This project aims to develop and demonstrate the benefits of using sensor-based methods for motion control of aerial vehicles in uncertain and dynamic environments. Recent studies support anecdotal evidence that a wide variety of animals adopt motion strategies that rely on a perceptual invariant. Building on this notion, we propose to develop a sensor-based control framework for performing tasks that involve close interaction with the environment and require a reactive and compliant behavior in response to unexpected changes. Both single vehicle and multi-vehicle mission scenarios in three-dimensional space will be considered.

To devise sensor-based control algorithms, a connection between the control objective and an invariant mapped in sensor space will be sought after. For example, in a follow the-leader problem, a possible vision-based invariant is given by the image coordinates (as perceived by the follower) of a collection of feature points marked on the leader. A sensor-based perspective can also be adopted to approach the dual problem of vehicle state estimation. In this case, the key concept lies in exploring the dynamics of sensor measurements and combining them with the knowledge of the system to devise dynamic algorithms that progressively improve the position and orientation estimates as more data is collected.

To implement sensor-based strategies, we propose to build an aerial sensing testbed based on micro quadrotors. Given their handling qualities, these vehicles are ideal for operation in confined environments and present some advantages over other rotorcraft. They are similar to helicopters in that they can describe extremely agile maneuvers in the low speed regimes, including hover, vertical take-off and landing. When compared with the main-rotor tail-rotor configuration, the four-rotor configuration is more stable and allows for a considerable downsize in rotor diameter, which greatly reduces the risks to the safety of operation in enclosed spaces. Special emphasis will be placed on designing the smallest possible platform, while ensuring that the payload requirements are met. Reducing the size and weight not only makes for a safer low-cost vehicle but also simplifies and expedites the development and test process.

The intended applications focus on indoor or urban environments, where GPS signals are unreliable or simply unavailable so that the control algorithms must rely on local sensor information. The basic quadrotor configuration will be equipped with a small low-power computer, an inertial measurement unit, and an ultrasonic range finder. Due to the payload constraints, either a camera or scanning laser range finder will also be installed onboard. We aim to address the topics of sensor-based control and navigation both from a theoretical and practical point of view. As such, we will concentrate not only on analyzing the stability and convergence of the proposed solutions but also on using the quadrotor testbed to demonstrate their applicability.

Envisaged tasks for a single vehicle include i) position and attitude stabilization with respect to a fixed target, ii) following a moving target, iii) tracking a specified feature in the environment keeping a security distance, iv) dynamic coverage of the working environment for data collection purposes. Multi-vehicle sensor networks open up a whole new ground of possibilities. Their spatial distribution and redundancy provide an extended sensory coverage ability and robustness to failure that can be exploited to surpass the capabilities of an individual vehicle. As an alternative to a single, complex vehicle burdened with equipment, a collection of smaller, less expensive, and simpler vehicles can increase efficiency and reliability, while reducing the costs of operation. Designing distributed algorithms for robotic networks poses unique challenges primarily related to the dynamic behavior of the sensing and communication topologies. Adopting a sensor-based approach, we will be concerned with accomplishing coordinated tasks such as gathering, pattern formation, or area coverage, using minimal communication between vehicles and relying mostly on the ability of each vehicle to sense the environment and the relative position of the surrounding vehicles.