Robust Engineering of Network Embedded Systems

M.I.T. Artificial Intelligence Laboratory

M.I.T. Space Systems Laboratory

Computer science is currently built on a foundation that largely assumes the existence of a perfect infrastructure. Integrated circuits are fabricated in clean-room environments, tested deterministically, and discarded if even a single defect is uncovered. Entire software systems fail with single-line errors. This foundation will not support the complex and flexibly reliable systems we will need in the future. Completely new approaches that provide dynamic adaptability are required. Those approaches will involve new models of computation, largely drawn from biologically motivated metaphors, and from new, adaptive control technology.

Among these new, large complex systems are Networked Embedded Systems, composed of hundreds or thousands of components, each of which communicates with near neighbors, and each of which is connected, via sensors or actuators, with its physical environment. Tasks for the ensemble are dynamically assignable, and require reliable and predictable performance even though the components are individually unreliable. There are enormous possibilities for application of such systems for defense as well as civilian applications, including maintenance and support of vehicles and weapons, large amorphous sensor grids, and smart materials, to name a few. These systems are now feasible because: 1) Moore's law continues to reduce the cost of computation and memory; 2) Wireless networking (e.g. IEEE 802.11) can provide high bandwidth (at least in a local area); 3) MEMS technology can affordably integrate computation with sensors and effectors; and 4) Bio-technology is beginning to offer us the integration of digital logic and biological capabilities.

Coordination, control and programming of such systems is currently an unsolved problem, and is the main barrier to effective use of such systems. Clearly, these systems must be designed and engineered to deliver system wide predictability and reliabilility, and programmed to dynamically respond to individual component failures. We believe that this kind of system wide control can only be accomplished using a model based approach. The kinds of models that we envision are biologically inspired models of organization and behavioral control, physical system models of the relationship of individual components to their environment, and overall models of the system structure and state.

Our work will focus on coordinated vehicles performing sensing and surveillance task. It will have three major technical thrusts. The first is concerned with how to assemble novel sensory arrays from large scale collections of relatively simple components. Key to this thrust will be the ability to dynamically construct models of the structure of the ensemble and to use these models to control the properties of aggregate sensors. The second thrust, model based autonomy, will be concerned with smaller collections of more complex components that collectively function as a team to achieve a common goal. The third thrust will be on novel software engineering approaches that allow one to construct the programs for such systems as if they were a single program. Software frameworks will allow dynamic updating, reconfiguration, and resource allocation of the entire ensemble.

This research will have revolutionary impact by creating the foundational technologies to engender coherent behavior in massive ensembles of fallible components.

Our Quad Chart


Progress Reports
Background Readings


Brian  Williams
Howard E. Shrobe 
Robert Laddaga

Related Projects

Our Dynamic Domain Architecture for Model Based Autonomy Project seeks to provide software frameworks for the control and coordination of embedded system software.
The Model-Based Embedded and Robotic Systems project is a joint project of the MIT Space Systems Lab and the MIT Artificial Intelligence Lab. It investigates model-based techniques for integrating autonomous systems including space systems, robots, mobile vehicles, etc.


Defense Advanced Research Projects Agency