Algorithmic Aspects of Perception-Aware Motion Planning on Resource-Constrained Platforms

Autonomous micro aerial vehicles (MAVs) are becoming an integral tool in numerous applications involving time-critical missions in GPS-denied environments. Due to their small size and lean energy budget, MAVs are often equipped with a camera to aid ego-localization. This introduces at least two fundamental challenges. First, cameras are of little use for state estimation if there is an insufficient quantity of visual information in the environment of the robot. Second, MAVs only display a limited amount of onboard computational resources. Should extracting motion estimates require excessive computational effort, in order to prevent fatal crashes, these agents would be confined to such low speeds that their deployment would be of questionable value. This thesis studies algorithmic aspects of the question: “How quickly can a vision-driven MAV traverse a given path, while maintaining accurate state estimates at all times?” We seek tractable families of problems involving designing a time-optimal open-loop sequence of controls for a MAV subject to both actuation as well as perception constraints that allow the robot leverage its onboard camera for accurate state estimation. Prior work has either focused on asymptotically optimal search-based approaches which are challenging to implement in real time, or fast local-optimization-based methods with no guarantees on global constraint satisfaction, stability, or optimality. We present three contributions. First, we extend optimality guarantees of a robust, computationally efficient algorithm for the time-optimal path parametrization problem. Second, we demonstrate the convexity of a general family of perception constraints which require a quadrotor to maintain a sufficient amount of information within field of view of its forward-facing onboard camera. Third, we devise computationally efficient algorithms for guiding the visual attention of a fully-actuated multirotor to traverse a path in minimum time while keeping the computational burden of extracting incremental motion estimates below a set threshold. Together, these contributions serve as stepping stones towards allowing MAVs execute missions autonomously at operational speeds.Ph.D