Autonomous navigation through small, confined spaces

GreenSight Agronomics (GSAg) is proposing to develop a man-portable, fully autonomous aerial mapping system to satisfy the goal of the solicitation: quickly and completely mapping a confined space with no human guidance. The vehicle will have a 400 mm frame size to provide an ideal balance between maneuverability, flight time, and payload capacity. We have chosen the Pixhawk platform running ArduPilot software as the autopilot, as it has proven to be a robust flight controller for vehicles of various size and application. The UAV will have a hot-swappable lithium-ion battery system for rapid deployment with uninterrupted power during mid-mission battery changes. The proposed vehicle will have multiple Intel RealSense RGB-D camera systems, arranged around the outside of the airframe body to give 360 degree visibility in the horizontal and vertical direction. These cameras will interface to an extremely powerful and compact graphics-based processing unit, the Nvidia Jetson TX2. GSAg is proposing a novel method of SLAM, called ConvNet-SLAM, which uses a convolutional neural network to enhance the feature matching step, giving a much more dense point cloud. This point cloud will then be used by the trajectory generation algorithm to generate collision free paths to explore the environment.