State Estimation, Control, and Planning for a Quadrotor Team

摘要

Teams of aerial robots have the potential to become effective information gathering instruments, specially in the fields of infrastructure inspection, surveillance, and environmental monitoring due to their ability to operate in environments which may be difficult or impassable for ground robots. This dissertation addresses the problem of developing a team of autonomous quadrotors that can be quickly deployed and controlled by a single human operator.\nFirst, we describe the system design and algorithms for robust single robot autonomy and demonstrate it with experiments in a variety of environments. The developed robot uses onboard cameras for state estimation and is capable of autonomous navigation through natural and man-made environments at high speeds.\nNext, we develop the system architecture and components to allow a single operator to deploy, control, and monitor a team of robots.\nIn order to focus on the multi-robot coordination problem, the robots used for this part utilized GPS for state estimation and required known maps of the environment for obstacle avoidance, which limited the usage of the system to known outdoor environments.\nFinally, we combine the robots from the first part with this multi-robot framework. The main challenge here is that due to using onboard sensing for state estimation, each robot has a different reference frame for its state estimates, typically with the origin at the starting pose of the robot while the high-level planner for the team requires all the robot state estimates to be in a common reference frame.\nWe propose a method that uses the relative position/bearing measurements of nearby robots detected using the onboard camera to solve this problem. The complication in this method is that these relative measurements do not contain any identity information since all the robots in the team are equivalent and look the same. In order to fuse these unlabeled measurements and estimate the states of the robots in a common reference frame, we used the Coupled Probabilistic Data Association Filter (CPDAF) and developed an approximation to reduce its computational complexity enabling online estimation with a team of up to 15 robots.