Micro aerial vehicles (MAVs) are agile and have unstable flight dynamics. They require a failsafe method to be navigated through areas even without GPS coverage. The approach of this work is to use only the feature matches between two consecutive images, i.e. optical flow (OF) and inertial cues. The vehicle’s pose and additional intrinsic as well as extrinsic states are estimated continuously to navigate and control the MAV through the area. Optical flow cues and inertial measurement readings are fused in an EKF (extended Kalman filter) framework to estimate a metric 3D body velocity, terrain plane-parameters, terrain plane relative 3D attitude including heading, and metric distance between the camera on the MAV and this plane. The estimates of the EKF provide the vehicle controller with accurate information about the vehicle and the environment in order to navigate the micro-helicopter autonomously through large areas. The system is fully self-contained and all computation is done onboard the MAV in real-time. This eliminates the need of a data link to a ground station and allows standalone operation.
The novelty of the technology is that the full information provided by OF and inertial readings is used to achieve a complete vehicle state estimation. Earlier work of the author on IOF (inertial-optical flow) state estimation for MAVs included estimates of the attitude with respect to a gravity-aligned navigation frame and a metric velocity including inertial biases, visual scale, and transformation between the camera and IMU (inertial measurement unit), turning the platform into a so-called self-calibrating power-on-and-go system. This extended approach additionally allows estimation of the inclination of the terrain towards gravity, the MAV’s heading with respect to this plane, and its metric distance to this plane. The heading and terrain-plane distance are two additional degrees of freedom which are crucial to keep the MAV on a specific course and away from the ground for terrain following. The approach is inherently failsafe since all these states are estimated by only using a minimum of three feature matches in two consecutive camera frames. Neither feature history nor (local) map — which could get corrupted over time — is required. In fact, tests show that the approach is robust enough such that the MAV can be deployed by simply throwing it into the air stabilizing itself quickly. This extends the platform from a power-on-and-go system to a true throw-and-go MAV.
For reconnaissance, exploration, and search and rescue, small-scale helicopters may be the most viable solution because of their hovering and navigational advantages. Small-scale helicopters can even be deployed in cluttered and narrow environments indoors and outdoors with a minimal risk to people or the environment. The light weight of MAVs makes them ideal to carry to and deploy in various different scenarios.