| dc.description.abstract | Multirotors could be used to autonomously perform tasks in search-and-rescue, reconnaissance, or infrastructure-monitoring applications. In these environments, the vehicle may have limited or degraded GPS access. Researchers have investigated methods for simultaneous localization and mapping (SLAM) using on-board vision sensors, allowing vehicles to navigate in GPS-denied environments. In particular, SLAM solutions based on a monocular camera offer low-cost, low-weight, and accurate navigation indoors and outdoors without explicit range limitations. However, a monocular camera is a bearing-only sensor. Additional sensors are required to achieve metric pose estimation, and the structure of a scene can only be recovered through camera motion. Because of these challenges, the performance of monocular-based navigation solutions is typically very sensitive to the environment and the vehicle’s trajectory. This work proposes an integrated estimation and guidance approach for improving the robustness of monocular SLAM to environmental uncertainty. It is specifically intended for a multirotor carrying a monocular camera, downward-facing rangefinder, and inertial measurement unit (IMU). A guidance maneuver is proposed that takes advantage of the metric rangefinder measurements. When the environmental uncertainty is high, the vehicle simply moves up and down, initializing features with a confident and accurate baseline. In order to demonstrate this technique, a vision-aided navigation solution is implemented which includes a unique approach to feature covariance initialization that is based on consider least squares. Features are only initialized if there is enough information to accurately triangulate their position, providing an indirect metric of environmental uncertainty that could be used to signal the guidance maneuver. The navigation filter is validated using hardware and simulated data. Finally, simulations show that the proposed initialization maneuver is a simple, practical, and effective way to improve the robustness of monocular-vision-aided-navigation and could increase the amount of autonomy that GPS-denied multirotors are capable of achieving. | en |
