A Covariance-Based View Planning Algorithm for Aerial Robots With Decoupled Vision System

The precision and robustness of visual odometry tasks for aerial robots may be compromised in featureless environments. To achieve robust visual odometry for micro aerial vehicles, it is crucial to utilize the advantages offered by a decoupled sensor system (e.g., gimbal camera) by actively controlling it based on perceptual requirements. In this article, by formulating the covariance information, sensor visibility, and the degree of exploration-exploitation carefully, we are able to build a model to evaluate the visual odometry robustness, namely, covariance-based feature exploration-exploitation (CoFEE) model. Building upon the CoFEE model, the study presents an active perception planning algorithm that enables the system to actively navigate toward scenes resulting in satisfactory visual odometry precision. Implementing these solutions leads to improved visual odometry quality and robustness. Several simulations and experiments are carried out to validate the practicability of our approach. We will release our implementation as an open-source package for the community.