directly used to reason about the next best view.…”

Potential information fields for mobile robot exploration

“…Bourgault et al [1] alternatively proposed a utility function for exploration that trades off the potential reduction of vehicle localization uncertainty, measured as entropy over a feature-based map, and the information gained over an occupancy grid. In contrast to these approaches, which consider independently the reduction of vehicle and map entropies, Vidal-Calleja et al, [20] tackled the issue of joint robot and map entropy reduction, taking into account robot and map cross correlations for the Visual SLAM EKF case.…”