An Integrated Approach to Multiple-vehicle Sensing, Coordination and Control

Controlling collections of unmanned or unmanned aerial and ground vehicles so as to accomplish their assigned mission remains a challenging task, with unsolved issues in the treatment of environment uncertainty, rapidly changing conditions, high dimensional state spaces, and information overload from sensor data. Control and sensing in such systems must be distributed in order to allow effective and scalable solutions, yet must be coordinated to attain global objectives. We propose to develop new computational methods for designing multi-vehicle sensing and control systems and for their on-line verification. Our project, called CoMotion (for Computational Methods for Collaborative Motion), aims at designs that will lead
to the deployment of high performance, safety critical, and scalable military and civilian systems.
Our underlying principle is that, while the physical systems we are interested in exist in a world in which time and state evolve continuously, it is easier, both phenomenologically and computationally, to reason about discrete objects and data. It is difficult to analyze and control a system of 10 aircraft, for
example, yet it is much easier if the system were represented by discrete data, such as flight modes and rules for transitioning between modes, the near neighbors of each aircraft in the system, and the clusters formed by groups of aircraft. In our research, we will develop and exploit dimensionality reduction techniques, and coarse-to-fine approximations. Our research develops three main
themes:
Distributed Hybrid Control. We have proposed a new paradigm for distributed control, which distributes the control systems in a way that avoids the high communication and computation costs of central control, at the same time limiting complexity. The distributed control must, nevertheless, permit centralized authority over those aspects of the system progress that are necessary to achieve high performance goals. Such a challenge can be met by organizing the distributed control in a hierarchical architecture that permits autonomy and thus the use of all the tools of central control, while introducing enough coordination and supervision to ensure the harmony of the distributed controllers necessary for high performance.
Task-Driven Sensing. We envisage systems of vehicles distributed throughout space, each equipped with suites of sensors; sensed information has an associated value towards the task at hand, and information from other vehicles may be necessary to perform the task. If all information were broadcast and processed at every timestep, a massive information glut would result. We propose that the act of sensing and storing information may be made more efficient if it is directed by control algorithms: the control may ask the sensors discrete questions, or task the sensors to determine if elementary
relationships between objects in the environment hold.
Online/Offline Verification.We are developing novel techniques for online and offline verification which will be designed to provide coarse results immediately, and then will gradually refine results as new data is received.